| """GPT-2 SAE last-token programs, layer 7 -- features 21900 to 21999 (100 programs). |
| |
| One function per SAE feature. Grouped 100 to a file so the set stays browsable; |
| the programs themselves are unchanged. |
| """ |
|
|
| def prog_L07F21900(input_tokens): |
| """40 token families, 42 surface forms (train F1 of the uncapped optimum: 0.539). |
| Fires at 7,094 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to Gross Domestic Product (GDP) and its related metrics""" |
| |
| fire = { |
| |
| ' growth', |
| |
| ' GDP', |
| |
| 'DP', |
| |
| ' productivity', |
| |
| ' output', |
| |
| ' graduation', |
| |
| ' Theft', |
| |
| 'GH', ' GH', |
| |
| ' capita', |
| |
| ' Auto', |
| |
| ' Gib', |
| |
| ' GST', |
| |
| ' homicide', |
| |
| ' births', |
| |
| ' bumper', |
| |
| ' prosperity', |
| |
| ' Civilization', |
| |
| ' OECD', |
| |
| ' biodiversity', |
| |
| ' competitiveness', |
| |
| ' AAA', |
| |
| 'fps', ' fps', |
| |
| ' Happiness', |
| |
| ' downside', |
| |
| ' IQ', |
| |
| ' VAT', |
| |
| ' diploma', |
| |
| ' positives', |
| |
| ' Socialism', |
| |
| 'iframe', |
| |
| ' noun', |
| |
| 'ECD', |
| |
| ' BMI', |
| |
| 'atoes', |
| |
| ' GPA', |
| |
| ' GN', |
| |
| ' genomes', |
| |
| 'itiveness', |
| |
| ' CSV', |
| |
| ' profitability', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'growth-family', 1 |
|
|
| def prog_L07F21901(input_tokens): |
| """26 token families, 59 surface forms (train F1 of the uncapped optimum: 0.303). |
| Fires at 97,961 of 36,786,566 positions.""" |
| |
| """prev SAE description: expressions of relief or positive emotional responses""" |
| |
| fire = { |
| |
| ' heart', ' hearts', 'heart', 'hearted', 'heartedly', |
| |
| 'ed', |
| |
| ' shock', ' shocked', 'shock', |
| |
| ' excited', |
| |
| ' concerned', ' concern', ' Concern', |
| |
| ' emotional', ' emotions', |
| |
| ' happy', 'happy', |
| |
| ' reaction', ' Reaction', |
| |
| ' tears', ' tear', ' Tears', |
| |
| ' feeling', ' feelings', |
| |
| 'ful', 'fully', 'fulness', |
| |
| ' pain', ' pains', ' Pain', 'pain', |
| |
| ' fear', ' fearful', ' fearing', |
| |
| ' disappointed', ' disappointment', |
| |
| ' surprised', |
| |
| ' upset', |
| |
| ' panic', ' Panic', 'panic', |
| |
| ' nervous', ' nerves', |
| |
| ' sad', ' sadness', ' Sad', |
| |
| ' excitement', |
| |
| ' confidence', 'confidence', |
| |
| ' relief', |
| |
| ' joy', 'joy', ' joyful', |
| |
| ' frustration', |
| |
| ' wonder', ' Wond', |
| |
| ' anger', 'angs', ' Anger', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'heart-family', 1 |
|
|
| def prog_L07F21902(input_tokens): |
| """21 token families, 26 surface forms (train F1 of the uncapped optimum: 0.133). |
| Fires at 61,125 of 36,786,566 positions.""" |
| |
| """prev SAE description: abbreviations and acronyms related to locations or entities""" |
| |
| fire = { |
| |
| 'en', 'EN', |
| |
| 'is', |
| |
| 'z', |
| |
| 'k', |
| |
| '\xad', |
| |
| 'ad', |
| |
| 'um', 'UM', |
| |
| 'ch', |
| |
| 'l', |
| |
| 'v', |
| |
| 'ens', 'ENS', ' Ens', |
| |
| 'se', |
| |
| 'em', |
| |
| 'IN', |
| |
| 'EG', |
| |
| 'qu', 'QU', |
| |
| 'ce', |
| |
| 'ph', |
| |
| 'ib', |
| |
| 'ip', |
| |
| 'AM', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'en-family', 1 |
|
|
| def prog_L07F21903(input_tokens): |
| """1 token families, 3 surface forms (train F1 of the uncapped optimum: 0.145). |
| Fires at 65,662 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases indicating a relationship or belonging""" |
| |
| fire = { |
| |
| ' of', ' Of', ' OF', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'of-family', 1 |
|
|
| def prog_L07F21904(input_tokens): |
| """1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.115). |
| Fires at 87,845 of 36,786,566 positions.""" |
| |
| """prev SAE description: the presence of the word "the" in various contexts""" |
| |
| fire = { |
| |
| ' the', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'the-family', 1 |
|
|
| def prog_L07F21905(input_tokens): |
| """22 token families, 38 surface forms (train F1 of the uncapped optimum: 0.674). |
| Fires at 16,486 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to prisons and the prison system""" |
| |
| fire = { |
| |
| ' prison', ' prisoners', ' prisons', ' Prison', ' prisoner', 'prison', |
| |
| ' jail', ' jailed', ' Jail', ' jails', |
| |
| ' detention', ' Detention', |
| |
| ' imprisoned', ' imprisonment', ' imprison', |
| |
| ' Corrections', ' Correctional', ' corrections', ' correctional', |
| |
| 'Pr', |
| |
| ' Pride', |
| |
| ' Guantanamo', |
| |
| ' execution', |
| |
| ' penal', ' Penal', |
| |
| ' parole', |
| |
| 'ISON', 'isons', |
| |
| ' confinement', |
| |
| ' incarceration', |
| |
| ' solitary', |
| |
| ' Quentin', |
| |
| ' Arkham', |
| |
| ' Mandela', |
| |
| ' incarcerated', |
| |
| ' dystopian', |
| |
| ' Auschwitz', |
| |
| ' totalitarian', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'prison-family', 1 |
|
|
| def prog_L07F21906(input_tokens): |
| """24 token families, 50 surface forms (train F1 of the uncapped optimum: 0.136). |
| Fires at 332,945 of 36,786,566 positions. |
| Fires at 0.91% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: phrases related to compliance and authority""" |
| |
| fire = { |
| |
| ' to', ' TO', |
| |
| ' be', |
| |
| ' been', ' Been', 'been', |
| |
| ' out', |
| |
| 't', |
| |
| ' being', |
| |
| ' up', |
| |
| ' into', |
| |
| ' now', ' NOW', |
| |
| ' appears', ' appear', ' appeared', ' appearing', ' Appears', |
| |
| ' off', ' OFF', |
| |
| ' going', |
| |
| ' come', |
| |
| ' set', |
| |
| ' started', ' starting', ' starts', 'started', ' Starts', |
| |
| ' take', |
| |
| ' remains', ' remain', ' remained', |
| |
| ' go', |
| |
| ' looks', ' looking', ' looked', 'look', ' Looks', |
| |
| ' already', |
| |
| ' coming', 'comes', ' Comes', |
| |
| ' under', ' undes', |
| |
| ' fall', ' falling', ' falls', ' Falling', |
| |
| ' less', ' Less', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'to-family', 1 |
|
|
| def prog_L07F21907(input_tokens): |
| """46 token families, 62 surface forms (train F1 of the uncapped optimum: 0.561). |
| Fires at 17,128 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to a specific vehicle model or related terminology""" |
| |
| fire = { |
| |
| 'TS', ' TS', ' ts', |
| |
| 'Reuters', |
| |
| ' 00', |
| |
| 'TO', |
| |
| ' AS', |
| |
| ' Anth', ' anth', 'anth', 'Anth', |
| |
| ' horn', ' horns', |
| |
| 'ANC', |
| |
| 'WATCH', |
| |
| 'NAS', ' NAS', |
| |
| 'MT', |
| |
| ' AV', |
| |
| ' TED', 'TED', |
| |
| 'TF', ' TF', |
| |
| 'NT', |
| |
| 'NYSE', |
| |
| ' sniper', ' snipers', |
| |
| ' TH', |
| |
| ' SK', |
| |
| ' Tales', ' tal', |
| |
| 'TON', |
| |
| ' som', ' Somers', |
| |
| ' Daly', |
| |
| ' Vand', |
| |
| ' tant', ' Tant', |
| |
| ' scorer', |
| |
| ' Ny', |
| |
| ' Watkins', |
| |
| ' Hastings', |
| |
| ' TC', |
| |
| ' Durant', |
| |
| ' Infinity', |
| |
| ' Turing', |
| |
| ' Toby', |
| |
| ' Spec', 'Spec', |
| |
| ' Thorn', |
| |
| ' THC', |
| |
| ' UCS', |
| |
| ' unders', |
| |
| ' Ware', 'WARE', |
| |
| ' Tav', |
| |
| ' Sung', |
| |
| ' SAS', |
| |
| ' Targ', ' targ', |
| |
| 'osaurus', |
| |
| ' Tobias', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'ts-family', 1 |
|
|
| def prog_L07F21908(input_tokens): |
| """50 token families, 73 surface forms (train F1 of the uncapped optimum: 0.711). |
| Fires at 32,515 of 36,786,566 positions.""" |
| |
| """prev SAE description: terms related to figures and their attributes in a broader context""" |
| |
| fire = { |
| |
| ' vest', ' Vest', |
| |
| ' Plaza', ' plaza', |
| |
| ' robotic', ' Robots', |
| |
| 'Week', ' WEEK', |
| |
| ' Sant', |
| |
| ' Viol', ' viol', 'viol', |
| |
| ' trainer', ' trainers', ' Trainer', |
| |
| 'Veter', ' Vet', ' Veter', |
| |
| 'Guard', 'guard', |
| |
| ' Gerald', |
| |
| 'Cla', ' Cla', ' cla', |
| |
| ' Period', 'period', |
| |
| ' Javier', |
| |
| ' lav', 'lav', |
| |
| 'VERTIS', |
| |
| ' Levant', |
| |
| ' Pand', 'Pand', |
| |
| ' Riv', 'riv', |
| |
| ' saf', ' SAF', |
| |
| ' Guardiola', |
| |
| ' Orwell', |
| |
| ' Supp', |
| |
| ' tot', ' Tot', |
| |
| ' sne', |
| |
| ' Pag', ' pag', |
| |
| ' extinct', |
| |
| ' Cout', ' cout', |
| |
| ' Moy', |
| |
| ' Tomas', |
| |
| ' usher', |
| |
| ' Cait', |
| |
| ' Carly', |
| |
| ' tant', ' Tant', |
| |
| ' Witcher', |
| |
| ' Argent', |
| |
| ' Lal', |
| |
| 'anto', |
| |
| ' lingu', |
| |
| ' Gat', |
| |
| ' Tarant', |
| |
| 'arant', |
| |
| 'quist', |
| |
| ' figur', |
| |
| ' Rowling', |
| |
| ' Nort', ' nort', |
| |
| 'ipel', |
| |
| ' Laurent', |
| |
| ' MGM', |
| |
| ' pediatric', |
| |
| ' Clown', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'vest-family', 1 |
|
|
| def prog_L07F21909(input_tokens): |
| """23 token families, 52 surface forms (train F1 of the uncapped optimum: 0.125). |
| Fires at 163,939 of 36,786,566 positions. |
| Fires at 0.45% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: references to notable public appearances or debuts""" |
| |
| fire = { |
| |
| ' in', |
| |
| ' season', ' seasons', 'season', ' seasonal', |
| |
| ' time', |
| |
| ' game', 'game', |
| |
| ' album', ' albums', ' Album', 'album', |
| |
| ' since', |
| |
| ' goal', ' goals', ' Goal', 'goal', |
| |
| ' win', ' wins', |
| |
| ' ever', 'ever', ' EVER', |
| |
| ' debut', |
| |
| ' book', ' books', |
| |
| ' games', |
| |
| ' match', ' matches', 'match', |
| |
| ' film', ' films', |
| |
| ' stories', ' stores', ' Stories', |
| |
| ' career', ' careers', |
| |
| ' latest', ' Latest', 'latest', |
| |
| ' project', ' projects', |
| |
| ' post', ' posts', |
| |
| ' appearance', ' Appearance', |
| |
| ' conference', ' Conference', 'conference', |
| |
| ' meeting', ' meetings', ' Meeting', |
| |
| ' history', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'in-family', 1 |
|
|
| def prog_L07F21910(input_tokens): |
| """23 token families, 62 surface forms (train F1 of the uncapped optimum: 0.123). |
| Fires at 353,806 of 36,786,566 positions. |
| Fires at 0.96% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: statistical information related to sports performance and records""" |
| |
| fire = { |
| |
| ' the', |
| |
| '-', |
| |
| ' their', ' Their', 'Their', ' theirs', |
| |
| ' season', 'season', ' seasons', |
| |
| ' win', ' wins', 'win', ' Win', 'Win', |
| |
| ' team', ' teams', 'team', ' Teams', |
| |
| ' against', 'Against', |
| |
| ' first', |
| |
| ' time', 'time', |
| |
| ' over', |
| |
| ' play', ' played', ' playing', ' plays', 'played', ' Playing', 'Playing', ' PLAY', |
| |
| ' game', 'game', |
| |
| ' year', |
| |
| ' top', ' Top', |
| |
| ' best', 'best', 'Best', |
| |
| ' since', 'since', |
| |
| ' games', |
| |
| ' good', 'good', |
| |
| ' head', ' heading', ' headed', ' Head', 'Head', 'header', 'HEAD', ' HEAD', |
| |
| ' three', 'three', |
| |
| ' playoff', ' playoffs', ' Playoffs', ' Playoff', |
| |
| "'", |
| |
| ' won', 'won', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'the-family', 1 |
|
|
| def prog_L07F21911(input_tokens): |
| """2 token families, 2 surface forms (train F1 of the uncapped optimum: 0.053). |
| Fires at 23,916 of 36,786,566 positions.""" |
| |
| """prev SAE description: instances of the character "ľ" in the text""" |
| |
| fire = { |
| |
| '�', |
| |
| 'Offline', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '�-family', 1 |
|
|
| def prog_L07F21912(input_tokens): |
| """46 token families, 46 surface forms (train F1 of the uncapped optimum: 0.442). |
| Fires at 198,685 of 36,786,566 positions. |
| Fires at 0.54% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: instances of temporal markers or phrases indicating time and context""" |
| |
| fire = { |
| |
| 'The', |
| |
| 'In', |
| |
| 'It', |
| |
| 'According', |
| |
| 'As', |
| |
| 'On', |
| |
| 'At', |
| |
| 'When', |
| |
| 'After', |
| |
| 'For', |
| |
| 'He', |
| |
| 'If', |
| |
| 'While', |
| |
| 'There', |
| |
| 'That', |
| |
| 'With', |
| |
| 'One', |
| |
| 'By', |
| |
| 'To', |
| |
| 'Now', |
| |
| 'An', |
| |
| 'However', |
| |
| 'Some', |
| |
| 'Last', |
| |
| 'During', |
| |
| 'Since', |
| |
| 'From', |
| |
| 'Speaking', |
| |
| 'Here', |
| |
| 'She', |
| |
| 'More', |
| |
| 'Mr', |
| |
| 'Over', |
| |
| 'Even', |
| |
| 'Two', |
| |
| 'Although', |
| |
| 'His', |
| |
| 'Of', |
| |
| 'Despite', |
| |
| 'Earlier', |
| |
| 'Under', |
| |
| 'Both', |
| |
| 'Most', |
| |
| 'Many', |
| |
| 'ADVERTISEMENT', |
| |
| 'Before', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'the-family', 1 |
|
|
| def prog_L07F21913(input_tokens): |
| """28 token families, 36 surface forms (train F1 of the uncapped optimum: 0.369). |
| Fires at 16,436 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to Santa and gift exchanges""" |
| |
| fire = { |
| |
| ' gift', ' gifts', ' Gift', ' Gifts', |
| |
| ' Santa', 'Santa', |
| |
| ' Christmas', |
| |
| ' package', |
| |
| 'anta', |
| |
| ' Valentine', |
| |
| ' Secret', |
| |
| ' collectors', ' collector', ' Collector', |
| |
| ' reddit', |
| |
| 'ibo', |
| |
| 'ETA', |
| |
| ' SS', |
| |
| ' Etsy', 'etsy', |
| |
| 'mas', 'MAS', |
| |
| ' gif', |
| |
| ' Claus', |
| |
| 'eddy', |
| |
| ' handmade', |
| |
| ' packaging', |
| |
| ' backers', |
| |
| 'hum', |
| |
| ' envelope', |
| |
| ' induct', |
| |
| ' thoughtful', |
| |
| ' Valent', |
| |
| ' Merry', |
| |
| ' parcel', |
| |
| 'arant', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'gift-family', 1 |
|
|
| def prog_L07F21914(input_tokens): |
| """37 token families, 46 surface forms (train F1 of the uncapped optimum: 0.474). |
| Fires at 19,189 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the national anthem""" |
| |
| fire = { |
| |
| ' flag', |
| |
| ' anthem', ' Anthem', |
| |
| ' opener', ' Opening', 'Opening', |
| |
| ' prayer', ' Prayer', ' prayers', |
| |
| ' Charlottesville', |
| |
| ' oath', ' Oath', |
| |
| ' Kaepernick', |
| |
| ' Alleg', ' alleg', |
| |
| ' Thanksgiving', |
| |
| ' patriotic', ' patriot', |
| |
| ' mascot', |
| |
| ' Independence', |
| |
| ' soundtrack', |
| |
| ' Standing', |
| |
| ' brutality', |
| |
| ' allegiance', |
| |
| ' salute', |
| |
| ' rainbow', |
| |
| ' kne', |
| |
| ' militar', |
| |
| ' patriotism', |
| |
| ' halftime', |
| |
| ' Monument', |
| |
| ' kneeling', |
| |
| 'discrimination', |
| |
| ' slurs', ' slur', |
| |
| 'assion', |
| |
| ' intro', |
| |
| ' Coliseum', |
| |
| 'zac', |
| |
| ' Shak', |
| |
| ' swast', |
| |
| ' Remem', |
| |
| 'rimination', |
| |
| 'amo', |
| |
| ' Pledge', |
| |
| ' kicker', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'flag-family', 1 |
|
|
| def prog_L07F21915(input_tokens): |
| """2 token families, 4 surface forms (train F1 of the uncapped optimum: 0.081). |
| Fires at 154,514 of 36,786,566 positions. |
| Fires at 0.42% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: terms related to various specialized fields of expertise and social issues""" |
| |
| fire = { |
| |
| ' social', ' Social', ' soc', |
| |
| ' high', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'soc-family', 1 |
|
|
| def prog_L07F21916(input_tokens): |
| """14 token families, 26 surface forms (train F1 of the uncapped optimum: 0.125). |
| Fires at 116,041 of 36,786,566 positions.""" |
| |
| """prev SAE description: specific technology and internet-related terms""" |
| |
| fire = { |
| |
| ' device', ' Device', 'Device', |
| |
| ' own', ' Own', |
| |
| ' phone', ' Phone', 'phone', |
| |
| ' life', |
| |
| ' favorite', ' Favorite', |
| |
| ' friends', 'friend', ' Friend', |
| |
| ' family', |
| |
| ' account', ' Account', |
| |
| ' body', ' Body', |
| |
| ' car', |
| |
| ' browser', ' browsing', ' Browser', |
| |
| ' email', |
| |
| ' computer', |
| |
| ' address', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'device-family', 1 |
|
|
| def prog_L07F21917(input_tokens): |
| """47 token families, 61 surface forms (train F1 of the uncapped optimum: 0.663). |
| Fires at 28,205 of 36,786,566 positions.""" |
| |
| """prev SAE description: the mention of the term "Isa" along with associated contexts""" |
| |
| fire = { |
| |
| 'ida', 'IDA', |
| |
| 'isa', 'ISA', ' Isa', 'Isa', |
| |
| ' Plaza', ' plaza', |
| |
| 'ISS', ' ISS', 'Iss', ' Iss', |
| |
| 'aca', 'ACA', |
| |
| ' Ellis', |
| |
| ' Visa', ' visa', |
| |
| ' Enix', |
| |
| 'itsu', |
| |
| ' Immigration', |
| |
| ' Nak', |
| |
| 'ESA', ' ESA', 'esa', |
| |
| 'ista', |
| |
| 'iza', |
| |
| 'sa', |
| |
| 'iko', |
| |
| 'asa', ' ASA', |
| |
| 'uka', |
| |
| ' da', |
| |
| 'umi', |
| |
| 'iba', |
| |
| 'issa', |
| |
| ' Genesis', ' genesis', |
| |
| 'achi', |
| |
| 'oshi', |
| |
| 'aci', |
| |
| 'obi', |
| |
| 'eda', |
| |
| ' Fam', |
| |
| 'olla', |
| |
| 'afa', |
| |
| 'DAQ', |
| |
| ' Maya', |
| |
| ' Lennon', |
| |
| 'esta', |
| |
| ' Estonia', |
| |
| ' Mayo', |
| |
| ' Santiago', |
| |
| ' Satoshi', |
| |
| 'ija', |
| |
| 'ivas', |
| |
| ' Drupal', |
| |
| 'iji', |
| |
| ' Paula', |
| |
| 'uta', |
| |
| ' EE', |
| |
| 'cci', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'ida-family', 1 |
|
|
| def prog_L07F21918(input_tokens): |
| """8 token families, 12 surface forms (train F1 of the uncapped optimum: 0.205). |
| Fires at 11,373 of 36,786,566 positions.""" |
| |
| """prev SAE description: negations or phrases emphasizing what one should avoid or not do""" |
| |
| fire = { |
| |
| "'t", |
| |
| ' Don', 'Don', |
| |
| ' NOT', 'NOT', |
| |
| 'Never', ' NEVER', |
| |
| ' beware', ' Beware', |
| |
| 'ASE', |
| |
| 'eware', |
| |
| ' dont', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 't-family', 1 |
|
|
| def prog_L07F21919(input_tokens): |
| """7 token families, 10 surface forms (train F1 of the uncapped optimum: 0.586). |
| Fires at 25,246 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to global topics or issues""" |
| |
| fire = { |
| |
| ' world', ' worlds', ' WORLD', |
| |
| ' worldwide', ' Worldwide', |
| |
| ' globe', |
| |
| ' globally', |
| |
| ' nationwide', |
| |
| ' internationally', |
| |
| ' continents', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'world-family', 1 |
|
|
| def prog_L07F21920(input_tokens): |
| """2 token families, 4 surface forms (train F1 of the uncapped optimum: 0.473). |
| Fires at 23,306 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases involving exploration or delving into various subjects or fields""" |
| |
| fire = { |
| |
| ' into', ' Into', ' INTO', |
| |
| ' foray', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'into-family', 1 |
|
|
| def prog_L07F21921(input_tokens): |
| """6 token families, 7 surface forms (train F1 of the uncapped optimum: 0.059). |
| Fires at 12,541 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to real-life applications and instances""" |
| |
| fire = { |
| |
| '-', |
| |
| ' Salt', |
| |
| ' Time', |
| |
| ' Clear', 'Clear', |
| |
| 'osc', |
| |
| 'world', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '--family', 1 |
|
|
| def prog_L07F21922(input_tokens): |
| """48 token families, 53 surface forms (train F1 of the uncapped optimum: 0.678). |
| Fires at 15,677 of 36,786,566 positions.""" |
| |
| """prev SAE description: occurrences of the name "Corey." """ |
| |
| fire = { |
| |
| ' 1986', '1986', |
| |
| ' Corey', |
| |
| ' Cent', |
| |
| ' Cory', |
| |
| ' Cody', |
| |
| ' Katie', |
| |
| ' Crystal', 'Crystal', |
| |
| ' Nicole', |
| |
| ' Newt', |
| |
| ' Anth', ' anth', 'Anth', |
| |
| ' Iv', 'Iv', |
| |
| ' Carey', |
| |
| ' trove', |
| |
| 'aci', |
| |
| ' Tyson', |
| |
| 'acey', |
| |
| ' Tracy', |
| |
| 'Core', |
| |
| ' mogul', |
| |
| ' Katy', |
| |
| 'Andre', |
| |
| ' Courtney', |
| |
| ' Trey', |
| |
| ' Lad', |
| |
| ' Michele', |
| |
| 'icky', |
| |
| ' Magn', |
| |
| ' subt', |
| |
| ' Collective', |
| |
| ' Axel', |
| |
| ' Bret', |
| |
| ' Madonna', |
| |
| 'andowski', |
| |
| ' gymn', |
| |
| ' Crypto', |
| |
| ' centrist', |
| |
| 'anche', |
| |
| ' Dillon', |
| |
| ' gent', |
| |
| ' Wyatt', |
| |
| ' Madagascar', |
| |
| ' Alicia', |
| |
| ' curator', |
| |
| ' civ', |
| |
| ' Kry', |
| |
| ' Feldman', |
| |
| ' Zac', |
| |
| 'antis', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '1986-family', 1 |
|
|
| def prog_L07F21923(input_tokens): |
| """41 token families, 53 surface forms (train F1 of the uncapped optimum: 0.612). |
| Fires at 7,391 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to rib structures or ribosomes in biological contexts""" |
| |
| fire = { |
| |
| 'rib', ' rib', ' Rib', ' ribs', 'ribes', |
| |
| 'rist', |
| |
| ' Trib', ' trib', |
| |
| 'rab', |
| |
| 'rim', |
| |
| 'Soc', 'soc', |
| |
| ' scrib', 'scrib', ' Scrib', |
| |
| 'ribe', |
| |
| 'ract', |
| |
| ' Britt', |
| |
| ' Rove', |
| |
| ' conj', |
| |
| 'ribut', |
| |
| ' Jord', |
| |
| ' gob', 'gob', |
| |
| 'tri', |
| |
| 'reb', |
| |
| 'minster', |
| |
| 'Desc', 'desc', |
| |
| ' drib', |
| |
| 'ρ', |
| |
| 'ribune', |
| |
| ')(', |
| |
| ' Jarrett', |
| |
| ' Clim', |
| |
| ' cler', |
| |
| ' ration', |
| |
| ' Zip', ' ZIP', |
| |
| ' Jamal', |
| |
| 'tab', 'Tab', |
| |
| ' Tib', |
| |
| ' Bib', |
| |
| ' PJ', |
| |
| 'ruff', |
| |
| ' Newsp', |
| |
| 'loc', |
| |
| ' pron', |
| |
| 'ibli', |
| |
| ' Ragnar', |
| |
| ' prelim', |
| |
| ' cris', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'rib-family', 1 |
|
|
| def prog_L07F21924(input_tokens): |
| """5 token families, 6 surface forms (train F1 of the uncapped optimum: 0.374). |
| Fires at 27,617 of 36,786,566 positions.""" |
| |
| """prev SAE description: punctuation marks, particularly single and double quotation marks""" |
| |
| fire = { |
| |
| '"', |
| |
| '",', |
| |
| '!"', |
| |
| "''", |
| |
| '..."', ' ..."', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '"-family', 1 |
|
|
| def prog_L07F21925(input_tokens): |
| """46 token families, 66 surface forms (train F1 of the uncapped optimum: 0.690). |
| Fires at 15,676 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to a person named Souma or variations of that name""" |
| |
| fire = { |
| |
| 'ou', ' Ou', |
| |
| ' Bou', ' bou', |
| |
| ' soup', ' Soup', |
| |
| ' Sou', ' sou', ' SOU', |
| |
| 'Ste', |
| |
| 'Che', |
| |
| ' Vo', 'Vo', 'VO', ' VO', |
| |
| ' Moy', |
| |
| ' Soul', 'Soul', |
| |
| ' Bour', ' bour', |
| |
| ' Rou', ' rou', |
| |
| ' cou', ' Cous', ' cous', |
| |
| 'hou', ' Hou', |
| |
| ' Mous', ' Mou', |
| |
| ' Tou', ' tou', |
| |
| ' SU', |
| |
| ' RO', |
| |
| ' Rouhani', |
| |
| ' Kun', |
| |
| ' Burg', |
| |
| ' gau', ' Gau', |
| |
| 'Mur', |
| |
| ' TED', |
| |
| ' Pure', |
| |
| ' Sana', |
| |
| ' Fou', ' fou', |
| |
| ' Gow', |
| |
| ' Dra', |
| |
| ' Garn', ' garn', |
| |
| ' Angus', |
| |
| ' orche', ' Orche', |
| |
| ' Og', |
| |
| ' Fare', |
| |
| ' Brookings', |
| |
| ' swe', |
| |
| ' Sao', |
| |
| ' Faul', |
| |
| ' Sweeney', |
| |
| ' Ludwig', |
| |
| ' Fay', |
| |
| ' Kou', |
| |
| 'stre', |
| |
| ' Boo', |
| |
| 'Dou', |
| |
| ' souven', |
| |
| ' Thorn', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'ou-family', 1 |
|
|
| def prog_L07F21926(input_tokens): |
| """7 token families, 13 surface forms (train F1 of the uncapped optimum: 0.676). |
| Fires at 6,367 of 36,786,566 positions.""" |
| |
| """prev SAE description: occurrences of the word "cast" and its variations in relation to voting or decision-making contexts""" |
| |
| fire = { |
| |
| ' cast', ' casting', ' Cast', 'Cast', ' casts', 'casting', ' Casting', |
| |
| ' portray', |
| |
| 'aga', |
| |
| ' audition', |
| |
| ' paints', |
| |
| ' assign', |
| |
| ' consec', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'cast-family', 1 |
|
|
| def prog_L07F21927(input_tokens): |
| """35 token families, 49 surface forms (train F1 of the uncapped optimum: 0.675). |
| Fires at 8,434 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to eggs as a food ingredient""" |
| |
| fire = { |
| |
| ' eggs', ' egg', ' Egg', ' Eggs', 'egg', |
| |
| ' nest', ' nests', ' nesting', |
| |
| ' Chick', ' chick', ' chicks', |
| |
| ' hatch', ' hatched', |
| |
| ' chickens', |
| |
| ' sperm', |
| |
| ' embryos', ' embryo', |
| |
| ' Orb', ' orb', ' orbs', |
| |
| ' Hearts', |
| |
| ' scrambled', |
| |
| ' Germ', ' germ', |
| |
| ' horn', |
| |
| ' Keys', 'Keys', |
| |
| ' hen', |
| |
| ' ov', |
| |
| ' oy', |
| |
| ' poultry', |
| |
| ' goose', |
| |
| ' Iz', |
| |
| 'Event', |
| |
| ' almond', |
| |
| ' sausage', |
| |
| ' ultr', |
| |
| 'Imm', |
| |
| ' gest', |
| |
| ' enzyme', |
| |
| ' yeast', |
| |
| 'mone', |
| |
| ' om', |
| |
| ' Eug', |
| |
| ' ovarian', |
| |
| ' Ender', |
| |
| ' enzymes', |
| |
| ' Ic', |
| |
| ' Yoshi', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'egg-family', 1 |
|
|
| def prog_L07F21928(input_tokens): |
| """22 token families, 60 surface forms (train F1 of the uncapped optimum: 0.106). |
| Fires at 916,471 of 36,786,566 positions. |
| Fires at 2.49% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: references to tasks, actions, or requirements that involve effort or productivity""" |
| |
| fire = { |
| |
| ' of', ' Of', |
| |
| ' a', ' A', 'ía', 'ça', |
| |
| ' is', ' Is', |
| |
| ' was', ' Was', ' WAS', |
| |
| ' are', ' Are', |
| |
| ' have', ' Have', |
| |
| ' I', ' i', |
| |
| ' be', |
| |
| ' not', ' noted', ' Not', ' notes', ' notable', ' NOT', ' noting', |
| |
| ' will', ' willing', ' Will', ' WILL', ' willingly', |
| |
| 't', "'t", ' t', |
| |
| ' you', |
| |
| ' said', |
| |
| ' also', ' Also', |
| |
| ' one', ' One', ' ONE', 'onement', |
| |
| ' were', ' Were', |
| |
| ' can', ' Can', ' CAN', |
| |
| ' up', ' Up', ' UP', |
| |
| ' who', ' Who', |
| |
| ' about', ' About', ' ABOUT', |
| |
| ' had', |
| |
| ' say', ' says', ' Say', ' Saying', ' SAY', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'of-family', 1 |
|
|
| def prog_L07F21929(input_tokens): |
| """13 token families, 16 surface forms (train F1 of the uncapped optimum: 0.551). |
| Fires at 8,095 of 36,786,566 positions.""" |
| |
| """prev SAE description: instances of the word "everywhere." """ |
| |
| fire = { |
| |
| ' everywhere', ' Everywhere', |
| |
| ' anywhere', |
| |
| ' everyday', ' Everyday', |
| |
| ' wherever', |
| |
| ' ubiquitous', |
| |
| 'every', ' EVERY', |
| |
| ' universally', |
| |
| ' omnip', |
| |
| ' Whenever', |
| |
| ' abound', |
| |
| ' Chic', |
| |
| ' Olympia', |
| |
| ' POP', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'everywhere-family', 1 |
|
|
| def prog_L07F21930(input_tokens): |
| """35 token families, 42 surface forms (train F1 of the uncapped optimum: 0.538). |
| Fires at 12,077 of 36,786,566 positions.""" |
| |
| """prev SAE description: mentions of "Night" as part of various events or shows""" |
| |
| fire = { |
| |
| ' Night', 'night', 'Night', ' Nights', ' nightly', ' NIGHT', |
| |
| 'unch', |
| |
| ' Tonight', |
| |
| ' Bat', |
| |
| ' Hour', 'Hour', |
| |
| ' Tomorrow', |
| |
| ' Evening', |
| |
| ' Mystery', |
| |
| ' Anniversary', |
| |
| 'Date', |
| |
| 'etime', |
| |
| ' Arkham', |
| |
| ' Carnival', |
| |
| ' Neighborhood', |
| |
| ' Midnight', |
| |
| ' Phantom', |
| |
| 'anity', |
| |
| ' Nightmare', |
| |
| 'htar', |
| |
| ' Murder', |
| |
| 'IGHT', |
| |
| ' Weekend', |
| |
| ' Maid', |
| |
| ' Node', |
| |
| ' Presents', |
| |
| 'romptu', |
| |
| ' Noon', 'noon', |
| |
| 'urnal', |
| |
| 'quet', |
| |
| ' Dread', |
| |
| ' Majesty', |
| |
| 'iday', |
| |
| ' nighttime', |
| |
| 'NING', |
| |
| 'Raven', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'night-family', 1 |
|
|
| def prog_L07F21931(input_tokens): |
| """25 token families, 52 surface forms (train F1 of the uncapped optimum: 0.115). |
| Fires at 130,812 of 36,786,566 positions.""" |
| |
| """prev SAE description: proper nouns and character names in a movie or narrative context""" |
| |
| fire = { |
| |
| '-', |
| |
| ':', |
| |
| ' B', 'b', 'B', |
| |
| 'h', ' H', ' h', |
| |
| '/', |
| |
| ' K', 'k', ' k', |
| |
| ' L', 'l', |
| |
| ' F', 'f', |
| |
| 'z', ' Z', 'Z', ' z', |
| |
| 'y', |
| |
| ' S', ' s', |
| |
| 'le', ' Le', ' le', |
| |
| ' County', ' COUNTY', |
| |
| ' J', 'j', |
| |
| 'v', ' v', |
| |
| ' G', |
| |
| ' M', ' m', |
| |
| ' de', ' De', |
| |
| 'man', ' Mans', |
| |
| ' Bl', 'bl', 'BL', |
| |
| ' R', 'ré', |
| |
| ' &', |
| |
| 'on', |
| |
| ' P', |
| |
| ' Sports', 'Sports', 'sports', 'Sport', ' SPORTS', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '--family', 1 |
|
|
| def prog_L07F21932(input_tokens): |
| """12 token families, 15 surface forms (train F1 of the uncapped optimum: 0.091). |
| Fires at 23,976 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to part-time work and its implications""" |
| |
| fire = { |
| |
| ' 1', |
| |
| ' Congress', |
| |
| ' states', |
| |
| 'time', |
| |
| 'playing', ' Playing', 'players', |
| |
| ' ways', |
| |
| ' continues', |
| |
| ' II', |
| |
| 'action', |
| |
| 'ire', 'IRE', |
| |
| ' Theft', |
| |
| ' Parliament', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '1-family', 1 |
|
|
| def prog_L07F21933(input_tokens): |
| """4 token families, 5 surface forms (train F1 of the uncapped optimum: 0.836). |
| Fires at 19,465 of 36,786,566 positions.""" |
| |
| """prev SAE description: mentions of the United States""" |
| |
| fire = { |
| |
| ' US', 'US', |
| |
| ' NZ', |
| |
| ' UAE', |
| |
| ' USDA', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'us-family', 1 |
|
|
| def prog_L07F21934(input_tokens): |
| """7 token families, 8 surface forms (train F1 of the uncapped optimum: 0.106). |
| Fires at 207,459 of 36,786,566 positions. |
| Fires at 0.56% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: command-like structures or technical terminology in the text""" |
| |
| fire = { |
| |
| ',', |
| |
| ' (', |
| |
| ' which', |
| |
| ' –', |
| |
| '),', ' ),', |
| |
| ' Could', |
| |
| '—', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return ',-family', 1 |
|
|
| def prog_L07F21935(input_tokens): |
| """47 token families, 51 surface forms (train F1 of the uncapped optimum: 0.820). |
| Fires at 19,504 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to Germany and its derivatives""" |
| |
| fire = { |
| |
| ' German', 'German', ' Germans', |
| |
| ' Germany', 'Germany', |
| |
| ' Berlin', |
| |
| ' Munich', |
| |
| ' euros', |
| |
| ' Merkel', |
| |
| ' Hitler', |
| |
| ' Bayern', |
| |
| 'Ber', |
| |
| ' Volkswagen', |
| |
| ' Hamburg', |
| |
| ' Frankfurt', |
| |
| ' Bundesliga', |
| |
| ' Cologne', |
| |
| 'LIN', |
| |
| ' Dortmund', |
| |
| ' Ghana', |
| |
| ' Deutsche', |
| |
| 'France', |
| |
| ' Bav', |
| |
| ' Adolf', |
| |
| ' Andreas', |
| |
| 'sburg', |
| |
| ' Greenland', |
| |
| 'utsch', |
| |
| ' Sax', |
| |
| ' Spiegel', |
| |
| ' Zeit', |
| |
| 'ß', |
| |
| 'gart', |
| |
| ' Wolfgang', |
| |
| 'zig', |
| |
| ' Fra', 'Fra', |
| |
| ' Ludwig', |
| |
| ' Franz', |
| |
| 'utsche', |
| |
| ' Rosenberg', |
| |
| ' Bundes', |
| |
| ' Friedrich', |
| |
| ' Blitz', |
| |
| ' Muller', |
| |
| ' Bayer', |
| |
| ' Dresden', |
| |
| ' Braun', |
| |
| ' Lutheran', |
| |
| ' Markus', |
| |
| ' Wilhelm', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'german-family', 1 |
|
|
| def prog_L07F21936(input_tokens): |
| """5 token families, 13 surface forms (train F1 of the uncapped optimum: 0.145). |
| Fires at 387,570 of 36,786,566 positions. |
| Fires at 1.05% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: terms related to health, physical attributes, and systems in a structured context""" |
| |
| fire = { |
| |
| '-', |
| |
| ' new', 'new', |
| |
| ' high', ' higher', ' highly', 'high', 'highly', |
| |
| ' long', 'long', ' LONG', |
| |
| 'year', ' yearly', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '--family', 1 |
|
|
| def prog_L07F21937(input_tokens): |
| """20 token families, 57 surface forms (train F1 of the uncapped optimum: 0.108). |
| Fires at 613,904 of 36,786,566 positions. |
| Fires at 1.67% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: references to awards and accomplishments""" |
| |
| fire = { |
| |
| ' the', 'The', ' The', |
| |
| ' a', 'A', |
| |
| ' was', ' Was', 'Was', |
| |
| ' has', 'has', |
| |
| ' an', 'An', |
| |
| ' but', ' But', 'But', 'but', |
| |
| ' have', 'Have', ' Have', 'have', |
| |
| ' he', ' He', 'He', |
| |
| ' are', ' Are', 'Are', |
| |
| ' will', ' Will', 'Will', 'will', ' willful', |
| |
| ' after', 'After', ' After', |
| |
| ' "', |
| |
| ' who', |
| |
| ' were', 'Were', |
| |
| ' new', |
| |
| 'N', ' N', |
| |
| ' help', ' helped', ' helping', ' helps', ' helper', |
| |
| ' she', 'She', ' She', |
| |
| ' people', 'People', |
| |
| ' play', ' playing', ' players', ' Playing', ' playful', 'Playing', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'the-family', 1 |
|
|
| def prog_L07F21938(input_tokens): |
| """40 token families, 45 surface forms (train F1 of the uncapped optimum: 0.765). |
| Fires at 9,711 of 36,786,566 positions.""" |
| |
| """prev SAE description: occurrences of the name "Richard" """ |
| |
| fire = { |
| |
| ' Richard', 'Richard', ' Richards', |
| |
| 'rich', 'Rich', |
| |
| 'rick', |
| |
| ' Nicolas', |
| |
| ' Nigel', |
| |
| ' Sherman', |
| |
| ' Richardson', |
| |
| ' Gerald', |
| |
| ' Salman', |
| |
| 'ICH', |
| |
| 'inton', |
| |
| 'Anthony', |
| |
| ' Darrell', |
| |
| ' Herbert', |
| |
| ' Sergey', |
| |
| ' Rahul', |
| |
| ' Burr', |
| |
| 'Charles', |
| |
| ' Angus', |
| |
| 'igel', |
| |
| ' RF', 'rf', |
| |
| ' Rutherford', |
| |
| ' Randolph', |
| |
| ' Rouhani', |
| |
| 'wayne', |
| |
| 'itchie', |
| |
| ' Richie', |
| |
| 'Roy', |
| |
| ' Maurice', |
| |
| ' Forrest', |
| |
| ' Norris', |
| |
| ' reclaimed', |
| |
| ' arsenic', |
| |
| ' Eleanor', |
| |
| ' Rusty', |
| |
| ' inhibitors', ' inhibitor', |
| |
| ' INS', |
| |
| 'aryl', |
| |
| 'で', |
| |
| 'ス', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'richard-family', 1 |
|
|
| def prog_L07F21939(input_tokens): |
| """17 token families, 60 surface forms (train F1 of the uncapped optimum: 0.126). |
| Fires at 602,594 of 36,786,566 positions. |
| Fires at 1.64% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: phrases and sentences that encourage audience engagement and interaction on a blog""" |
| |
| fire = { |
| |
| ' to', ' TO', |
| |
| ' I', ' i', |
| |
| ' this', ' This', 'this', ' THIS', 'THIS', |
| |
| ' you', ' YOU', 'YOU', |
| |
| ' we', ' We', ' WE', |
| |
| ' be', ' BE', |
| |
| ' our', ' Our', ' OUR', |
| |
| ' my', ' MY', |
| |
| ' post', ' posted', ' posts', ' Posted', ' posting', 'Posted', 'post', 'posted', ' POST', 'posts', ' postings', 'POST', |
| |
| ' 1', '1', |
| |
| ' some', ' SOME', |
| |
| ' what', ' whats', ' WHAT', |
| |
| ' your', ' yours', ' YOUR', |
| |
| ' if', ' If', ' IF', |
| |
| ' most', |
| |
| ' here', ' HERE', 'here', |
| |
| ' Thank', ' thanks', ' Thanks', ' thank', 'Thanks', 'Thank', 'thanks', ' THANK', ' thanking', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'to-family', 1 |
|
|
| def prog_L07F21940(input_tokens): |
| """12 token families, 29 surface forms (train F1 of the uncapped optimum: 0.148). |
| Fires at 192,418 of 36,786,566 positions. |
| Fires at 0.52% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: nouns that represent categories or classifications""" |
| |
| fire = { |
| |
| 's', |
| |
| ' services', ' Services', 'services', |
| |
| ' system', ' systems', ' System', ' Systems', ' SYSTEM', |
| |
| ' games', |
| |
| ' officials', |
| |
| ' groups', 'groups', ' Groups', |
| |
| ' account', ' accounts', ' accountable', ' Accounts', |
| |
| ' members', ' Member', |
| |
| ' Street', ' streets', ' Streets', |
| |
| ' service', ' Service', |
| |
| ' power', ' powers', ' Powers', |
| |
| ' data', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 's-family', 1 |
|
|
| def prog_L07F21941(input_tokens): |
| """42 token families, 55 surface forms (train F1 of the uncapped optimum: 0.582). |
| Fires at 9,375 of 36,786,566 positions.""" |
| |
| """prev SAE description: terms related to sacredness and holy significance""" |
| |
| fire = { |
| |
| ' holy', 'holy', 'Holy', |
| |
| ' ancient', |
| |
| ' sacred', ' Sacred', |
| |
| ' sanctuary', ' Sanctuary', |
| |
| ' healing', ' Healing', |
| |
| ' precious', |
| |
| ' revered', ' reverence', |
| |
| 'Sac', 'sac', ' sac', |
| |
| ' magical', |
| |
| ' politic', |
| |
| ' forbidden', ' Forbidden', |
| |
| ' shrine', ' Shrine', |
| |
| ' sanct', ' Sanct', |
| |
| ' mythical', |
| |
| ' ceremonial', |
| |
| ' blessed', ' Blessed', |
| |
| ' pristine', |
| |
| ' mystical', ' mystic', |
| |
| ' pilgrimage', |
| |
| ' taboo', |
| |
| ' cherished', |
| |
| ' Burning', |
| |
| ' solemn', |
| |
| ' shameful', |
| |
| ' communal', |
| |
| ' devout', |
| |
| 'brance', |
| |
| ' purity', |
| |
| ' altar', |
| |
| 'cca', |
| |
| ' Mecca', |
| |
| 'ctuary', |
| |
| ' ancestral', |
| |
| ' holiest', |
| |
| ' vener', |
| |
| 'asured', |
| |
| ' cleansing', |
| |
| ' entrusted', |
| |
| ' rite', |
| |
| ' foundational', |
| |
| ' virtuous', |
| |
| ' pious', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'holy-family', 1 |
|
|
| def prog_L07F21942(input_tokens): |
| """47 token families, 52 surface forms (train F1 of the uncapped optimum: 0.547). |
| Fires at 17,198 of 36,786,566 positions.""" |
| |
| """prev SAE description: expressions of strong fixation or preoccupation with certain subjects or ideas""" |
| |
| fire = { |
| |
| ' devoted', |
| |
| ' obsessed', ' obsessive', |
| |
| ' obsession', ' obsess', |
| |
| ' affair', |
| |
| ' frenzy', |
| |
| ' jugg', ' Jugg', |
| |
| ' fascinated', |
| |
| ' sne', |
| |
| ' fever', |
| |
| 'occupied', 'occup', |
| |
| ' craz', |
| |
| ' cra', |
| |
| ' addicted', |
| |
| ' synonymous', |
| |
| ' Obs', ' obs', |
| |
| ' eccentric', |
| |
| ' fascination', |
| |
| ' devote', |
| |
| ' crusade', |
| |
| ' hysteria', |
| |
| ' intrigued', |
| |
| ' vigilant', |
| |
| ' bored', |
| |
| ' fantasies', |
| |
| ' relentlessly', |
| |
| ' stuffed', |
| |
| 'essed', |
| |
| 'ulsive', |
| |
| ' devotion', |
| |
| ' relentless', |
| |
| ' stud', |
| |
| ' incess', |
| |
| 'icion', |
| |
| ' fanatic', |
| |
| 'urious', |
| |
| ' endlessly', |
| |
| ' intensely', |
| |
| 'obia', |
| |
| ' lust', |
| |
| ' fetish', |
| |
| ' experimentation', |
| |
| ' maniac', |
| |
| ' zeal', |
| |
| ' exhaustive', |
| |
| ' glued', |
| |
| ' hoop', |
| |
| ' flirt', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'devot-family', 1 |
|
|
| def prog_L07F21943(input_tokens): |
| """13 token families, 18 surface forms (train F1 of the uncapped optimum: 0.079). |
| Fires at 101,913 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases expressing agreement or acceptance""" |
| |
| fire = { |
| |
| ' this', ' THIS', |
| |
| ' your', |
| |
| ' her', ' hers', ' HER', |
| |
| ' what', ' WHAT', |
| |
| ' our', |
| |
| ' these', ' THESE', |
| |
| ' those', |
| |
| ' me', |
| |
| ' idea', |
| |
| ' anything', |
| |
| ' anyone', |
| |
| ' everything', |
| |
| ' current', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'thi-family', 1 |
|
|
| def prog_L07F21944(input_tokens): |
| """28 token families, 59 surface forms (train F1 of the uncapped optimum: 0.167). |
| Fires at 146,432 of 36,786,566 positions. |
| Fires at 0.40% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: words or fragments related to character names or titles""" |
| |
| fire = { |
| |
| '-', |
| |
| 'o', 'O', ' O', 'ño', 'oğ', |
| |
| 'y', 'Y', |
| |
| 'en', 'EN', ' EN', |
| |
| 'on', 'ON', |
| |
| 'er', 'ER', |
| |
| 'a', 'ña', 'ía', 'ça', |
| |
| 'al', 'AL', |
| |
| 'an', 'AN', |
| |
| 'as', 'AS', |
| |
| 'e', 'ée', |
| |
| 'i', |
| |
| 'z', |
| |
| 'ad', ' Ad', 'AD', |
| |
| 'os', 'OS', |
| |
| 'u', |
| |
| 'ar', 'AR', |
| |
| 'h', |
| |
| 'at', |
| |
| ' alerts', ' Alert', 'Alert', |
| |
| 'ne', |
| |
| ' Sports', 'Sports', 'sports', ' SPORTS', |
| |
| 'ist', 'ister', 'ISTER', |
| |
| 'le', |
| |
| 'ORE', 'ore', 'ores', |
| |
| 'ia', 'IA', |
| |
| 'am', |
| |
| 'ville', 'VILLE', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '--family', 1 |
|
|
| def prog_L07F21945(input_tokens): |
| """10 token families, 17 surface forms (train F1 of the uncapped optimum: 0.718). |
| Fires at 14,942 of 36,786,566 positions.""" |
| |
| """prev SAE description: mentions of Twitter and its related functionalities""" |
| |
| fire = { |
| |
| ' Twitter', 'Twitter', ' twitter', |
| |
| ' tweets', ' tweeted', ' tweeting', |
| |
| ' Tw', ' TW', 'TW', |
| |
| ' followers', |
| |
| ' Twe', 'Twe', |
| |
| ' Colbert', |
| |
| ' Twitch', |
| |
| ' Waterloo', |
| |
| 'izens', |
| |
| ' IRC', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'twitt-family', 1 |
|
|
| def prog_L07F21946(input_tokens): |
| """3 token families, 3 surface forms (train F1 of the uncapped optimum: 0.084). |
| Fires at 70,949 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to health-related topics and conditions""" |
| |
| fire = { |
| |
| ',', |
| |
| '),', |
| |
| ';', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return ',-family', 1 |
|
|
| def prog_L07F21947(input_tokens): |
| """22 token families, 68 surface forms (train F1 of the uncapped optimum: 0.341). |
| Fires at 67,898 of 36,786,566 positions.""" |
| |
| """prev SAE description: technical specifications and features related to computer hardware""" |
| |
| fire = { |
| |
| ' Core', 'core', ' core', 'Core', |
| |
| ' chip', ' chips', 'chip', ' Chips', 'Chip', |
| |
| ' performance', 'performance', ' Performance', 'Performance', |
| |
| ' processor', ' processors', ' Processor', 'processor', |
| |
| ' Intel', 'Intel', ' intel', 'intel', |
| |
| ' graphics', ' Graphics', 'Graphics', |
| |
| ' display', ' displays', ' Display', 'Display', |
| |
| ' GPU', ' GPUs', 'GPU', 'gpu', |
| |
| ' memory', ' Memory', 'Memory', 'memory', |
| |
| ' CPU', ' CPUs', 'CPU', ' cpu', |
| |
| ' architecture', ' Architecture', |
| |
| ' cooling', ' cooler', 'cool', ' Cool', 'Cool', |
| |
| ' micro', ' Micro', 'Micro', |
| |
| ' AMD', 'AMD', |
| |
| 'generation', |
| |
| ' GTX', |
| |
| ' benchmark', ' benchmarks', |
| |
| ' specs', ' spec', 'Spec', ' SPEC', |
| |
| ' dual', ' Dual', 'Dual', |
| |
| 'bit', 'bits', |
| |
| 'GHz', ' GHz', |
| |
| ' GeForce', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'core-family', 1 |
|
|
| def prog_L07F21948(input_tokens): |
| """18 token families, 45 surface forms (train F1 of the uncapped optimum: 0.094). |
| Fires at 710,557 of 36,786,566 positions. |
| Fires at 1.93% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: phrases that involve political authority and mandates""" |
| |
| fire = { |
| |
| ' of', ' OF', |
| |
| ' to', ' TO', |
| |
| ' for', ' FOR', |
| |
| ' on', ' ON', |
| |
| ' with', ' WITH', |
| |
| ' by', |
| |
| ' his', ' HIS', |
| |
| ' their', ' THEIR', ' theirs', |
| |
| ' years', ' year', ' Year', ' Years', ' YEAR', 'Year', ' yearly', |
| |
| ' he', ' HE', |
| |
| ' last', ' lasted', ' lasting', ' lasts', ' LAST', 'lasting', |
| |
| ' about', ' About', 'about', ' ABOUT', |
| |
| ' they', ' THEY', |
| |
| ' who', ' WHO', |
| |
| ' its', |
| |
| ' her', ' HER', |
| |
| ' than', ' Than', |
| |
| ' $', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'of-family', 1 |
|
|
| def prog_L07F21949(input_tokens): |
| """26 token families, 38 surface forms (train F1 of the uncapped optimum: 0.508). |
| Fires at 9,453 of 36,786,566 positions.""" |
| |
| """prev SAE description: mentions of savings""" |
| |
| fire = { |
| |
| ' savings', ' saved', ' saving', ' saves', 'saving', 'sav', ' Saving', ' Savings', 'Sav', |
| |
| ' save', ' Save', 'Save', |
| |
| ' wilderness', ' Wilderness', |
| |
| 'wich', |
| |
| ' Salvation', ' salvation', |
| |
| ' browsing', |
| |
| ' Fortress', |
| |
| ' sol', |
| |
| ' shaving', |
| |
| 'overs', |
| |
| ' reductions', |
| |
| ' Sophia', |
| |
| 'cutting', |
| |
| 'bucks', |
| |
| 'hew', |
| |
| ' Wisdom', |
| |
| ' insulation', |
| |
| ' smokers', |
| |
| ' masturbation', |
| |
| ' avoidance', |
| |
| ' stash', |
| |
| ' 🙂', |
| |
| ' scissors', |
| |
| ' shortfall', |
| |
| ' withdrawals', |
| |
| 'iencies', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'sav-family', 1 |
|
|
| def prog_L07F21950(input_tokens): |
| """4 token families, 4 surface forms (train F1 of the uncapped optimum: 0.073). |
| Fires at 29,876 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the NCAA D-League or related basketball topics""" |
| |
| fire = { |
| |
| '-', |
| |
| '_', |
| |
| '&', |
| |
| ' MORE', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '--family', 1 |
|
|
| def prog_L07F21951(input_tokens): |
| """14 token families, 28 surface forms (train F1 of the uncapped optimum: 0.159). |
| Fires at 431,912 of 36,786,566 positions. |
| Fires at 1.17% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: references to demographics, attributes, or characteristics of subjects""" |
| |
| fire = { |
| |
| ')', |
| |
| ' years', ' year', |
| |
| '"', |
| |
| ' people', |
| |
| ' home', |
| |
| ' world', |
| |
| 'old', ' old', ' older', 'olds', ' olds', 'older', |
| |
| ' government', ' governments', ' governance', ' governed', |
| |
| ' law', ' laws', 'law', |
| |
| ' University', ' universities', ' universes', |
| |
| ' name', |
| |
| ' life', |
| |
| ' States', ' states', |
| |
| ' police', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return ')-family', 1 |
|
|
| def prog_L07F21952(input_tokens): |
| """34 token families, 68 surface forms (train F1 of the uncapped optimum: 0.480). |
| Fires at 134,244 of 36,786,566 positions.""" |
| |
| """prev SAE description: objects and items often associated with logistical or security contexts""" |
| |
| fire = { |
| |
| 'ers', |
| |
| ' cars', ' Cars', 'cars', |
| |
| ' cards', ' Cards', 'cards', |
| |
| ' vehicles', ' Vehicles', |
| |
| ' cameras', |
| |
| ' phones', 'phones', 'Phones', |
| |
| ' officers', |
| |
| ' cells', ' Cells', 'cells', |
| |
| ' homes', ' Homes', |
| |
| ' ads', 'ads', ' Ads', |
| |
| ' bags', 'bags', |
| |
| ' dogs', ' Dogs', 'dogs', |
| |
| ' bombs', ' bombers', ' bombings', ' Bombs', |
| |
| ' hands', |
| |
| ' tickets', ' Tickets', 'Tickets', |
| |
| ' boxes', 'boxes', |
| |
| ' machines', ' Machines', |
| |
| ' accounts', |
| |
| ' signs', |
| |
| ' guns', 'guns', |
| |
| ' computers', |
| |
| ' units', ' Units', 'units', |
| |
| ' tests', ' testers', 'tests', |
| |
| ' trucks', |
| |
| ' planes', 'planes', ' Planes', |
| |
| ' houses', 'houses', ' Houses', |
| |
| ' lights', 'lights', |
| |
| ' missiles', |
| |
| ' letters', 'letters', |
| |
| ' messages', |
| |
| ' jets', |
| |
| ' trains', ' trainers', |
| |
| ' buses', |
| |
| ' walls', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'ers-family', 1 |
|
|
| def prog_L07F21953(input_tokens): |
| """20 token families, 26 surface forms (train F1 of the uncapped optimum: 0.134). |
| Fires at 91,364 of 36,786,566 positions.""" |
| |
| """prev SAE description: technical specifications or version information of software or devices""" |
| |
| fire = { |
| |
| ':', |
| |
| ' years', 'year', |
| |
| ']', |
| |
| '5', |
| |
| '0', |
| |
| ' weeks', 'week', ' Weeks', |
| |
| 'nd', |
| |
| '30', |
| |
| '4', |
| |
| ' hours', 'hour', |
| |
| '8', |
| |
| '6', |
| |
| ' billion', 'billion', ' Billion', |
| |
| '00', |
| |
| '7', |
| |
| '][', |
| |
| 'time', |
| |
| '500', |
| |
| 'x', |
| |
| 'day', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return ':-family', 1 |
|
|
| def prog_L07F21954(input_tokens): |
| """15 token families, 28 surface forms (train F1 of the uncapped optimum: 0.076). |
| Fires at 336,110 of 36,786,566 positions. |
| Fires at 0.91% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: phrases and words that convey emotional intensity or descriptions of experiences""" |
| |
| fire = { |
| |
| ',', |
| |
| ' and', ' And', 'And', |
| |
| ' as', |
| |
| ' all', ' All', 'all', |
| |
| ' —', '—', |
| |
| ' which', 'which', |
| |
| ' It', |
| |
| ' so', |
| |
| ' when', 'when', |
| |
| ' –', '–', |
| |
| ';', ' ;', |
| |
| ' like', ' Like', |
| |
| ' while', 'while', |
| |
| ' -', |
| |
| ' even', ' Even', 'even', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return ',-family', 1 |
|
|
| def prog_L07F21955(input_tokens): |
| """2 token families, 4 surface forms (train F1 of the uncapped optimum: 0.504). |
| Fires at 5,431 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases indicating distance or separation""" |
| |
| fire = { |
| |
| ' far', 'far', ' FAR', |
| |
| ' insofar', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'far-family', 1 |
|
|
| def prog_L07F21956(input_tokens): |
| """15 token families, 30 surface forms (train F1 of the uncapped optimum: 0.759). |
| Fires at 12,152 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to "source" material or citations""" |
| |
| fire = { |
| |
| ' source', 'Source', ' Source', 'source', 'SOURCE', |
| |
| ' sources', ' sourced', ' Sources', ' sourcing', |
| |
| ' destination', ' Destination', |
| |
| ' supplier', ' suppliers', |
| |
| ' origin', |
| |
| ' fountain', ' Fountain', |
| |
| ' withhold', ' withholding', |
| |
| 'ourced', 'ources', 'ourcing', |
| |
| ' informant', ' informants', |
| |
| ' insider', |
| |
| ' trove', |
| |
| ' originating', |
| |
| 'ource', 'OURCE', |
| |
| 'ribut', |
| |
| ' supplemental', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'source-family', 1 |
|
|
| def prog_L07F21957(input_tokens): |
| """10 token families, 19 surface forms (train F1 of the uncapped optimum: 0.109). |
| Fires at 174,240 of 36,786,566 positions. |
| Fires at 0.47% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: references to specific individuals, organizations, or authorities in a formal context""" |
| |
| fire = { |
| |
| 's', |
| |
| ' Trump', ' TRUMP', |
| |
| ' Obama', |
| |
| ' government', ' Government', ' governments', ' Governments', |
| |
| 'com', |
| |
| ' administration', ' Administration', |
| |
| ' Party', ' party', |
| |
| ' media', |
| |
| ' Department', ' department', ' departments', |
| |
| ' Clinton', ' Clintons', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 's-family', 1 |
|
|
| def prog_L07F21958(input_tokens): |
| """12 token families, 12 surface forms (train F1 of the uncapped optimum: 0.459). |
| Fires at 6,017 of 36,786,566 positions.""" |
| |
| """prev SAE description: the phrase "continued below." """ |
| |
| fire = { |
| |
| ' highlights', |
| |
| ' Continued', |
| |
| ' Below', |
| |
| 'Spread', |
| |
| 'cript', |
| |
| 'FIELD', |
| |
| ' Akin', |
| |
| ' Latest', |
| |
| 'Trend', |
| |
| ' Replay', |
| |
| 'Repe', |
| |
| ' DK', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'highlight-family', 1 |
|
|
| def prog_L07F21959(input_tokens): |
| """44 token families, 68 surface forms (train F1 of the uncapped optimum: 0.617). |
| Fires at 12,404 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to texture in various contexts""" |
| |
| fire = { |
| |
| ' Tex', 'Tex', 'tex', ' tex', |
| |
| 'uv', ' UV', 'UV', |
| |
| ' terrain', |
| |
| ' texture', ' Texture', 'Texture', 'texture', |
| |
| ' thermal', ' Thermal', |
| |
| ' sculpt', |
| |
| ' brutality', |
| |
| 'struct', ' struct', 'Struct', ' Struct', |
| |
| ' Tes', ' Tess', 'Tes', |
| |
| ' mesh', ' Mesh', ' meshes', 'Mesh', |
| |
| 'essa', |
| |
| ' Text', ' TEXT', |
| |
| ' Nicole', |
| |
| ' Matte', ' matte', |
| |
| ' Sergei', |
| |
| ' Melania', |
| |
| ' textures', 'Textures', |
| |
| 'etz', |
| |
| 'otte', |
| |
| ' Tek', 'tek', |
| |
| ' consistency', |
| |
| ' Melee', ' melee', |
| |
| ' scenery', |
| |
| ' landscapes', |
| |
| ' sweater', |
| |
| ' Univers', ' univers', |
| |
| ' tyre', |
| |
| ' spatial', |
| |
| ' tim', |
| |
| ' Ulster', |
| |
| ' Florence', |
| |
| ' palette', |
| |
| ' geometry', |
| |
| ' flavour', |
| |
| ' moisture', |
| |
| ' Trey', |
| |
| ' fract', ' Fract', |
| |
| ' textile', |
| |
| ' tyres', |
| |
| ' Tina', |
| |
| ' bruises', |
| |
| ' stamina', |
| |
| ' filesystem', |
| |
| ' tactile', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'tex-family', 1 |
|
|
| def prog_L07F21960(input_tokens): |
| """33 token families, 44 surface forms (train F1 of the uncapped optimum: 0.416). |
| Fires at 18,085 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to sports conferences""" |
| |
| fire = { |
| |
| ' League', |
| |
| ' division', ' Division', 'division', |
| |
| ' Conference', 'conference', |
| |
| ' Championship', |
| |
| ' SEC', 'SEC', |
| |
| ' AFC', |
| |
| ' NFC', |
| |
| ' Eastern', |
| |
| ' standings', |
| |
| ' bracket', ' brackets', |
| |
| ' continental', 'continental', ' Continental', |
| |
| ' Ten', |
| |
| 'FC', |
| |
| ' ACC', 'ACC', |
| |
| ' calendar', |
| |
| ' Playoffs', ' Playoff', |
| |
| ' corridor', ' Corridor', |
| |
| ' AL', |
| |
| ' CONC', |
| |
| ' divisions', |
| |
| ' NL', |
| |
| ' Premiership', |
| |
| ' northeastern', |
| |
| ' AHL', |
| |
| 'igue', |
| |
| 'acket', 'ackets', |
| |
| ' Northeast', |
| |
| ' abdominal', |
| |
| ' Hemisphere', |
| |
| ' relegation', |
| |
| ' NW', |
| |
| ' lockout', |
| |
| ' scheduling', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'league-family', 1 |
|
|
| def prog_L07F21961(input_tokens): |
| """25 token families, 48 surface forms (train F1 of the uncapped optimum: 0.440). |
| Fires at 35,096 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to specific legislative bills""" |
| |
| fire = { |
| |
| ' bill', ' Bill', ' bills', 'Bill', 'BILL', |
| |
| ' law', ' Law', ' Laws', 'laws', |
| |
| ' Act', ' Acts', |
| |
| ' legislation', ' Legislation', |
| |
| ' measure', ' Measure', |
| |
| ' constitution', ' Constitution', |
| |
| ' resolution', ' Resolution', |
| |
| ' amendment', ' amendments', ' Amendments', |
| |
| ' initiative', ' Initiative', |
| |
| ' motion', ' motions', |
| |
| ' ballot', |
| |
| ' Plan', |
| |
| ' ordinance', |
| |
| ' Prop', 'Prop', |
| |
| ' Order', ' Ord', |
| |
| ' SB', 'SB', |
| |
| ' HB', 'HB', |
| |
| ' Charter', ' charter', |
| |
| ' Budget', |
| |
| ' Section', 'Section', |
| |
| ' decree', |
| |
| ' Proposition', |
| |
| ' Declaration', |
| |
| ' Stat', 'Stat', |
| |
| ' Rules', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'bill-family', 1 |
|
|
| def prog_L07F21962(input_tokens): |
| """12 token families, 18 surface forms (train F1 of the uncapped optimum: 0.174). |
| Fires at 473,264 of 36,786,566 positions. |
| Fires at 1.29% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: specific formal organizations or associations mentioned in the text""" |
| |
| fire = { |
| |
| '\n', |
| |
| '.', |
| |
| 'The', ' The', |
| |
| '"', |
| |
| 'In', |
| |
| '."', ' ."', |
| |
| 'As', ' As', |
| |
| ').', ' ).', |
| |
| ',"', ' ,"', |
| |
| ']', |
| |
| 'According', ' According', |
| |
| 'He', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '\\n-family', 1 |
|
|
| def prog_L07F21963(input_tokens): |
| """28 token families, 43 surface forms (train F1 of the uncapped optimum: 0.378). |
| Fires at 47,706 of 36,786,566 positions.""" |
| |
| """prev SAE description: words related to observation-related structures or locations""" |
| |
| fire = { |
| |
| 'atory', |
| |
| 'room', 'rooms', |
| |
| 'ation', 'ATION', |
| |
| 'house', |
| |
| 'ary', 'ARY', |
| |
| 'ity', 'ITY', |
| |
| 'ator', 'atories', 'atorial', 'ATOR', |
| |
| 'ry', |
| |
| 'ival', 'ivals', |
| |
| 'ery', 'ERY', |
| |
| 'ium', 'IUM', |
| |
| ' sanctuary', ' Sanctuary', |
| |
| 'aries', 'arious', |
| |
| 'yard', 'yards', |
| |
| 'ities', |
| |
| 'ious', |
| |
| 'oir', |
| |
| ' barn', |
| |
| 'ocese', |
| |
| 'ows', |
| |
| 'ency', 'ENCY', |
| |
| 'arium', |
| |
| 'aph', |
| |
| 'asis', |
| |
| ' cathedral', ' Cathedral', |
| |
| 'oration', |
| |
| 'itarian', |
| |
| ' respiratory', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'atory-family', 1 |
|
|
| def prog_L07F21964(input_tokens): |
| """33 token families, 39 surface forms (train F1 of the uncapped optimum: 0.582). |
| Fires at 8,231 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases indicating negative impacts or consequences""" |
| |
| fire = { |
| |
| ' blow', ' blows', ' Blow', 'blow', |
| |
| ' punch', |
| |
| ' blew', |
| |
| ' setback', |
| |
| ' Pastebin', |
| |
| ' bombshell', |
| |
| ' Wim', |
| |
| ' Andrews', |
| |
| ' Cum', 'cum', |
| |
| ' insult', |
| |
| ' Tide', |
| |
| ' fro', |
| |
| ' payday', |
| |
| ' boon', |
| |
| ' slap', |
| |
| ' Craw', ' craw', |
| |
| ' Beam', |
| |
| ' swipe', |
| |
| 'breaker', ' breaker', |
| |
| ' candle', |
| |
| ' tumble', |
| |
| ' casualty', |
| |
| ' Pow', |
| |
| ' sham', |
| |
| 'ront', |
| |
| ' rag', |
| |
| ' bearer', |
| |
| 'bye', |
| |
| ' cutter', |
| |
| ' bleach', |
| |
| 'beard', |
| |
| 'bang', |
| |
| ' Grail', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'blow-family', 1 |
|
|
| def prog_L07F21965(input_tokens): |
| """3 token families, 6 surface forms (train F1 of the uncapped optimum: 0.094). |
| Fires at 41,859 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to organizations or institutions""" |
| |
| fire = { |
| |
| ' of', ' Of', 'of', ' OF', |
| |
| ' de', |
| |
| ' des', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'of-family', 1 |
|
|
| def prog_L07F21966(input_tokens): |
| """29 token families, 38 surface forms (train F1 of the uncapped optimum: 0.585). |
| Fires at 6,360 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to exit polls""" |
| |
| fire = { |
| |
| ' exit', 'Exit', ' exited', ' exiting', ' exits', ' Exit', 'exit', |
| |
| ' escape', ' Escape', |
| |
| ' entrance', |
| |
| ' Trevor', |
| |
| ' withdrawal', |
| |
| ' ejected', ' eject', |
| |
| ' elimination', |
| |
| 'enter', ' ENTER', |
| |
| ' Arri', |
| |
| ' eliminated', |
| |
| 'vals', |
| |
| ' unh', |
| |
| ' exodus', |
| |
| ' unl', |
| |
| ' Eyes', |
| |
| ' diseng', |
| |
| ' disembark', |
| |
| 'abeth', |
| |
| ' reversible', |
| |
| 'xit', |
| |
| ' Doors', |
| |
| ' departures', |
| |
| ' Investor', |
| |
| ' ignition', |
| |
| ' Obst', |
| |
| ' unfavorable', |
| |
| ' EB', |
| |
| ' abort', |
| |
| ' REM', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'exit-family', 1 |
|
|
| def prog_L07F21967(input_tokens): |
| """9 token families, 16 surface forms (train F1 of the uncapped optimum: 0.794). |
| Fires at 21,289 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to sources or reports in the text""" |
| |
| fire = { |
| |
| ' according', 'According', ' According', 'according', |
| |
| 'Based', ' Based', |
| |
| 'Per', |
| |
| ' Due', 'Due', |
| |
| 'Sources', |
| |
| ' Depending', 'Depending', |
| |
| ' Interestingly', 'Interestingly', |
| |
| 'rary', |
| |
| ' Needless', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'accord-family', 1 |
|
|
| def prog_L07F21968(input_tokens): |
| """40 token families, 49 surface forms (train F1 of the uncapped optimum: 0.624). |
| Fires at 11,657 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the Marine Corps""" |
| |
| fire = { |
| |
| ' Marine', ' marine', 'marine', |
| |
| 'Mar', 'MAR', |
| |
| ' Marines', ' Marin', ' marines', |
| |
| ' Oracle', 'Oracle', |
| |
| ' jungle', ' Jungle', |
| |
| ' Rogue', 'Rogue', |
| |
| ' Corps', |
| |
| ' PayPal', |
| |
| ' Ninja', |
| |
| ' amphib', |
| |
| ' Marion', |
| |
| ' Marina', |
| |
| ' Ranger', ' ranger', |
| |
| 'aroo', |
| |
| ' camouflage', |
| |
| 'Philipp', |
| |
| 'opter', |
| |
| ' Normandy', |
| |
| ' biologist', |
| |
| ' Mare', |
| |
| 'Marie', |
| |
| 'opus', |
| |
| ' SAS', |
| |
| 'antis', |
| |
| 'Scar', |
| |
| ' sailor', |
| |
| 'amar', |
| |
| ' butterfly', |
| |
| 'Brit', |
| |
| 'orge', |
| |
| ' SEAL', |
| |
| ' Koran', |
| |
| 'Mech', |
| |
| ' Serge', |
| |
| 'autical', |
| |
| ' Fisheries', |
| |
| ' Commando', |
| |
| 'ungle', |
| |
| ' Scarlett', |
| |
| 'iator', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'marine-family', 1 |
|
|
| def prog_L07F21969(input_tokens): |
| """12 token families, 16 surface forms (train F1 of the uncapped optimum: 0.094). |
| Fires at 52,196 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to societal structures and relationships""" |
| |
| fire = { |
| |
| ' day', |
| |
| ' thing', |
| |
| ' person', ' PERSON', |
| |
| ' step', ' Step', |
| |
| ' count', |
| |
| ' source', |
| |
| ' piece', ' Piece', |
| |
| ' teams', |
| |
| ' reason', 'Reason', |
| |
| ' hand', |
| |
| ' word', |
| |
| ' species', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'day-family', 1 |
|
|
| def prog_L07F21970(input_tokens): |
| """5 token families, 7 surface forms (train F1 of the uncapped optimum: 0.101). |
| Fires at 57,202 of 36,786,566 positions.""" |
| |
| """prev SAE description: intensifiers or modifiers, particularly the word "very" and its variations""" |
| |
| fire = { |
| |
| ' no', |
| |
| ' very', 'very', ' Very', |
| |
| ' once', |
| |
| ' quite', |
| |
| ' either', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'no-family', 1 |
|
|
| def prog_L07F21971(input_tokens): |
| """18 token families, 62 surface forms (train F1 of the uncapped optimum: 0.309). |
| Fires at 82,848 of 36,786,566 positions.""" |
| |
| """prev SAE description: terms related to beer types and their characteristics""" |
| |
| fire = { |
| |
| ' beer', ' beers', ' Beer', 'Beer', 'beer', |
| |
| ' Brewing', ' brew', ' brewing', ' Brew', ' brewers', ' brewed', ' brewer', 'brew', 'Brew', ' Brewers', |
| |
| ' brewery', ' Brewery', |
| |
| ' craft', ' Craft', 'crafted', |
| |
| ' brand', ' brands', ' Brands', |
| |
| ' flavors', ' flavor', ' flavorful', ' flavored', ' Flavor', |
| |
| ' sweet', ' Sweet', ' sweetness', 'sweet', 'Sweet', |
| |
| ' flagship', |
| |
| ' wine', ' Wine', 'wine', |
| |
| ' Ale', ' ale', 'Ale', ' ALE', |
| |
| ' Hop', ' hops', ' hop', 'Hop', |
| |
| ' barrel', ' barrels', ' Barrel', |
| |
| ' tap', ' Tap', ' taps', 'tap', |
| |
| ' breweries', |
| |
| ' coffee', ' Coffee', |
| |
| ' tasting', ' tasted', ' tastes', ' tast', |
| |
| ' Company', |
| |
| ' dist', ' Dist', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'beer-family', 1 |
|
|
| def prog_L07F21972(input_tokens): |
| """11 token families, 13 surface forms (train F1 of the uncapped optimum: 0.091). |
| Fires at 71,315 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to specific groups or entities in various contexts""" |
| |
| fire = { |
| |
| ' them', ' themes', |
| |
| ' those', |
| |
| ' these', ' THESE', |
| |
| ' players', |
| |
| ' researchers', |
| |
| ' reasons', |
| |
| ' victims', |
| |
| ' friends', |
| |
| ' scientists', |
| |
| ' whom', |
| |
| ' organizations', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'them-family', 1 |
|
|
| def prog_L07F21973(input_tokens): |
| """8 token families, 20 surface forms (train F1 of the uncapped optimum: 0.150). |
| Fires at 57,175 of 36,786,566 positions.""" |
| |
| """prev SAE description: programming functions and code-related structures""" |
| |
| fire = { |
| |
| 'x', ' X', ' x', |
| |
| '_', |
| |
| '://', |
| |
| ' log', 'log', ' logs', ' Log', 'Log', |
| |
| 'of', |
| |
| 'www', ' www', |
| |
| '()', ' ()', |
| |
| ' function', 'function', ' Function', 'func', ' func', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'x-family', 1 |
|
|
| def prog_L07F21974(input_tokens): |
| """18 token families, 29 surface forms (train F1 of the uncapped optimum: 0.205). |
| Fires at 106,196 of 36,786,566 positions.""" |
| |
| """prev SAE description: various forms of sports and entertainment activities""" |
| |
| fire = { |
| |
| ' game', |
| |
| ' games', |
| |
| ' video', ' videos', ' VIDEO', |
| |
| ' book', ' books', 'books', |
| |
| ' music', |
| |
| ' football', |
| |
| ' food', ' foods', |
| |
| ' beer', ' beers', |
| |
| ' movie', |
| |
| ' movies', ' Movies', |
| |
| ' flag', ' flags', |
| |
| ' marijuana', |
| |
| ' coffee', |
| |
| ' ball', ' balls', 'balls', |
| |
| ' training', |
| |
| ' surgery', |
| |
| ' drink', ' drinks', |
| |
| ' drugs', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'game-family', 1 |
|
|
| def prog_L07F21975(input_tokens): |
| """12 token families, 18 surface forms (train F1 of the uncapped optimum: 0.127). |
| Fires at 52,046 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to a specific geographic location, particularly Timbuktoo and its cultural significance""" |
| |
| fire = { |
| |
| 'y', |
| |
| 'an', |
| |
| 'on', |
| |
| ' al', |
| |
| 'el', |
| |
| ' Disney', |
| |
| 'IS', |
| |
| 'pe', ' Pe', 'PE', |
| |
| 'mer', 'MER', |
| |
| 'per', 'pered', 'pers', |
| |
| ' Kaine', |
| |
| ' Cook', ' Cookies', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'y-family', 1 |
|
|
| def prog_L07F21976(input_tokens): |
| """17 token families, 24 surface forms (train F1 of the uncapped optimum: 0.116). |
| Fires at 523,278 of 36,786,566 positions. |
| Fires at 1.42% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: instances of teaching or educational experiences""" |
| |
| fire = { |
| |
| ' in', |
| |
| ' are', |
| |
| ' as', |
| |
| ' from', |
| |
| ' by', |
| |
| ' be', |
| |
| ' have', 'have', |
| |
| ' were', 'were', ' Were', |
| |
| ' which', 'which', |
| |
| ' not', |
| |
| 't', ' t', |
| |
| ' than', |
| |
| ' up', |
| |
| ' into', |
| |
| ' them', 'them', |
| |
| ' off', ' offer', |
| |
| ' only', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'in-family', 1 |
|
|
| def prog_L07F21977(input_tokens): |
| """1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.560). |
| Fires at 47,836 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the term "this" within various contexts""" |
| |
| fire = { |
| |
| ' this', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'thi-family', 1 |
|
|
| def prog_L07F21978(input_tokens): |
| """3 token families, 3 surface forms (train F1 of the uncapped optimum: 0.061). |
| Fires at 18,697 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases indicating something that has not yet occurred or been confirmed""" |
| |
| fire = { |
| |
| ' yet', |
| |
| ' never', |
| |
| 'to', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'yet-family', 1 |
|
|
| def prog_L07F21979(input_tokens): |
| """11 token families, 15 surface forms (train F1 of the uncapped optimum: 0.048). |
| Fires at 102,554 of 36,786,566 positions.""" |
| |
| """prev SAE description: instances of hypothetical situations or conditional statements""" |
| |
| fire = { |
| |
| ' to', |
| |
| ' out', |
| |
| ' were', |
| |
| ' them', |
| |
| ' me', |
| |
| ' go', ' GO', |
| |
| ' do', |
| |
| ' every', ' EVERY', |
| |
| ' any', |
| |
| ' going', |
| |
| ' done', 'done', ' Done', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'to-family', 1 |
|
|
| def prog_L07F21980(input_tokens): |
| """1 token families, 2 surface forms (train F1 of the uncapped optimum: 0.597). |
| Fires at 11,911 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to actions or tasks that are being done or completed""" |
| |
| fire = { |
| |
| ' things', 'things', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'thing-family', 1 |
|
|
| def prog_L07F21981(input_tokens): |
| """15 token families, 33 surface forms (train F1 of the uncapped optimum: 0.121). |
| Fires at 106,601 of 36,786,566 positions.""" |
| |
| """prev SAE description: topics related to societal issues and structures""" |
| |
| fire = { |
| |
| ' social', ' Social', ' soc', |
| |
| ' media', |
| |
| ' security', |
| |
| ' health', 'health', |
| |
| ' family', 'family', |
| |
| ' political', |
| |
| ' economic', ' economies', 'economic', |
| |
| ' culture', ' Culture', |
| |
| ' technology', 'technology', |
| |
| ' gender', ' Gender', ' genders', |
| |
| ' rights', 'rights', |
| |
| ' education', 'education', |
| |
| ' politics', ' Politics', 'politics', |
| |
| ' climate', ' Climate', 'climate', |
| |
| ' personal', ' Personal', 'personal', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'soc-family', 1 |
|
|
| def prog_L07F21982(input_tokens): |
| """22 token families, 30 surface forms (train F1 of the uncapped optimum: 0.648). |
| Fires at 10,623 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases that describe something being completely occupied or filled""" |
| |
| fire = { |
| |
| ' filled', ' fill', ' filling', 'filled', ' Fill', ' fills', 'fill', 'Fill', |
| |
| 'cel', |
| |
| ' Ing', 'Ing', |
| |
| ' populated', |
| |
| ' flooded', |
| |
| ' stuffed', |
| |
| 'icken', |
| |
| 'illed', |
| |
| ' Og', |
| |
| ' cubic', |
| |
| ' decorated', |
| |
| 'ayette', |
| |
| ' Dungeons', |
| |
| ' grat', |
| |
| ' Excel', |
| |
| ' cavern', |
| |
| ' Pens', |
| |
| ' overflowing', |
| |
| 'onge', |
| |
| ' Fountain', |
| |
| 'empty', |
| |
| 'capacity', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'fill-family', 1 |
|
|
| def prog_L07F21983(input_tokens): |
| """1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.472). |
| Fires at 87,707 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases that involve providing or getting something for someone or something""" |
| |
| fire = { |
| |
| ' for', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'for-family', 1 |
|
|
| def prog_L07F21984(input_tokens): |
| """1 token families, 2 surface forms (train F1 of the uncapped optimum: 0.223). |
| Fires at 80,362 of 36,786,566 positions.""" |
| |
| """prev SAE description: instances of the word "in" indicating various contexts or categories""" |
| |
| fire = { |
| |
| ' in', ' IN', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'in-family', 1 |
|
|
| def prog_L07F21985(input_tokens): |
| """12 token families, 17 surface forms (train F1 of the uncapped optimum: 0.760). |
| Fires at 7,303 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases related to political majorities""" |
| |
| fire = { |
| |
| ' majority', ' Majority', 'majority', ' majorities', |
| |
| ' minority', ' Minority', |
| |
| ' overwhelmingly', |
| |
| 'thirds', ' thirds', |
| |
| ' bulk', |
| |
| ' predominantly', |
| |
| 'large', |
| |
| ' landslide', |
| |
| ' plurality', |
| |
| ' Rasmussen', |
| |
| 'orum', |
| |
| 'ority', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'major-family', 1 |
|
|
| def prog_L07F21986(input_tokens): |
| """14 token families, 20 surface forms (train F1 of the uncapped optimum: 0.062). |
| Fires at 84,921 of 36,786,566 positions.""" |
| |
| """prev SAE description: statements that emphasize negation or denial""" |
| |
| fire = { |
| |
| ' be', |
| |
| ' not', ' NOT', |
| |
| ' just', ' JUST', |
| |
| ' good', |
| |
| ' very', ' VERY', |
| |
| ' part', ' partly', |
| |
| ' best', ' BEST', |
| |
| ' isn', |
| |
| ' really', ' REALLY', |
| |
| ' great', |
| |
| ' kind', |
| |
| ' something', |
| |
| ' case', |
| |
| ' thing', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'be-family', 1 |
|
|
| def prog_L07F21987(input_tokens): |
| """2 token families, 2 surface forms (train F1 of the uncapped optimum: 0.348). |
| Fires at 25,997 of 36,786,566 positions.""" |
| |
| """prev SAE description: numerical values associated with distances or measurements""" |
| |
| fire = { |
| |
| ')', |
| |
| '%)', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return ')-family', 1 |
|
|
| def prog_L07F21988(input_tokens): |
| """16 token families, 42 surface forms (train F1 of the uncapped optimum: 0.275). |
| Fires at 208,219 of 36,786,566 positions. |
| Fires at 0.57% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: proper nouns, particularly names of individuals mentioned in criminal contexts""" |
| |
| fire = { |
| |
| ' John', 'John', 'JOHN', ' JOHN', |
| |
| ' David', 'David', |
| |
| ' Michael', 'Michael', |
| |
| 'old', 'Old', |
| |
| ' James', 'James', ' Jamal', 'Jam', |
| |
| 'ie', |
| |
| ' Robert', 'Robert', ' Roberts', |
| |
| ' William', ' Williams', 'William', 'Williams', |
| |
| ' Chris', 'Chris', |
| |
| ' Richard', 'Richard', ' Richards', |
| |
| ' Mike', 'Mike', |
| |
| ' Scott', 'Scott', |
| |
| 'ah', 'Ah', ' Ah', |
| |
| ' Thomas', 'Thomas', |
| |
| ' Daniel', 'Daniel', ' Daniels', |
| |
| ' Eric', 'Eric', 'eric', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'john-family', 1 |
|
|
| def prog_L07F21989(input_tokens): |
| """46 token families, 53 surface forms (train F1 of the uncapped optimum: 0.840). |
| Fires at 14,494 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to Italy and the Italian language""" |
| |
| fire = { |
| |
| ' Italian', 'Italian', ' Italians', |
| |
| ' Italy', 'Italy', |
| |
| ' Rome', |
| |
| ' Milan', |
| |
| ' Mediterranean', |
| |
| ' Ferrari', |
| |
| ' Rodriguez', |
| |
| ' Andrea', |
| |
| ' Juventus', |
| |
| ' mafia', ' Mafia', |
| |
| ' Angelo', 'angelo', 'Angelo', |
| |
| ' fascist', ' fascists', |
| |
| ' Roma', |
| |
| ' Maharashtra', |
| |
| ' Ortiz', |
| |
| 'coni', |
| |
| ' pasta', |
| |
| ' Fi', |
| |
| ' Gi', |
| |
| ' Roberto', |
| |
| ' Serie', |
| |
| ' Lomb', |
| |
| ' Fitzgerald', |
| |
| ' Florence', |
| |
| ' Frances', |
| |
| ' Venice', |
| |
| ' Via', |
| |
| ' Farrell', |
| |
| ' Renaissance', |
| |
| ' Carlo', |
| |
| 'ione', |
| |
| ' Illustrated', |
| |
| ' Matte', |
| |
| ' Lorenzo', |
| |
| ' Fiat', |
| |
| ' fascism', |
| |
| ' Rossi', |
| |
| ' Alps', |
| |
| ' Circus', |
| |
| ' Triumph', |
| |
| ' Romans', |
| |
| ' Stef', |
| |
| ' Salv', |
| |
| ' Luigi', |
| |
| ' Marino', |
| |
| ' Naples', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'italian-family', 1 |
|
|
| def prog_L07F21990(input_tokens): |
| """37 token families, 43 surface forms (train F1 of the uncapped optimum: 0.521). |
| Fires at 4,597 of 36,786,566 positions.""" |
| |
| """prev SAE description: a specific character or symbol used in various contexts""" |
| |
| fire = { |
| |
| 'pert', ' pert', |
| |
| ' certified', ' Certified', |
| |
| ' Elliot', |
| |
| ' chops', ' chop', ' Chop', |
| |
| 'oint', |
| |
| 'oped', |
| |
| ' Matrix', 'Matrix', |
| |
| ' multit', |
| |
| 'inates', |
| |
| 'stant', |
| |
| 'arter', |
| |
| 'phy', |
| |
| ' neurot', |
| |
| '000000', |
| |
| 'mmm', |
| |
| 'ophe', |
| |
| 'ensor', |
| |
| ' Marty', |
| |
| 'verbal', |
| |
| 'vec', ' vec', |
| |
| '455', |
| |
| ' doct', |
| |
| 'cooked', |
| |
| 'eret', |
| |
| 'cision', |
| |
| 'ometry', |
| |
| ' Δ', |
| |
| 'ochond', |
| |
| '::::::::', |
| |
| '────────', |
| |
| 'ynt', |
| |
| 'mmmm', |
| |
| 'entin', |
| |
| 'チ', |
| |
| 'urat', |
| |
| '294', |
| |
| '456', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'pert-family', 1 |
|
|
| def prog_L07F21991(input_tokens): |
| """34 token families, 45 surface forms (train F1 of the uncapped optimum: 0.548). |
| Fires at 10,865 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to patents and intellectual property""" |
| |
| fire = { |
| |
| ' patent', ' patents', ' Patent', ' patented', |
| |
| ' trademark', ' trademarks', |
| |
| ' Pir', ' pir', 'Pir', |
| |
| ' intellectual', ' Intellectual', |
| |
| ' invention', |
| |
| ' inventor', |
| |
| ' Prize', |
| |
| ' tariffs', ' tariff', |
| |
| ' Patterson', |
| |
| ' vulnerabilities', |
| |
| ' pardon', |
| |
| ' proprietary', |
| |
| ' Theodore', |
| |
| ' Marvin', |
| |
| ' Treaty', |
| |
| ' warranty', ' Warranty', |
| |
| 'emark', |
| |
| ' Trad', |
| |
| ' inventions', |
| |
| 'otechnology', |
| |
| ' Institutes', |
| |
| ' Regulatory', |
| |
| 'yrights', 'yright', |
| |
| ' monopoly', |
| |
| ' Property', |
| |
| ' prototypes', |
| |
| 'ellectual', |
| |
| ' Redmond', |
| |
| ' RFC', 'RFC', |
| |
| 'prints', |
| |
| ' dissertation', |
| |
| 'ulnerability', |
| |
| 'proven', |
| |
| ' royalties', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'patent-family', 1 |
|
|
| def prog_L07F21992(input_tokens): |
| """28 token families, 43 surface forms (train F1 of the uncapped optimum: 0.224). |
| Fires at 263,264 of 36,786,566 positions. |
| Fires at 0.72% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: words related to launching or initiating actions or campaigns""" |
| |
| fire = { |
| |
| ' been', |
| |
| ' announced', ' announcing', ' announces', |
| |
| ' released', ' releasing', |
| |
| ' found', |
| |
| ' come', |
| |
| ' revealed', ' reveals', |
| |
| ' confirmed', ' confirm', ' confirms', ' confirming', |
| |
| ' decided', |
| |
| ' taken', |
| |
| ' launched', ' launching', |
| |
| ' become', |
| |
| ' signed', 'signed', |
| |
| ' begun', |
| |
| ' arrested', |
| |
| ' given', |
| |
| ' reportedly', ' reporting', |
| |
| ' agreed', |
| |
| ' charged', |
| |
| ' warned', ' warning', ' warns', |
| |
| ' received', |
| |
| ' issued', ' issuing', |
| |
| ' learned', |
| |
| ' gone', |
| |
| ' created', |
| |
| ' died', |
| |
| ' filed', |
| |
| ' joined', ' joining', |
| |
| ' ordered', ' ordering', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'been-family', 1 |
|
|
| def prog_L07F21993(input_tokens): |
| """12 token families, 19 surface forms (train F1 of the uncapped optimum: 0.077). |
| Fires at 155,728 of 36,786,566 positions. |
| Fires at 0.42% of positions, and |
| token rules do not serve features this dense: median test F1 0.086 |
| across the 37,173 features above 0.39% of positions, against 0.763 for |
| the 76,283 below 0.016%. Read this fire set as the best available |
| approximation, not as what the feature does.""" |
| |
| """prev SAE description: terms related to legal charges and investigations""" |
| |
| fire = { |
| |
| ' is', |
| |
| ' are', |
| |
| ' will', |
| |
| ' been', |
| |
| ' which', 'which', |
| |
| ' were', 'were', |
| |
| ' would', |
| |
| ' including', ' includes', ' included', |
| |
| ' could', |
| |
| ' charges', 'charges', ' Charges', 'Charges', |
| |
| ' came', |
| |
| ' issued', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'is-family', 1 |
|
|
| def prog_L07F21994(input_tokens): |
| """4 token families, 14 surface forms (train F1 of the uncapped optimum: 0.594). |
| Fires at 19,753 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases related to the necessity of help or assistance""" |
| |
| fire = { |
| |
| ' need', ' needed', ' needs', ' Need', ' needing', 'Need', 'needed', 'need', ' Needs', 'needs', ' NEED', |
| |
| ' deserve', |
| |
| ' deserved', |
| |
| ' necessity', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'need-family', 1 |
|
|
| def prog_L07F21995(input_tokens): |
| """1 token families, 4 surface forms (train F1 of the uncapped optimum: 0.106). |
| Fires at 8,366 of 36,786,566 positions.""" |
| |
| """prev SAE description: the phrase "could have" and its variations, often in a context suggesting missed opportunities or potential actions""" |
| |
| fire = { |
| |
| ' have', ' Have', 'have', ' HAVE', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'have-family', 1 |
|
|
| def prog_L07F21996(input_tokens): |
| """50 token families, 56 surface forms (train F1 of the uncapped optimum: 0.716). |
| Fires at 35,462 of 36,786,566 positions.""" |
| |
| """prev SAE description: mentions of academic institutions, particularly Cornell University""" |
| |
| fire = { |
| |
| ' ACC', 'ACC', |
| |
| ' Armstrong', |
| |
| ' Hannity', ' Hann', |
| |
| 'ahn', |
| |
| ' Gandhi', |
| |
| 'Activ', |
| |
| ' Cont', |
| |
| ' Lauren', |
| |
| 'allah', |
| |
| ' Nigel', |
| |
| ' Fairfax', |
| |
| ' Cornell', |
| |
| ' Rebecca', |
| |
| 'BD', ' BD', 'bd', |
| |
| ' Gross', |
| |
| ' Canon', |
| |
| ' Clemson', |
| |
| ' Hayes', |
| |
| ' Thomson', |
| |
| ' Durant', |
| |
| ' EUR', ' Eur', |
| |
| ' Goodman', |
| |
| ' Donovan', |
| |
| ' Allan', |
| |
| ' Banks', |
| |
| 'INGTON', |
| |
| ' Ghana', |
| |
| ' Lennon', |
| |
| ' Goodell', |
| |
| ' Elliott', |
| |
| ' Neg', |
| |
| ' Fargo', |
| |
| ' overweight', |
| |
| ' Swan', |
| |
| ' Garner', |
| |
| ' cloth', |
| |
| ' Randall', |
| |
| ' Wagner', |
| |
| ' Allison', |
| |
| ' contr', ' CONTR', |
| |
| 'hn', |
| |
| ' Gregg', |
| |
| ' Cameroon', |
| |
| ' TT', |
| |
| ' Greenpeace', |
| |
| 'anya', |
| |
| ' Gallagher', |
| |
| ' Holden', |
| |
| ' Dunn', |
| |
| ' Noel', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'acc-family', 1 |
|
|
| def prog_L07F21997(input_tokens): |
| """19 token families, 31 surface forms (train F1 of the uncapped optimum: 0.139). |
| Fires at 115,834 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases indicating movement or progression through space or time""" |
| |
| fire = { |
| |
| ' on', ' ON', |
| |
| ' into', ' Into', ' INTO', |
| |
| ' over', |
| |
| ' up', ' UP', |
| |
| ' through', |
| |
| ' around', |
| |
| ' off', ' Off', ' OFF', |
| |
| ' back', ' BACK', |
| |
| ' down', ' DOWN', |
| |
| ' across', |
| |
| ' away', ' Away', |
| |
| ' along', |
| |
| ' ahead', ' Ahead', |
| |
| ' toward', ' towards', ' Towards', |
| |
| ' together', |
| |
| ' forward', |
| |
| ' behind', |
| |
| ' onto', |
| |
| ' closer', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'on-family', 1 |
|
|
| def prog_L07F21998(input_tokens): |
| """48 token families, 59 surface forms (train F1 of the uncapped optimum: 0.751). |
| Fires at 18,925 of 36,786,566 positions.""" |
| |
| """prev SAE description: mentions of specific places and individuals, particularly associated with the names "Knox" and "Tate." """ |
| |
| fire = { |
| |
| ' bars', ' Bars', |
| |
| ' Athens', ' Athen', |
| |
| ' Gregory', |
| |
| ' Watt', ' Watts', |
| |
| ' Ridge', |
| |
| ' punk', ' Punk', |
| |
| ' Barrett', |
| |
| ' madness', ' Madness', ' MAD', |
| |
| ' Memphis', |
| |
| 'Tok', ' Tok', |
| |
| 'itsu', |
| |
| ' Knox', |
| |
| ' Bent', |
| |
| ' Kaz', |
| |
| ' ransomware', |
| |
| ' Ortiz', |
| |
| ' Hardy', |
| |
| ' yen', |
| |
| ' Hicks', |
| |
| ' Williamson', |
| |
| 'istrates', |
| |
| ' Tate', |
| |
| ' Kingston', |
| |
| ' Norton', |
| |
| ' Berry', |
| |
| 'ollah', |
| |
| ' blues', |
| |
| ' moons', ' Moons', |
| |
| ' Kerr', |
| |
| ' Bates', |
| |
| ' Pike', |
| |
| ' Raven', 'Raven', |
| |
| ' mysteries', |
| |
| ' Thatcher', |
| |
| 'ayette', |
| |
| 'Kn', |
| |
| ' Ton', 'Ton', |
| |
| ' Maid', |
| |
| ' Byrne', |
| |
| ' Gin', |
| |
| ' Loll', |
| |
| 'ß', |
| |
| ' Towns', |
| |
| ' Ney', |
| |
| ' Grimes', |
| |
| ' Ware', ' ware', |
| |
| ' Tos', |
| |
| ' Meredith', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'bar-family', 1 |
|
|
| def prog_L07F21999(input_tokens): |
| """44 token families, 87 surface forms (train F1 of the uncapped optimum: 0.415). |
| Fires at 91,443 of 36,786,566 positions.""" |
| |
| """prev SAE description: terms related to changes in status or position within an organization or context""" |
| |
| fire = { |
| |
| ' release', 'release', |
| |
| ' move', |
| |
| ' transfer', ' transfers', ' Transfer', 'Transfer', 'transfer', |
| |
| ' return', ' Return', ' Returns', 'Return', 'Returns', 'return', ' Returning', |
| |
| ' launch', 'launch', 'Launch', |
| |
| ' signing', ' signings', 'sign', |
| |
| 'ation', 'ATION', |
| |
| ' trade', 'trade', |
| |
| 'ization', |
| |
| ' sale', |
| |
| ' nomination', |
| |
| 'ment', |
| |
| ' resignation', |
| |
| ' arrest', ' Arrest', |
| |
| ' award', ' awards', ' awarding', |
| |
| ' departure', |
| |
| ' appointment', ' appointments', |
| |
| ' acquisition', ' Acquisition', |
| |
| ' removal', ' Removal', ' Removed', 'Removed', |
| |
| ' exit', ' exits', ' Exit', 'exit', |
| |
| ' extension', |
| |
| ' expansion', ' Expansion', |
| |
| ' selection', ' Selection', 'selection', |
| |
| ' retirement', ' Retirement', |
| |
| ' closure', 'closure', |
| |
| ' transition', ' Transition', |
| |
| ' adoption', |
| |
| ' registration', ' Registration', 'Registration', |
| |
| ' deployment', ' deployments', ' Deploy', |
| |
| ' divorce', |
| |
| ' endorsement', ' endorsements', |
| |
| 'isation', |
| |
| ' repeal', |
| |
| ' suspension', |
| |
| ' confirmation', |
| |
| ' promotion', |
| |
| ' firing', |
| |
| ' membership', ' Membership', |
| |
| ' upgrade', 'Upgrade', |
| |
| ' recall', ' Recall', |
| |
| ' withdrawal', |
| |
| ' diagnosis', |
| |
| ' conversion', ' Conversion', |
| |
| ' merger', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'release-family', 1 |
|
|
