| """GPT-2 SAE last-token programs, layer 8 -- 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_L08F21900(input_tokens): |
| """49 token families, 50 surface forms (train F1 of the uncapped optimum: 0.728). |
| Fires at 11,656 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to individuals named Michelle, particularly Michelle Obama""" |
| |
| fire = { |
| |
| ' Michelle', 'Michelle', |
| |
| ' Jesse', |
| |
| ' Huckabee', |
| |
| ' Emma', |
| |
| ' Abraham', |
| |
| ' Ivanka', |
| |
| ' Grace', |
| |
| ' Stephanie', |
| |
| ' Debbie', |
| |
| ' Michele', |
| |
| ' Melania', |
| |
| ' Megan', |
| |
| 'issa', |
| |
| ' Regina', |
| |
| ' Mickey', |
| |
| ' Rainbow', |
| |
| ' Sandra', |
| |
| ' Rahul', |
| |
| ' Maggie', |
| |
| ' Andrews', |
| |
| ' Hank', |
| |
| ' Marilyn', |
| |
| ' Malik', |
| |
| ' Sonia', |
| |
| 'ieg', |
| |
| ' Ng', |
| |
| ' Gloria', |
| |
| ' Gutierrez', |
| |
| 'iang', |
| |
| ' Denise', |
| |
| ' Gina', |
| |
| ' Manny', |
| |
| ' Valerie', |
| |
| 'annie', |
| |
| ' Heidi', |
| |
| ' Nina', |
| |
| ' Gibbs', |
| |
| ' Lois', |
| |
| ' Alicia', |
| |
| ' Imag', |
| |
| 'hillary', |
| |
| ' Harriet', |
| |
| ' Sophia', |
| |
| ' Prairie', |
| |
| ' grandma', |
| |
| ' Simone', |
| |
| ' Sasha', |
| |
| 'Category', |
| |
| ' Jiang', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'michelle-family', 1 |
|
|
| def prog_L08F21901(input_tokens): |
| """40 token families, 58 surface forms (train F1 of the uncapped optimum: 0.623). |
| Fires at 10,853 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to propositions, particularly numbered ones""" |
| |
| fire = { |
| |
| ' prop', ' Prop', 'Prop', 'prop', ' props', |
| |
| ' Ralph', |
| |
| ' proposition', ' Proposition', |
| |
| ' proponents', ' proponent', |
| |
| ' decl', 'Decl', 'decl', ' Decl', |
| |
| ' Opp', 'Opp', |
| |
| ' Janet', |
| |
| ' Pant', ' pant', |
| |
| 'LGBT', |
| |
| ' Champ', 'Champ', |
| |
| ' Prot', ' PROT', |
| |
| 'vertising', |
| |
| ' Gavin', |
| |
| ' Marin', |
| |
| ' Enc', |
| |
| ' SB', |
| |
| ' Californ', |
| |
| ' Porn', |
| |
| ' Penny', |
| |
| ' DOM', 'DOM', |
| |
| ' Marvin', |
| |
| ' Flag', 'FLAG', |
| |
| ' PRE', |
| |
| ' postp', |
| |
| ' Conv', 'conv', |
| |
| 'endra', |
| |
| ' Proceedings', |
| |
| ' Adv', ' ADV', |
| |
| ' FP', |
| |
| ' Patent', |
| |
| ' PG', |
| |
| 'ardon', |
| |
| ' propulsion', |
| |
| ' Spons', |
| |
| ' Philos', |
| |
| 'alph', |
| |
| ' phr', |
| |
| ' controvers', |
| |
| ' Vend', |
| |
| 'Arg', 'Args', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'prop-family', 1 |
|
|
| def prog_L08F21902(input_tokens): |
| """12 token families, 46 surface forms (train F1 of the uncapped optimum: 0.097). |
| Fires at 342,548 of 36,786,566 positions. |
| Fires at 0.93% 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: references to friendship and social connections""" |
| |
| fire = { |
| |
| ' of', |
| |
| 's', 'és', |
| |
| ' (', |
| |
| ' years', ' year', 'years', ' YEAR', |
| |
| ' or', |
| |
| ' man', 'man', ' MAN', ' mans', 'mans', |
| |
| ' 1', '1', |
| |
| ' world', ' worlds', 'world', ' Worlds', 'worldly', ' WORLD', |
| |
| ' group', ' groups', ' Group', 'group', ' grouping', ' Groups', 'groups', ' grouped', |
| |
| ' team', ' teams', ' Team', 'team', ' Teams', |
| |
| ' old', ' oldest', ' older', ' Old', 'olds', ' olds', 'older', |
| |
| ' day', ' days', ' Day', 'days', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'of-family', 1 |
|
|
| def prog_L08F21903(input_tokens): |
| """2 token families, 2 surface forms (train F1 of the uncapped optimum: 0.037). |
| Fires at 25,925 of 36,786,566 positions.""" |
| |
| """prev SAE description: sentence-ending punctuation, specifically focusing on periods and commas""" |
| |
| fire = { |
| |
| '.', |
| |
| '."', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '.-family', 1 |
|
|
| def prog_L08F21904(input_tokens): |
| """6 token families, 12 surface forms (train F1 of the uncapped optimum: 0.257). |
| Fires at 41,407 of 36,786,566 positions.""" |
| |
| """prev SAE description: repeated phrases emphasizing possession or inquiries in a conversational context""" |
| |
| fire = { |
| |
| ' have', 'Have', ' Have', ' HAVE', |
| |
| 've', "'ve", |
| |
| ' haven', |
| |
| 'Having', ' Having', |
| |
| ' Got', 'Got', |
| |
| ' hadn', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'have-family', 1 |
|
|
| def prog_L08F21905(input_tokens): |
| """18 token families, 57 surface forms (train F1 of the uncapped optimum: 0.170). |
| Fires at 102,811 of 36,786,566 positions.""" |
| |
| """prev SAE description: direct speech or dialogue within the text""" |
| |
| fire = { |
| |
| ' said', 'said', |
| |
| ' say', ' saying', ' Say', ' Saying', 'say', ' SAY', |
| |
| ' asked', ' ask', ' asking', ' asks', ' Asked', |
| |
| ' speaking', ' speak', ' speaks', ' Speaking', 'speak', ' Speak', |
| |
| ' gave', |
| |
| ' words', 'word', |
| |
| ' shouting', ' shouted', ' shout', ' shouts', |
| |
| ' talking', ' talked', 'talk', 'talking', |
| |
| ' spoke', |
| |
| ' telling', ' tells', 'tell', |
| |
| ' giving', ' gives', 'Giving', |
| |
| ' delivered', ' delivering', |
| |
| ' singing', ' sing', ' sings', 'sing', |
| |
| ' screaming', ' screamed', ' scream', ' screams', |
| |
| ' perform', ' performing', ' performs', |
| |
| ' yelling', ' yelled', ' yell', ' yells', |
| |
| ' offered', |
| |
| ' deliver', ' delivers', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'said-family', 1 |
|
|
| def prog_L08F21906(input_tokens): |
| """18 token families, 29 surface forms (train F1 of the uncapped optimum: 0.110). |
| Fires at 22,452 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases related to accounting and percentages""" |
| |
| fire = { |
| |
| ' for', |
| |
| ' represent', ' represents', ' representing', ' represented', |
| |
| ' explain', ' explains', |
| |
| ' behind', |
| |
| ' provide', |
| |
| ' responsible', |
| |
| ' provided', ' provides', 'provided', |
| |
| ' caused', |
| |
| ' dominated', ' dominating', ' dominates', |
| |
| ' comprises', ' comprising', |
| |
| ' comprise', |
| |
| ' handle', |
| |
| ' accounted', ' accounting', |
| |
| ' captured', ' capturing', |
| |
| ' garnered', |
| |
| ' constitute', |
| |
| ' comprised', |
| |
| ' dominate', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'for-family', 1 |
|
|
| def prog_L08F21907(input_tokens): |
| """39 token families, 60 surface forms (train F1 of the uncapped optimum: 0.621). |
| Fires at 15,757 of 36,786,566 positions.""" |
| |
| """prev SAE description: terms related to animation and animated content""" |
| |
| fire = { |
| |
| ' anime', ' Anime', |
| |
| ' animated', ' Animated', |
| |
| ' animation', ' Animation', 'Animation', |
| |
| ' Animal', ' anim', 'anim', ' Anim', 'Anim', 'animal', |
| |
| ' GIF', ' gif', 'gif', |
| |
| ' cartoon', ' cartoons', ' Cartoon', |
| |
| ' Yang', 'Yang', |
| |
| ' limbs', ' limb', |
| |
| 'atar', 'atars', |
| |
| 'action', 'ACTION', |
| |
| ' UI', |
| |
| ' Frozen', |
| |
| ' Miy', |
| |
| ' Arnold', |
| |
| ' embryos', ' embryo', |
| |
| ' animations', |
| |
| ' Avatar', ' avatar', |
| |
| ' Toy', |
| |
| 'acer', |
| |
| ' amphib', |
| |
| ' Pixar', |
| |
| ' skeleton', |
| |
| ' cinematic', |
| |
| 'bender', 'bending', |
| |
| ' apparel', |
| |
| ' anatomy', |
| |
| ' immortal', |
| |
| 'rozen', |
| |
| ' frogs', |
| |
| 'airo', |
| |
| ' Musical', |
| |
| ' CGI', |
| |
| ' Harmon', |
| |
| ' CG', 'CG', |
| |
| ' decap', |
| |
| ' Harmony', |
| |
| ' Hanna', |
| |
| 'motion', |
| |
| ' scripted', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'anime-family', 1 |
|
|
| def prog_L08F21908(input_tokens): |
| """8 token families, 32 surface forms (train F1 of the uncapped optimum: 0.118). |
| Fires at 290,405 of 36,786,566 positions. |
| Fires at 0.79% 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: key themes related to education, policy, and social issues""" |
| |
| fire = { |
| |
| '.', |
| |
| ' American', ' Americans', 'American', 'Americans', |
| |
| ' government', ' Government', ' governments', ' governance', 'government', 'Government', ' Governments', |
| |
| ' state', ' State', ' statement', ' statements', 'state', 'State', ' Statement', ' STATE', |
| |
| ' media', ' Media', 'media', |
| |
| ' US', |
| |
| ' Google', 'Google', ' google', |
| |
| ' world', ' World', 'world', 'World', ' Worlds', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '.-family', 1 |
|
|
| def prog_L08F21909(input_tokens): |
| """9 token families, 18 surface forms (train F1 of the uncapped optimum: 0.077). |
| Fires at 160,106 of 36,786,566 positions. |
| Fires at 0.44% 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: expressions related to personal growth and resilience""" |
| |
| fire = { |
| |
| ' you', |
| |
| ' then', ' Then', 'Then', 'then', ' THEN', |
| |
| '?', |
| |
| ' get', ' gets', |
| |
| ' your', |
| |
| ' find', ' finding', ' Finding', 'finding', |
| |
| ' without', |
| |
| ' start', ' starts', |
| |
| ' things', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'you-family', 1 |
|
|
| def prog_L08F21910(input_tokens): |
| """1 token families, 2 surface forms (train F1 of the uncapped optimum: 0.298). |
| Fires at 14,043 of 36,786,566 positions.""" |
| |
| """prev SAE description: the phrase "do what" in various contexts""" |
| |
| fire = { |
| |
| ' what', 'what', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'what-family', 1 |
|
|
| def prog_L08F21911(input_tokens): |
| """9 token families, 14 surface forms (train F1 of the uncapped optimum: 0.136). |
| Fires at 206,983 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: words related to security and safety mechanisms""" |
| |
| fire = { |
| |
| '-', |
| |
| ')', |
| |
| '"', |
| |
| ' way', ' ways', |
| |
| ' information', 'information', |
| |
| ' access', 'access', |
| |
| ' experience', |
| |
| ' level', ' levels', 'level', |
| |
| ' news', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '--family', 1 |
|
|
| def prog_L08F21912(input_tokens): |
| """3 token families, 9 surface forms (train F1 of the uncapped optimum: 0.774). |
| Fires at 28,157 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases that emphasize specific actions or outcomes""" |
| |
| fire = { |
| |
| ' these', ' These', 'These', 'these', ' THESE', |
| |
| ' Those', 'Those', |
| |
| ' Such', 'Such', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'these-family', 1 |
|
|
| def prog_L08F21913(input_tokens): |
| """8 token families, 20 surface forms (train F1 of the uncapped optimum: 0.205). |
| Fires at 117,414 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases that specify examples or categories of items or entities""" |
| |
| fire = { |
| |
| ' including', ' includes', ' included', 'including', ' Including', 'includes', |
| |
| ' like', 'like', |
| |
| ' such', 'such', |
| |
| ' called', ' Called', |
| |
| ' include', ' Include', |
| |
| ' especially', 'especially', ' Especially', |
| |
| ' featuring', |
| |
| ' particularly', 'particularly', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'includ-family', 1 |
|
|
| def prog_L08F21914(input_tokens): |
| """47 token families, 55 surface forms (train F1 of the uncapped optimum: 0.628). |
| Fires at 15,512 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the stomach and gastrointestinal health issues""" |
| |
| fire = { |
| |
| ' gut', ' guts', |
| |
| ' stomach', |
| |
| ' throat', ' throats', |
| |
| ' Taco', ' taco', ' tacos', |
| |
| 'elly', |
| |
| ' oceans', |
| |
| ' appetite', |
| |
| ' payroll', |
| |
| ' belly', |
| |
| ' Gast', ' gast', |
| |
| ' respiratory', |
| |
| ' waist', |
| |
| ' hamstring', |
| |
| ' cardiac', |
| |
| 'utton', |
| |
| ' breathing', |
| |
| ' panc', ' Panc', |
| |
| ' vomiting', ' vomit', |
| |
| ' digest', |
| |
| ' vagina', |
| |
| ' conscience', |
| |
| ' seaw', |
| |
| ' metabolism', |
| |
| ' atmospheric', |
| |
| ' bowel', |
| |
| ' abdomen', |
| |
| ' nausea', |
| |
| ' carniv', |
| |
| ' groin', |
| |
| ' Nicaragua', |
| |
| ' backstage', |
| |
| 'agus', |
| |
| ' Varg', |
| |
| ' Investig', |
| |
| ' appet', |
| |
| ' torso', |
| |
| ' abdominal', |
| |
| ' eyeb', |
| |
| ' intestinal', 'intestinal', |
| |
| ' GI', |
| |
| ' digestive', |
| |
| ' Sylvia', |
| |
| ' Bloody', |
| |
| 'aryn', |
| |
| ' pancreat', |
| |
| ' bladder', |
| |
| ' microbiome', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'gut-family', 1 |
|
|
| def prog_L08F21915(input_tokens): |
| """27 token families, 78 surface forms (train F1 of the uncapped optimum: 0.221). |
| Fires at 148,919 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: terms related to various fields of study or expertise""" |
| |
| fire = { |
| |
| ' social', ' Social', ' soc', 'social', |
| |
| ' political', ' Political', 'Political', ' polit', 'political', ' POLIT', |
| |
| ' public', ' Public', |
| |
| ' business', ' Business', 'Business', 'business', |
| |
| ' security', 'security', ' securities', |
| |
| ' health', 'health', |
| |
| ' digital', ' Digital', 'Digital', ' digit', 'digital', |
| |
| ' law', |
| |
| ' foreign', 'foreign', |
| |
| ' human', ' Human', ' humanities', |
| |
| ' sports', ' sport', ' sporting', 'Sports', 'sports', |
| |
| ' economic', ' Economic', 'econom', 'Econom', 'Economic', |
| |
| ' medical', ' Medical', |
| |
| ' computer', ' Computer', 'Computer', ' computational', ' computed', |
| |
| ' financial', ' Financial', 'financial', |
| |
| ' science', ' sci', 'Science', 'science', |
| |
| ' music', ' musical', ' mus', |
| |
| ' tech', ' Tech', 'Tech', |
| |
| ' climate', ' Climate', 'climate', |
| |
| ' web', |
| |
| ' anti', |
| |
| ' art', ' arts', |
| |
| ' consumer', ' Consumer', 'Consumer', |
| |
| ' civil', |
| |
| ' environmental', |
| |
| ' cyber', ' Cyber', |
| |
| ' investment', ' Investment', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'soc-family', 1 |
|
|
| def prog_L08F21916(input_tokens): |
| """28 token families, 32 surface forms (train F1 of the uncapped optimum: 0.418). |
| Fires at 11,278 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the rear aspect of objects, particularly vehicles""" |
| |
| fire = { |
| |
| ' front', 'Front', ' frontal', |
| |
| ' rear', ' Rear', |
| |
| ' exterior', |
| |
| ' reverse', ' Reverse', |
| |
| ' tail', |
| |
| ' hind', |
| |
| ' resur', |
| |
| ' exhaust', |
| |
| ' steering', |
| |
| ' Ret', |
| |
| ' shel', |
| |
| ' onboard', |
| |
| ' overhead', |
| |
| 'arter', |
| |
| 'quarter', |
| |
| ' brake', |
| |
| ' outward', |
| |
| ' detached', |
| |
| ' stereo', |
| |
| ' anterior', |
| |
| 'facing', |
| |
| ' stern', |
| |
| ' cockpit', |
| |
| 'terior', |
| |
| ' Bare', |
| |
| ' protr', |
| |
| ' lateral', |
| |
| ' downstream', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'front-family', 1 |
|
|
| def prog_L08F21917(input_tokens): |
| """35 token families, 60 surface forms (train F1 of the uncapped optimum: 0.451). |
| Fires at 40,878 of 36,786,566 positions.""" |
| |
| """prev SAE description: words and phrases indicating significant increases in value or intensity""" |
| |
| fire = { |
| |
| ' rising', ' rises', ' Rising', 'rising', |
| |
| ' rise', |
| |
| ' surge', ' Surge', |
| |
| ' soaring', ' soared', ' soar', |
| |
| ' booming', ' boom', |
| |
| ' flying', 'flying', |
| |
| ' exploded', ' exploding', ' explodes', |
| |
| ' surged', ' surging', ' surges', |
| |
| ' plunged', ' plung', |
| |
| ' jumped', ' jumps', ' jumper', |
| |
| ' risen', |
| |
| ' rose', |
| |
| ' climbed', ' climbing', ' climb', |
| |
| ' hyper', 'hyper', |
| |
| ' escalating', ' escalated', |
| |
| ' falling', ' Falling', |
| |
| ' dramatically', |
| |
| ' spir', ' spiral', |
| |
| ' staggering', |
| |
| ' upward', ' upwards', |
| |
| ' plummeted', ' plummet', |
| |
| 'umbled', 'umbling', 'umbles', |
| |
| ' spike', |
| |
| ' sharply', |
| |
| ' whopping', |
| |
| ' rampant', |
| |
| ' plunge', |
| |
| ' erupted', |
| |
| ' doubled', |
| |
| ' slim', |
| |
| ' epidemic', |
| |
| ' Janet', |
| |
| ' trajectory', |
| |
| ' elevated', |
| |
| ' bul', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'ris-family', 1 |
|
|
| def prog_L08F21918(input_tokens): |
| """24 token families, 42 surface forms (train F1 of the uncapped optimum: 0.368). |
| Fires at 18,232 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to images or photographs in context""" |
| |
| fire = { |
| |
| ' seen', 'seen', ' Seen', |
| |
| ' photo', ' PHOTO', 'PHOTOS', 'photo', |
| |
| ' shown', 'shown', |
| |
| ' pictured', 'pictured', |
| |
| 'Picture', ' Picture', |
| |
| ' photographed', ' photograph', ' Photograph', |
| |
| 'FILE', ' FILE', |
| |
| ' depicted', ' depicts', ' depicting', |
| |
| 'see', |
| |
| 'left', |
| |
| 'Pict', ' Pict', ' pict', 'pict', |
| |
| ' Courtesy', 'Courtesy', |
| |
| ' filmed', |
| |
| 'TY', |
| |
| ' surrounded', |
| |
| ' displayed', |
| |
| 'above', ' Above', |
| |
| 'VIDEO', |
| |
| ' onstage', |
| |
| ' flanked', |
| |
| ' acquitted', |
| |
| 'played', |
| |
| 'below', |
| |
| ' circa', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'seen-family', 1 |
|
|
| def prog_L08F21919(input_tokens): |
| """45 token families, 49 surface forms (train F1 of the uncapped optimum: 0.606). |
| Fires at 15,662 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to parsing or the term "ars." """ |
| |
| fire = { |
| |
| 'ars', 'ARS', |
| |
| 'AR', |
| |
| 'ared', |
| |
| 'ares', |
| |
| 'aring', |
| |
| 'arp', |
| |
| ' mountains', |
| |
| ' Cars', |
| |
| ' Lars', |
| |
| 'oust', |
| |
| 'las', |
| |
| ' rockets', |
| |
| 'arin', |
| |
| ' Sirius', |
| |
| ' Ras', |
| |
| ' Arlington', |
| |
| 'aris', |
| |
| ' Marcos', |
| |
| ' staging', |
| |
| ' Parsons', |
| |
| ' pars', ' Pars', |
| |
| 'Sir', |
| |
| ' Katy', |
| |
| ' parsing', ' parser', ' parsed', |
| |
| 'arson', |
| |
| ' Zar', |
| |
| 'arro', |
| |
| ' rehearsal', |
| |
| 'arius', |
| |
| 'arer', |
| |
| ' Pasadena', |
| |
| 'irs', |
| |
| ' Kardashian', |
| |
| 'myra', |
| |
| 'rar', |
| |
| 'iffin', |
| |
| ' rehears', |
| |
| 'asar', |
| |
| 'yrus', |
| |
| 'aryn', |
| |
| 'oras', |
| |
| ' Lumpur', |
| |
| ' Aram', |
| |
| 'arse', |
| |
| ' flares', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'ars-family', 1 |
|
|
| def prog_L08F21920(input_tokens): |
| """28 token families, 45 surface forms (train F1 of the uncapped optimum: 0.354). |
| Fires at 83,645 of 36,786,566 positions.""" |
| |
| """prev SAE description: names of people or proper nouns""" |
| |
| fire = { |
| |
| 'ie', |
| |
| ' Ryan', 'Ryan', 'ryan', |
| |
| 'on', |
| |
| 'ley', 'LEY', |
| |
| ' Scott', 'Scott', |
| |
| ' Martin', 'Martin', |
| |
| ' Lee', 'Lee', 'lee', |
| |
| ' Jordan', 'Jordan', |
| |
| 'ton', |
| |
| 'ne', |
| |
| ' Russell', 'Russell', |
| |
| ' Wayne', 'wayne', |
| |
| ' Marcus', 'Marcus', |
| |
| 'ane', |
| |
| ' Colin', |
| |
| ' Dennis', |
| |
| ' Warren', 'Warren', |
| |
| ' Anthony', |
| |
| ' Gordon', 'Gordon', |
| |
| ' Ross', 'Ross', |
| |
| ' Tyler', 'Tyler', |
| |
| ' Grant', 'Grant', |
| |
| ' Howard', 'Howard', |
| |
| ' Bryan', |
| |
| 'rick', |
| |
| ' Lindsey', |
| |
| ' Corey', |
| |
| ' Julian', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'ie-family', 1 |
|
|
| def prog_L08F21921(input_tokens): |
| """1 token families, 2 surface forms (train F1 of the uncapped optimum: 0.203). |
| Fires at 82,975 of 36,786,566 positions.""" |
| |
| """prev SAE description: repeated conjunctions and their use in connecting various clauses or ideas""" |
| |
| fire = { |
| |
| ' and', ' AND', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'and-family', 1 |
|
|
| def prog_L08F21922(input_tokens): |
| """21 token families, 35 surface forms (train F1 of the uncapped optimum: 0.234). |
| Fires at 46,092 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases related to economic inequality and taxation policies""" |
| |
| fire = { |
| |
| ' top', |
| |
| ' rich', ' richest', ' richer', |
| |
| ' poor', ' poorest', ' Poor', ' poorer', |
| |
| ' middle', 'middle', |
| |
| ' inequality', ' inequalities', |
| |
| ' super', |
| |
| ' class', |
| |
| ' income', 'income', |
| |
| ' corporations', |
| |
| ' wealthy', |
| |
| ' wealth', ' Wealth', 'wealth', |
| |
| ' taxes', |
| |
| ' households', |
| |
| ' 500', |
| |
| ' sector', ' sectors', |
| |
| ' bottom', |
| |
| ' wealthiest', ' wealthier', |
| |
| ' earners', ' earning', 'Earn', |
| |
| ' elite', |
| |
| ' billionaires', |
| |
| ' 99', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'top-family', 1 |
|
|
| def prog_L08F21923(input_tokens): |
| """2 token families, 6 surface forms (train F1 of the uncapped optimum: 0.214). |
| Fires at 72,964 of 36,786,566 positions.""" |
| |
| """prev SAE description: statements about the importance and value of various concepts or subjects""" |
| |
| fire = { |
| |
| ' is', ' Is', ' IS', |
| |
| ' are', ' Are', ' ARE', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'is-family', 1 |
|
|
| def prog_L08F21924(input_tokens): |
| """1 token families, 3 surface forms (train F1 of the uncapped optimum: 0.113). |
| Fires at 95,829 of 36,786,566 positions.""" |
| |
| """prev SAE description: the definite article "the" """ |
| |
| fire = { |
| |
| ' the', 'the', ' THE', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'the-family', 1 |
|
|
| def prog_L08F21925(input_tokens): |
| """12 token families, 48 surface forms (train F1 of the uncapped optimum: 0.133). |
| Fires at 246,475 of 36,786,566 positions. |
| Fires at 0.67% 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 various types of biases and social issues""" |
| |
| fire = { |
| |
| ' games', ' Games', ' gamers', ' Gaming', 'games', ' gamer', 'gaming', 'Gaming', |
| |
| ' system', ' systems', ' System', ' Systems', |
| |
| ' version', ' Version', 'version', ' VERS', 'VERSION', |
| |
| ' game', ' GAME', |
| |
| 'ism', 'isms', 'ISM', |
| |
| ' School', ' schools', ' Schools', |
| |
| ' market', ' markets', ' marketing', ' Market', ' Marketing', ' Markets', ' marketers', |
| |
| ' data', ' Data', ' DATA', |
| |
| 'ists', 'ist', |
| |
| ' Act', ' Acts', 'acting', |
| |
| ' coach', ' coaches', ' Coach', |
| |
| ' Minister', ' minister', ' ministers', ' Ministers', 'ministic', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'gam-family', 1 |
|
|
| def prog_L08F21926(input_tokens): |
| """40 token families, 58 surface forms (train F1 of the uncapped optimum: 0.565). |
| Fires at 29,133 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the word "smash" and its variations""" |
| |
| fire = { |
| |
| ' upset', |
| |
| ' smashed', ' Smash', ' smash', ' smashing', |
| |
| 'ashing', 'ashed', 'ashes', |
| |
| ' crushed', ' crushing', ' crush', ' Crush', |
| |
| ' clashes', ' clashed', ' clash', |
| |
| ' flagship', |
| |
| ' overturn', ' overturned', |
| |
| ' thrilled', |
| |
| ' pleased', |
| |
| ' shattered', ' shattering', ' Shattered', |
| |
| ' furious', |
| |
| ' devastated', |
| |
| ' scattered', ' scattering', |
| |
| ' vandal', |
| |
| ' trump', 'trump', |
| |
| ' crashing', |
| |
| ' havoc', |
| |
| ' unse', |
| |
| ' Rebel', ' Rebels', |
| |
| ' delighted', |
| |
| ' bruised', ' bru', |
| |
| ' sensational', |
| |
| ' shred', |
| |
| ' vandalism', |
| |
| ' violently', |
| |
| ' bruising', ' bruises', |
| |
| ' tram', |
| |
| ' hur', |
| |
| ' craz', |
| |
| ' Sunset', |
| |
| ' fiercely', |
| |
| ' fractured', |
| |
| ' shaken', |
| |
| ' smear', |
| |
| ' topp', |
| |
| ' stomp', |
| |
| ' blur', |
| |
| ' squash', |
| |
| ' bud', |
| |
| ' tabloid', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'upset-family', 1 |
|
|
| def prog_L08F21927(input_tokens): |
| """15 token families, 17 surface forms (train F1 of the uncapped optimum: 0.093). |
| Fires at 109,234 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to notable events or developments related to technology and culture""" |
| |
| fire = { |
| |
| 'The', |
| |
| 'A', |
| |
| 'In', |
| |
| 'This', 'THIS', |
| |
| 'As', |
| |
| 'When', |
| |
| 'After', |
| |
| 'On', |
| |
| 'By', |
| |
| 'If', |
| |
| 'At', |
| |
| 'One', |
| |
| 'An', |
| |
| 'New', 'NEW', |
| |
| 'With', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'the-family', 1 |
|
|
| def prog_L08F21928(input_tokens): |
| """3 token families, 8 surface forms (train F1 of the uncapped optimum: 0.640). |
| Fires at 6,566 of 36,786,566 positions.""" |
| |
| """prev SAE description: mentions of the name "Car" and its variations throughout the document""" |
| |
| fire = { |
| |
| ' Car', 'Car', 'car', 'CAR', ' CAR', ' Cars', |
| |
| ' Tor', |
| |
| ' Kel', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'car-family', 1 |
|
|
| def prog_L08F21929(input_tokens): |
| """10 token families, 20 surface forms (train F1 of the uncapped optimum: 0.492). |
| Fires at 12,012 of 36,786,566 positions.""" |
| |
| """prev SAE description: the repetition of the word "Then." """ |
| |
| fire = { |
| |
| ' then', ' Then', 'Then', 'then', ' THEN', |
| |
| ' Later', 'Later', |
| |
| ' Plus', 'Plus', |
| |
| 'Next', |
| |
| ' Finally', |
| |
| ' Otherwise', 'Otherwise', |
| |
| 'Untitled', |
| |
| ' Eventually', 'Eventually', |
| |
| ' Else', 'Else', 'else', |
| |
| 'Thus', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'then-family', 1 |
|
|
| def prog_L08F21930(input_tokens): |
| """29 token families, 34 surface forms (train F1 of the uncapped optimum: 0.142). |
| Fires at 64,260 of 36,786,566 positions.""" |
| |
| """prev SAE description: words related to political discussions or topics""" |
| |
| fire = { |
| |
| 'o', |
| |
| 'at', |
| |
| 'k', |
| |
| 'en', |
| |
| 'on', |
| |
| 'ar', 'AR', |
| |
| 'z', |
| |
| 'um', 'UM', |
| |
| 'ter', |
| |
| 'it', |
| |
| 'v', |
| |
| 'ch', |
| |
| 'ac', |
| |
| 'ot', |
| |
| 'ol', |
| |
| 'ay', |
| |
| ' Disney', |
| |
| 'te', |
| |
| 'OW', |
| |
| 'IS', |
| |
| 'ov', |
| |
| 'att', 'atter', |
| |
| 'ize', |
| |
| 'ap', |
| |
| 'ost', 'oster', |
| |
| 'la', 'LA', |
| |
| 'ber', |
| |
| ' Research', |
| |
| 'qu', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'o-family', 1 |
|
|
| def prog_L08F21931(input_tokens): |
| """43 token families, 51 surface forms (train F1 of the uncapped optimum: 0.566). |
| Fires at 12,089 of 36,786,566 positions.""" |
| |
| """prev SAE description: concepts related to transparency and accountability""" |
| |
| fire = { |
| |
| ' transparency', ' Transparency', |
| |
| ' whistleblower', ' whistleblowers', |
| |
| ' transparent', |
| |
| ' disclosure', ' Disclosure', |
| |
| ' insider', ' insiders', |
| |
| ' accountability', ' Accountability', |
| |
| ' privat', ' privatization', |
| |
| ' secretive', |
| |
| ' watchdog', |
| |
| ' secrecy', |
| |
| ' procedural', |
| |
| ' sustainability', |
| |
| ' hypocrisy', |
| |
| ' visibility', |
| |
| ' honesty', |
| |
| ' fairness', |
| |
| ' straightforward', |
| |
| ' aud', |
| |
| 'corruption', ' Corruption', |
| |
| ' openness', |
| |
| ' disclosures', |
| |
| ' encaps', |
| |
| ' FOIA', |
| |
| 'esty', |
| |
| ' outsider', |
| |
| ' murky', |
| |
| 'parent', |
| |
| ' opaque', |
| |
| ' shameful', |
| |
| ' simplicity', |
| |
| ' sunshine', |
| |
| ' democrat', |
| |
| ' HTTPS', |
| |
| ' shrouded', |
| |
| 'rina', |
| |
| ' obfusc', |
| |
| ' confidentiality', |
| |
| ' Honest', 'Honest', |
| |
| ' translucent', |
| |
| 'recy', |
| |
| ' modesty', |
| |
| 'spection', |
| |
| ' humility', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'transparency-family', 1 |
|
|
| def prog_L08F21932(input_tokens): |
| """28 token families, 36 surface forms (train F1 of the uncapped optimum: 0.730). |
| Fires at 17,136 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the abbreviation "AL" """ |
| |
| fire = { |
| |
| ' Al', ' al', 'AL', 'Al', ' AL', |
| |
| ' alley', ' Alley', |
| |
| 'Ill', |
| |
| ' Albany', |
| |
| ' Alonso', |
| |
| ' Alison', |
| |
| ' Salem', |
| |
| ' algae', |
| |
| ' Alph', 'alph', |
| |
| ' Jal', |
| |
| ' almond', ' almonds', |
| |
| ' Alto', |
| |
| ' Alb', |
| |
| ' AJ', |
| |
| ' ale', ' ALE', |
| |
| ' Alberto', |
| |
| ' fal', |
| |
| ' CAL', |
| |
| ' Alf', |
| |
| ' Alc', |
| |
| 'RAL', |
| |
| ' abdominal', |
| |
| 'ر', |
| |
| 'Alice', |
| |
| ' anal', |
| |
| 'ulsion', |
| |
| ' Alpine', |
| |
| ' exasper', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'al-family', 1 |
|
|
| def prog_L08F21933(input_tokens): |
| """21 token families, 46 surface forms (train F1 of the uncapped optimum: 0.172). |
| Fires at 157,327 of 36,786,566 positions. |
| Fires at 0.43% 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 industries and sectors related to job opportunities""" |
| |
| fire = { |
| |
| ' health', 'health', |
| |
| ' social', ' soc', |
| |
| ' public', |
| |
| ' science', ' sci', 'sci', |
| |
| ' email', |
| |
| ' financial', 'financial', 'Financial', |
| |
| ' music', 'music', |
| |
| ' education', 'Education', |
| |
| ' medical', ' MED', 'medical', |
| |
| ' business', |
| |
| ' development', 'development', 'Development', 'develop', |
| |
| ' sports', ' sporting', 'sports', |
| |
| ' food', 'food', |
| |
| ' care', 'care', |
| |
| ' TV', ' tv', |
| |
| ' energy', 'energy', |
| |
| ' web', 'web', |
| |
| ' services', 'services', |
| |
| ' art', ' arts', ' arter', |
| |
| ' oil', ' oils', |
| |
| ' personal', 'personal', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'health-family', 1 |
|
|
| def prog_L08F21934(input_tokens): |
| """8 token families, 14 surface forms (train F1 of the uncapped optimum: 0.094). |
| Fires at 234,787 of 36,786,566 positions. |
| Fires at 0.64% 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: themes related to enjoyment and appreciation of nature or gardens""" |
| |
| fire = { |
| |
| ' "', |
| |
| ' not', 'Notes', |
| |
| ' which', |
| |
| ' there', 'there', ' THERE', |
| |
| ';', ' ;', |
| |
| ' no', |
| |
| ' day', 'DAY', ' DAY', |
| |
| ' made', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '"-family', 1 |
|
|
| def prog_L08F21935(input_tokens): |
| """13 token families, 15 surface forms (train F1 of the uncapped optimum: 0.120). |
| Fires at 16,401 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the New York Times""" |
| |
| fire = { |
| |
| ' Twitter', |
| |
| ' cool', |
| |
| ' LGBT', |
| |
| ' Daily', |
| |
| ' organizations', |
| |
| ' Caption', |
| |
| ' Reddit', 'Reddit', |
| |
| ' FC', |
| |
| ' Transcript', |
| |
| ' GIF', |
| |
| 'news', ' NEW', |
| |
| ' Full', |
| |
| ' PR', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'twitt-family', 1 |
|
|
| def prog_L08F21936(input_tokens): |
| """15 token families, 71 surface forms (train F1 of the uncapped optimum: 0.241). |
| Fires at 452,297 of 36,786,566 positions. |
| Fires at 1.23% 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 layoffs and job cuts in the workforce""" |
| |
| fire = { |
| |
| ' work', ' workers', ' working', ' worked', ' worker', ' works', ' Workers', 'workers', ' Work', 'work', 'working', 'Work', 'worked', ' Worker', 'worker', ' Working', 'Working', ' WORK', ' workings', |
| |
| ' for', |
| |
| ' job', ' jobs', ' Job', ' Jobs', 'job', 'Job', 'jobs', |
| |
| ' been', |
| |
| ' new', |
| |
| ' day', ' days', 'day', 'days', |
| |
| ' former', |
| |
| ' employees', ' Employees', |
| |
| ' hours', ' hour', 'hour', ' hourly', 'Hours', 'hours', ' Hours', ' Hour', 'Hour', |
| |
| '000', |
| |
| ' staff', ' staffers', ' staffer', ' staffing', 'staff', ' staffed', 'Staff', |
| |
| ' women', ' Women', 'Women', 'women', |
| |
| ' contract', ' contracts', 'contract', ' contracted', ' Contract', ' contracting', 'Contract', |
| |
| ' fired', ' firing', ' Fired', ' fir', |
| |
| ' months', 'month', 'months', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'work-family', 1 |
|
|
| def prog_L08F21937(input_tokens): |
| """17 token families, 43 surface forms (train F1 of the uncapped optimum: 0.146). |
| Fires at 245,295 of 36,786,566 positions. |
| Fires at 0.67% 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: terms related to immigration and asylum issues""" |
| |
| fire = { |
| |
| ' are', |
| |
| ' people', |
| |
| ' who', |
| |
| ' were', |
| |
| ' them', ' Them', |
| |
| ' refugees', ' refuge', ' Refugees', |
| |
| ' immigrants', ' immigrant', 'immigrant', |
| |
| ' migrants', ' migrant', |
| |
| ' return', ' returned', ' returning', |
| |
| ' arrived', ' arrival', ' arriving', |
| |
| ' illegal', ' illeg', ' Illegal', 'illegal', |
| |
| ' citizens', ' citizen', 'citizens', |
| |
| ' children', 'Children', |
| |
| ' living', 'living', |
| |
| ' workers', ' worker', ' Work', ' Worker', 'worker', |
| |
| ' asylum', ' Asylum', |
| |
| ' seekers', ' seeking', ' seeker', 'seekers', 'seeking', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'are-family', 1 |
|
|
| def prog_L08F21938(input_tokens): |
| """8 token families, 9 surface forms (train F1 of the uncapped optimum: 0.067). |
| Fires at 318,775 of 36,786,566 positions. |
| Fires at 0.87% 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: indicators of urgency or critical conditions""" |
| |
| fire = { |
| |
| ' a', 'ça', |
| |
| ' with', |
| |
| ' as', |
| |
| ' more', |
| |
| ' up', |
| |
| ' than', |
| |
| ' like', |
| |
| ' no', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'a-family', 1 |
|
|
| def prog_L08F21939(input_tokens): |
| """13 token families, 16 surface forms (train F1 of the uncapped optimum: 0.120). |
| Fires at 75,787 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases indicating sufficiency or adequacy in various contexts""" |
| |
| fire = { |
| |
| ' make', ' MAKE', |
| |
| ' get', |
| |
| ' keep', |
| |
| ' win', |
| |
| ' give', |
| |
| ' stop', |
| |
| ' hold', |
| |
| ' allow', |
| |
| ' ensure', |
| |
| ' prevent', ' prevented', |
| |
| ' cause', |
| |
| ' earn', ' earning', |
| |
| ' buy', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'make-family', 1 |
|
|
| def prog_L08F21940(input_tokens): |
| """2 token families, 3 surface forms (train F1 of the uncapped optimum: 0.269). |
| Fires at 17,806 of 36,786,566 positions.""" |
| |
| """prev SAE description: instances of the phrase "pick up." """ |
| |
| fire = { |
| |
| ' up', |
| |
| ' pickup', ' pickups', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'up-family', 1 |
|
|
| def prog_L08F21941(input_tokens): |
| """5 token families, 21 surface forms (train F1 of the uncapped optimum: 0.741). |
| Fires at 21,110 of 36,786,566 positions.""" |
| |
| """prev SAE description: instances of the word "hold" and its variations in various contexts""" |
| |
| fire = { |
| |
| ' hold', ' holding', ' holds', 'hold', ' holders', ' holder', ' Holder', ' Holdings', ' Hold', 'Hold', ' holdings', 'holders', ' Holding', 'holding', |
| |
| ' held', 'held', ' Held', |
| |
| ' Holden', |
| |
| ' owning', |
| |
| ' cling', ' clinging', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'hold-family', 1 |
|
|
| def prog_L08F21942(input_tokens): |
| """40 token families, 55 surface forms (train F1 of the uncapped optimum: 0.421). |
| Fires at 93,968 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to government and its activities""" |
| |
| fire = { |
| |
| ' Court', 'Court', ' Courts', ' COURT', |
| |
| ' Government', 'Government', ' Governments', |
| |
| ' State', |
| |
| ' Council', 'Council', |
| |
| ' Federal', |
| |
| ' Commission', 'Commission', |
| |
| ' Senate', |
| |
| ' Party', |
| |
| ' Army', 'Army', |
| |
| ' Parliament', |
| |
| ' Police', |
| |
| ' Board', ' Boards', |
| |
| ' Administration', |
| |
| ' Ministry', |
| |
| ' Church', ' Churches', |
| |
| ' Committee', |
| |
| ' Coast', |
| |
| ' Office', |
| |
| ' Constitution', |
| |
| ' Congressional', |
| |
| ' Republic', 'Republic', |
| |
| ' Team', ' Teams', |
| |
| ' Season', |
| |
| ' Navy', |
| |
| ' Draft', |
| |
| ' Presidential', ' Presidents', |
| |
| ' Authority', ' Authorization', |
| |
| ' Legislature', |
| |
| ' Cabinet', |
| |
| ' Left', 'Left', |
| |
| ' Coalition', |
| |
| ' Report', |
| |
| ' Bureau', |
| |
| ' Assembly', |
| |
| ' Side', |
| |
| ' Right', |
| |
| ' Empire', |
| |
| ' Province', |
| |
| ' Forces', |
| |
| ' Embassy', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'court-family', 1 |
|
|
| def prog_L08F21943(input_tokens): |
| """4 token families, 8 surface forms (train F1 of the uncapped optimum: 0.131). |
| Fires at 365,805 of 36,786,566 positions. |
| Fires at 0.99% 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 events or entities in sports and legal contexts""" |
| |
| fire = { |
| |
| '-', |
| |
| ' �', |
| |
| ' White', ' white', 'white', 'White', |
| |
| ' U', ' u', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '--family', 1 |
|
|
| def prog_L08F21944(input_tokens): |
| """39 token families, 51 surface forms (train F1 of the uncapped optimum: 0.662). |
| Fires at 10,670 of 36,786,566 positions.""" |
| |
| """prev SAE description: instances of the word "boat" in various contexts""" |
| |
| fire = { |
| |
| ' boat', ' boats', 'boat', ' Boat', 'boats', |
| |
| ' ship', ' ships', |
| |
| ' vessel', ' vessels', ' Vessel', |
| |
| ' sail', ' sailing', ' sailed', ' sails', |
| |
| ' ferry', ' Ferry', |
| |
| ' helicopter', |
| |
| ' submarine', |
| |
| ' dock', ' docks', |
| |
| ' raft', |
| |
| ' balloon', |
| |
| ' kay', |
| |
| 'acht', |
| |
| ' submarines', |
| |
| ' porch', |
| |
| 'oat', |
| |
| ' fishermen', |
| |
| ' Torres', |
| |
| ' yacht', |
| |
| ' buoy', |
| |
| ' padd', |
| |
| ' balcony', |
| |
| 'orry', |
| |
| 'latable', |
| |
| ' dolphin', |
| |
| 'arge', |
| |
| ' tanker', |
| |
| ' canoe', |
| |
| ' torpedo', |
| |
| ' RV', |
| |
| ' Titanic', |
| |
| ' dams', |
| |
| ' paddle', |
| |
| ' fisherman', |
| |
| ' afloat', |
| |
| 'marine', |
| |
| 'tub', |
| |
| ' ashore', |
| |
| ' lighthouse', |
| |
| ' overboard', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'boat-family', 1 |
|
|
| def prog_L08F21945(input_tokens): |
| """3 token families, 4 surface forms (train F1 of the uncapped optimum: 0.101). |
| Fires at 57,216 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to specifications and requirements in various contexts""" |
| |
| fire = { |
| |
| ':', ' :', |
| |
| ' ]', |
| |
| ' Edit', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return ':-family', 1 |
|
|
| def prog_L08F21946(input_tokens): |
| """3 token families, 3 surface forms (train F1 of the uncapped optimum: 0.214). |
| Fires at 41,555 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases indicating academic positions and affiliations""" |
| |
| fire = { |
| |
| ' at', |
| |
| ' attended', |
| |
| ' prestigious', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'at-family', 1 |
|
|
| def prog_L08F21947(input_tokens): |
| """50 token families, 57 surface forms (train F1 of the uncapped optimum: 0.709). |
| Fires at 28,062 of 36,786,566 positions.""" |
| |
| """prev SAE description: geographical locations and names of places, specifically in Canada""" |
| |
| fire = { |
| |
| ' Edmonton', |
| |
| ' SK', 'sk', 'SK', |
| |
| ' Winnipeg', |
| |
| ' Idaho', |
| |
| ' Finland', |
| |
| ' Rapids', |
| |
| 'ula', |
| |
| ' Kashmir', |
| |
| 'ault', 'aults', |
| |
| 'BU', ' BU', |
| |
| ' Saskatchewan', |
| |
| ' Jacksonville', |
| |
| ' Wellington', |
| |
| ' Kabul', |
| |
| ' Manitoba', |
| |
| ' Wings', |
| |
| ' Finnish', |
| |
| ' Falls', |
| |
| ' bargaining', |
| |
| 'iga', |
| |
| ' Gat', |
| |
| ' Churchill', |
| |
| ' Aboriginal', ' aboriginal', |
| |
| 'iba', |
| |
| ' Arbor', |
| |
| ' Legion', |
| |
| ' Regina', |
| |
| ' Belarus', |
| |
| ' Punjab', |
| |
| 'atoon', |
| |
| ' Sask', |
| |
| ' Ghana', |
| |
| 'Fin', ' FIN', |
| |
| ' Donetsk', |
| |
| ' Waterloo', |
| |
| ' Curiosity', |
| |
| ' Kingston', |
| |
| ' Stafford', |
| |
| ' Leaf', |
| |
| ' Arg', 'Arg', |
| |
| ' Sherlock', |
| |
| ' Joan', |
| |
| ' Kun', |
| |
| ' wildfires', |
| |
| 'zhou', |
| |
| ' Caldwell', |
| |
| ' wildfire', |
| |
| 'ammu', |
| |
| ' CFL', |
| |
| ' Mandela', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'edmonton-family', 1 |
|
|
| def prog_L08F21948(input_tokens): |
| """46 token families, 55 surface forms (train F1 of the uncapped optimum: 0.473). |
| Fires at 79,565 of 36,786,566 positions.""" |
| |
| """prev SAE description: terms related to sports teams and their draft selections""" |
| |
| fire = { |
| |
| 'year', 'years', |
| |
| 'day', |
| |
| 'time', |
| |
| 'term', 'Term', |
| |
| 'level', 'levels', |
| |
| 'season', |
| |
| 'class', 'Class', |
| |
| 'round', 'rounder', |
| |
| 'game', |
| |
| 'end', |
| |
| 'inch', |
| |
| 'party', |
| |
| 'weight', |
| |
| 'degree', |
| |
| 'sex', |
| |
| 'minute', |
| |
| 'page', |
| |
| 'century', |
| |
| 'generation', |
| |
| 'sized', |
| |
| 'point', |
| |
| 'star', |
| |
| 'state', |
| |
| 'ever', |
| |
| 'style', |
| |
| 'night', |
| |
| 'selling', 'seller', |
| |
| 'era', |
| |
| 'school', |
| |
| 'driving', |
| |
| 'life', |
| |
| 'range', 'Range', |
| |
| 'bound', |
| |
| 'team', |
| |
| 'largest', |
| |
| 'ranked', 'ranking', |
| |
| 'quality', |
| |
| 'world', |
| |
| 'market', |
| |
| 'door', |
| |
| 'grade', |
| |
| 'screen', |
| |
| 'size', |
| |
| 'language', |
| |
| 'government', |
| |
| 'length', 'Length', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'year-family', 1 |
|
|
| def prog_L08F21949(input_tokens): |
| """5 token families, 5 surface forms (train F1 of the uncapped optimum: 0.067). |
| Fires at 46,600 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases indicating actions or states, particularly using the verb "is" in various contexts""" |
| |
| fire = { |
| |
| ':', |
| |
| '?', |
| |
| ' |', |
| |
| ' include', |
| |
| ' simply', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return ':-family', 1 |
|
|
| def prog_L08F21950(input_tokens): |
| """11 token families, 16 surface forms (train F1 of the uncapped optimum: 0.165). |
| Fires at 168,956 of 36,786,566 positions. |
| Fires at 0.46% 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: predictions about future events""" |
| |
| fire = { |
| |
| ' will', 'will', |
| |
| ' be', |
| |
| ' soon', |
| |
| 'll', "'ll", |
| |
| ' become', ' Become', |
| |
| ' eventually', ' Eventually', |
| |
| ' start', |
| |
| ' finally', |
| |
| ' able', |
| |
| ' future', |
| |
| ' becoming', ' becomes', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'will-family', 1 |
|
|
| def prog_L08F21951(input_tokens): |
| """13 token families, 21 surface forms (train F1 of the uncapped optimum: 0.675). |
| Fires at 11,256 of 36,786,566 positions.""" |
| |
| """prev SAE description: the word "particularly" and related expressions emphasizing significance or importance""" |
| |
| fire = { |
| |
| ' particularly', ' particular', 'particularly', ' Particularly', |
| |
| ' especially', ' Especially', 'especially', 'Especially', |
| |
| ' specifically', ' Specifically', 'Specifically', |
| |
| ' unusually', |
| |
| ' specially', |
| |
| ' downright', |
| |
| ' uniquely', |
| |
| ' exceptionally', |
| |
| ' notoriously', |
| |
| ' disproportionately', |
| |
| ' acutely', |
| |
| 'uly', |
| |
| ' chiefly', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'particular-family', 1 |
|
|
| def prog_L08F21952(input_tokens): |
| """43 token families, 52 surface forms (train F1 of the uncapped optimum: 0.542). |
| Fires at 15,873 of 36,786,566 positions.""" |
| |
| """prev SAE description: terms related to emergency situations or services""" |
| |
| fire = { |
| |
| ' emergency', ' Emergency', 'Emergency', |
| |
| ' disaster', ' disasters', ' Disaster', |
| |
| ' urgent', ' urgently', |
| |
| ' investigative', ' Investigative', |
| |
| ' rescue', |
| |
| ' interim', |
| |
| ' evacuation', |
| |
| ' acute', |
| |
| ' Joint', |
| |
| ' evacuated', |
| |
| ' panic', |
| |
| ' hospitalized', |
| |
| ' exped', |
| |
| 'emer', |
| |
| ' urgency', |
| |
| ' canned', |
| |
| ' evac', |
| |
| ' Amber', |
| |
| 'saving', |
| |
| ' respiratory', |
| |
| ' ER', |
| |
| ' improvised', ' improv', |
| |
| ' emergencies', |
| |
| ' ult', |
| |
| ' intensive', |
| |
| 'threatening', |
| |
| 'traumatic', |
| |
| 'rial', |
| |
| ' specialty', |
| |
| 'omsday', |
| |
| 'mber', |
| |
| 'gency', |
| |
| 'ampa', |
| |
| 'romptu', |
| |
| 'ordinary', |
| |
| 'extra', 'Extra', |
| |
| ' Expedition', |
| |
| ' Jaime', |
| |
| ' EMS', 'EMS', |
| |
| ' contingency', |
| |
| 'ourmet', |
| |
| ' evacuate', |
| |
| ' discretionary', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'emergency-family', 1 |
|
|
| def prog_L08F21953(input_tokens): |
| """34 token families, 47 surface forms (train F1 of the uncapped optimum: 0.695). |
| Fires at 18,114 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to shooting incidents or violent events""" |
| |
| fire = { |
| |
| ' shooting', ' shooter', ' shootings', ' shoot', ' shooters', ' Shooting', ' Shoot', 'shoot', ' Shooter', |
| |
| ' massacre', ' Massacre', |
| |
| ' bombing', ' bombings', |
| |
| ' gunman', |
| |
| ' killings', |
| |
| ' gunshot', ' gunshots', |
| |
| ' firing', |
| |
| ' gunfire', |
| |
| ' gunmen', |
| |
| ' murders', |
| |
| ' assassination', |
| |
| ' stabbing', |
| |
| ' rampage', |
| |
| ' homicide', |
| |
| ' Hook', |
| |
| ' poisoning', |
| |
| ' thriller', |
| |
| ' sniper', ' snipers', ' Sniper', |
| |
| ' gunned', |
| |
| ' Tarant', |
| |
| ' Newtown', |
| |
| ' shootout', |
| |
| ' homicides', |
| |
| ' Rifle', |
| |
| ' ambush', |
| |
| ' robberies', |
| |
| ' spree', |
| |
| 'ooting', |
| |
| ' Paddock', |
| |
| ' Pist', |
| |
| ' Hebdo', |
| |
| ' massacres', |
| |
| ' Columb', |
| |
| ' Pulse', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'shoot-family', 1 |
|
|
| def prog_L08F21954(input_tokens): |
| """14 token families, 16 surface forms (train F1 of the uncapped optimum: 0.589). |
| Fires at 17,048 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases that include the word "based." """ |
| |
| fire = { |
| |
| ' based', ' Based', 'Based', |
| |
| ' founded', |
| |
| ' focused', |
| |
| ' centered', |
| |
| ' derived', |
| |
| ' rooted', |
| |
| ' relies', |
| |
| ' stems', |
| |
| ' dependent', |
| |
| ' revolves', |
| |
| ' modeled', |
| |
| ' grounded', |
| |
| ' headquartered', |
| |
| ' rests', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'bas-family', 1 |
|
|
| def prog_L08F21955(input_tokens): |
| """16 token families, 25 surface forms (train F1 of the uncapped optimum: 0.119). |
| Fires at 101,614 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases that describe different types or categories of things""" |
| |
| fire = { |
| |
| ' a', |
| |
| ' an', |
| |
| ' "', |
| |
| ' like', |
| |
| " '", |
| |
| 'the', |
| |
| 'called', ' call', |
| |
| ' essentially', ' Essentially', |
| |
| ' effectively', |
| |
| ' own', |
| |
| ' basically', |
| |
| ' true', 'true', |
| |
| ' act', ' acts', ' acting', ' acted', |
| |
| ' ultimate', 'ultimate', 'Ultimate', |
| |
| ' giant', |
| |
| ' official', 'official', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'a-family', 1 |
|
|
| def prog_L08F21956(input_tokens): |
| """2 token families, 2 surface forms (train F1 of the uncapped optimum: 0.075). |
| Fires at 125,000 of 36,786,566 positions.""" |
| |
| """prev SAE description: concepts related to separation and distinction between various entities or ideas""" |
| |
| fire = { |
| |
| ' another', |
| |
| ' Clinton', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'anoth-family', 1 |
|
|
| def prog_L08F21957(input_tokens): |
| """33 token families, 45 surface forms (train F1 of the uncapped optimum: 0.256). |
| Fires at 38,094 of 36,786,566 positions.""" |
| |
| """prev SAE description: occurrences of the character sequence "ch" """ |
| |
| fire = { |
| |
| 'z', |
| |
| 'ch', |
| |
| 'ach', ' Ach', |
| |
| 'ek', ' Ek', |
| |
| 'ak', 'Ak', |
| |
| 'iz', |
| |
| 'ik', ' Ik', |
| |
| 'ich', ' Ich', |
| |
| 'ish', ' Ish', |
| |
| 'ash', 'ASH', |
| |
| 'az', |
| |
| 'ok', ' Ok', 'Ok', |
| |
| 'av', |
| |
| ' Space', |
| |
| 'q', |
| |
| 'aj', |
| |
| 'oh', |
| |
| 'och', |
| |
| 'eb', ' Eb', |
| |
| 'ff', |
| |
| 'KY', |
| |
| 'ev', |
| |
| 'oj', |
| |
| 'osh', |
| |
| ' Ark', 'Ark', |
| |
| 'ush', |
| |
| 'akh', |
| |
| 'uch', |
| |
| 'esh', |
| |
| 'kh', |
| |
| ' Bed', 'Bed', |
| |
| 'ech', |
| |
| 'iy', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'z-family', 1 |
|
|
| def prog_L08F21958(input_tokens): |
| """8 token families, 9 surface forms (train F1 of the uncapped optimum: 0.128). |
| Fires at 52,421 of 36,786,566 positions.""" |
| |
| """prev SAE description: names of people and their social media handles""" |
| |
| fire = { |
| |
| 'New', 'News', |
| |
| '_', |
| |
| ' (@', |
| |
| 'J', |
| |
| '&', |
| |
| 'IL', |
| |
| 'Z', |
| |
| 'AS', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'new-family', 1 |
|
|
| def prog_L08F21959(input_tokens): |
| """2 token families, 2 surface forms (train F1 of the uncapped optimum: 0.216). |
| Fires at 10,540 of 36,786,566 positions.""" |
| |
| """prev SAE description: instances of the contraction "won't" """ |
| |
| fire = { |
| |
| 't', |
| |
| ' wont', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 't-family', 1 |
|
|
| def prog_L08F21960(input_tokens): |
| """12 token families, 15 surface forms (train F1 of the uncapped optimum: 0.655). |
| Fires at 14,111 of 36,786,566 positions.""" |
| |
| """prev SAE description: occurrences of the word "House" in various contexts, particularly in relation to legislative discussions""" |
| |
| fire = { |
| |
| ' House', 'House', ' HOUSE', |
| |
| ' Representatives', ' Representative', |
| |
| ' congressional', |
| |
| ' Majority', |
| |
| 'HOU', |
| |
| ' Boehner', |
| |
| ' Sabha', |
| |
| 'Represent', |
| |
| ' Lords', |
| |
| ' Reps', |
| |
| ' Houses', |
| |
| ' Pulse', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'house-family', 1 |
|
|
| def prog_L08F21961(input_tokens): |
| """25 token families, 61 surface forms (train F1 of the uncapped optimum: 0.137). |
| Fires at 99,668 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases related to risk and consequences""" |
| |
| fire = { |
| |
| ' to', |
| |
| ' can', |
| |
| ' allowing', ' allows', ' allow', ' allowed', 'allows', |
| |
| ' more', |
| |
| ' means', ' meaning', ' mean', |
| |
| ' makes', ' making', |
| |
| ' prevent', ' preventing', ' prevented', ' prevents', |
| |
| ' result', ' resulting', ' resulted', 'result', |
| |
| ' leading', ' lead', ' leads', |
| |
| ' avoid', ' avoiding', ' avoids', ' avoided', 'avoid', ' Avoid', |
| |
| ' causing', ' caused', ' causes', |
| |
| ' creating', ' created', ' creates', |
| |
| ' why', |
| |
| ' able', |
| |
| ' difficult', ' difficulties', |
| |
| ' leaving', ' leaves', |
| |
| ' possible', ' possibility', |
| |
| ' prompting', ' prompted', ' prompt', ' prompts', |
| |
| ' less', |
| |
| ' access', |
| |
| ' reduce', |
| |
| ' thus', ' Thus', 'thus', |
| |
| ' reducing', ' reduces', ' reduced', ' reduction', |
| |
| ' opportunity', ' opportunities', |
| |
| ' ability', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'to-family', 1 |
|
|
| def prog_L08F21962(input_tokens): |
| """14 token families, 23 surface forms (train F1 of the uncapped optimum: 0.682). |
| Fires at 12,131 of 36,786,566 positions.""" |
| |
| """prev SAE description: mentions of the term "Open" and related concepts""" |
| |
| fire = { |
| |
| ' open', ' Open', 'Open', 'open', ' Opening', ' OPEN', ' openness', 'opens', |
| |
| ' Op', |
| |
| ' Brew', 'Brew', |
| |
| ' Modern', |
| |
| ' Clear', 'Clear', |
| |
| ' Apache', |
| |
| ' Clean', |
| |
| ' Libre', |
| |
| ' Fen', |
| |
| ' DirectX', |
| |
| 'Oak', |
| |
| ' OpenGL', |
| |
| ' Closed', |
| |
| ' PLoS', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'open-family', 1 |
|
|
| def prog_L08F21963(input_tokens): |
| """7 token families, 12 surface forms (train F1 of the uncapped optimum: 0.079). |
| Fires at 90,087 of 36,786,566 positions.""" |
| |
| """prev SAE description: programming constructs and function definitions""" |
| |
| fire = { |
| |
| ':', ' :', |
| |
| 'a', 'ça', 'ía', 'ña', |
| |
| '00', |
| |
| '30', |
| |
| ' .', |
| |
| ' de', |
| |
| ' |', '|', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return ':-family', 1 |
|
|
| def prog_L08F21964(input_tokens): |
| """1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.294). |
| Fires at 102,111 of 36,786,566 positions.""" |
| |
| """prev SAE description: prepositions, particularly "in," used in various contexts""" |
| |
| fire = { |
| |
| ' in', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'in-family', 1 |
|
|
| def prog_L08F21965(input_tokens): |
| """1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.070). |
| Fires at 34,880 of 36,786,566 positions.""" |
| |
| """prev SAE description: instances of the phrase "you can't" or variations thereof""" |
| |
| fire = { |
| |
| '�', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '�-family', 1 |
|
|
| def prog_L08F21966(input_tokens): |
| """17 token families, 49 surface forms (train F1 of the uncapped optimum: 0.098). |
| Fires at 285,689 of 36,786,566 positions. |
| Fires at 0.78% 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 indicating alerts or announcements of significant events""" |
| |
| fire = { |
| |
| ' a', |
| |
| ' an', ' An', |
| |
| ' you', ' You', 'You', 'you', |
| |
| ' your', ' Your', 'Your', ' yours', 'your', |
| |
| ' or', |
| |
| ' had', |
| |
| ' what', ' What', 'What', 'what', 'WHAT', ' WHAT', |
| |
| 'Last', ' Last', |
| |
| ' my', ' My', ' MY', |
| |
| '!', ' !', |
| |
| ' work', ' working', 'worker', ' Work', 'workers', ' WORK', |
| |
| ' something', 'Something', ' Something', 'something', |
| |
| ' going', ' Going', |
| |
| ' me', |
| |
| ' coming', 'coming', ' Coming', |
| |
| 've', "'ve", |
| |
| ' someone', ' Someone', 'Someone', 'someone', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'a-family', 1 |
|
|
| def prog_L08F21967(input_tokens): |
| """32 token families, 54 surface forms (train F1 of the uncapped optimum: 0.331). |
| Fires at 20,027 of 36,786,566 positions.""" |
| |
| """prev SAE description: terms related to exercising rights and powers""" |
| |
| fire = { |
| |
| ' exercise', ' Exercise', |
| |
| ' abusing', ' abused', |
| |
| ' wielding', ' wield', ' wielded', |
| |
| ' assumed', ' assuming', ' assumes', |
| |
| ' regain', ' regained', |
| |
| ' resort', ' resorted', |
| |
| ' granted', |
| |
| ' invoked', ' invoking', |
| |
| ' lift', ' lifting', |
| |
| ' freely', |
| |
| ' exert', ' exerted', |
| |
| ' flex', 'flex', |
| |
| ' enlisted', ' enlist', |
| |
| ' assert', ' asserting', 'assert', |
| |
| ' exercising', |
| |
| ' resumed', ' resumes', |
| |
| 'erc', 'ERC', |
| |
| ' relinqu', |
| |
| ' wrest', |
| |
| ' exhausted', ' exhaust', ' exhausting', |
| |
| ' practicing', |
| |
| ' exercised', ' exerc', |
| |
| ' refrain', |
| |
| ' relieve', |
| |
| ' resume', |
| |
| ' imb', |
| |
| ' invoke', 'invoke', |
| |
| ' tread', |
| |
| ' impaired', |
| |
| ' revoked', |
| |
| ' discretion', |
| |
| ' behaving', ' behaved', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'exercise-family', 1 |
|
|
| def prog_L08F21968(input_tokens): |
| """2 token families, 3 surface forms (train F1 of the uncapped optimum: 0.054). |
| Fires at 24,106 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases expressing disappointment or frustration about unmet expectations""" |
| |
| fire = { |
| |
| 've', |
| |
| '\xad', ' \xad', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 've-family', 1 |
|
|
| def prog_L08F21969(input_tokens): |
| """6 token families, 10 surface forms (train F1 of the uncapped optimum: 0.076). |
| Fires at 29,969 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases expressing hypothetical scenarios and their consequences""" |
| |
| fire = { |
| |
| ' have', ' Have', |
| |
| ' able', |
| |
| ' without', |
| |
| ' having', 'having', |
| |
| ' never', ' NEVER', |
| |
| ' expect', ' expecting', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'have-family', 1 |
|
|
| def prog_L08F21970(input_tokens): |
| """24 token families, 61 surface forms (train F1 of the uncapped optimum: 0.180). |
| Fires at 324,633 of 36,786,566 positions. |
| Fires at 0.88% 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 express desires or goals""" |
| |
| fire = { |
| |
| ' to', ' TO', |
| |
| ' with', ' WITH', |
| |
| ':', ' :', |
| |
| ' about', |
| |
| '?', ' ?', |
| |
| ' want', ' wanted', ' wants', ' wanting', ' WANT', |
| |
| ' aimed', ' aim', ' aims', ' aiming', ' Aim', 'Aim', |
| |
| ' plans', ' plan', |
| |
| ' going', 'going', |
| |
| ' goal', ' goals', ' Goals', |
| |
| ' decided', ' decides', ' deciding', |
| |
| ' series', ' serial', |
| |
| ' set', ' SET', |
| |
| ' take', 'take', |
| |
| ' long', 'long', ' longing', ' LONG', |
| |
| ' efforts', ' effort', |
| |
| ' go', |
| |
| ' mission', ' Mission', 'Mission', |
| |
| ' campaign', ' campaigns', ' campaigning', |
| |
| ' focused', ' focusing', ' focuses', 'focused', |
| |
| ' focus', 'focus', |
| |
| ' attacks', ' attack', 'attacks', |
| |
| ' looking', |
| |
| ' love', 'love', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'to-family', 1 |
|
|
| def prog_L08F21971(input_tokens): |
| """1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.145). |
| Fires at 115,811 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the United States and notable events or attacks related to it""" |
| |
| fire = { |
| |
| ' the', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'the-family', 1 |
|
|
| def prog_L08F21972(input_tokens): |
| """30 token families, 113 surface forms (train F1 of the uncapped optimum: 0.303). |
| Fires at 65,075 of 36,786,566 positions.""" |
| |
| """prev SAE description: terms related to software development and version control""" |
| |
| fire = { |
| |
| ' project', ' projects', ' Project', 'Project', ' Projects', 'project', 'projects', |
| |
| ' release', ' Release', 'Release', 'release', |
| |
| ' version', ' Version', 'version', 'Version', 'VERSION', ' VERS', |
| |
| ' source', ' Source', 'source', 'SOURCE', |
| |
| ' build', ' builds', ' Build', 'Build', 'build', |
| |
| ' changes', ' chang', 'Changes', ' Changes', |
| |
| ' update', |
| |
| ' updates', ' updating', ' Updates', |
| |
| ' license', ' License', 'License', 'license', |
| |
| 'Arch', ' archive', ' Arch', ' Archives', 'arch', ' archives', 'archive', 'archives', |
| |
| ' package', ' Package', 'package', 'Package', |
| |
| ' beta', ' Beta', 'beta', 'Beta', |
| |
| ' bug', ' bugs', ' Bug', 'Bug', 'bugs', |
| |
| ' fixes', 'fixes', 'fix', ' Fixes', |
| |
| ' releases', ' Releases', ' Released', |
| |
| ' library', 'library', |
| |
| ' download', ' downloads', 'download', ' Downloads', |
| |
| ' install', 'install', 'Install', ' Install', ' installs', ' installer', |
| |
| ' deploy', ' deployment', ' deploying', ' Deploy', ' deployments', 'Deploy', |
| |
| ' packages', ' packaging', 'packages', |
| |
| ' patch', ' patches', 'Patch', 'patch', |
| |
| ' versions', |
| |
| ' maintain', ' maintaining', |
| |
| ' GitHub', ' github', 'github', ' Github', |
| |
| ' stable', 'stable', |
| |
| ' libraries', ' Libraries', |
| |
| ' branch', ' branches', ' branching', |
| |
| ' Debian', ' debian', 'debian', |
| |
| ' git', 'git', ' Git', |
| |
| ' documentation', ' Documentation', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'project-family', 1 |
|
|
| def prog_L08F21973(input_tokens): |
| """47 token families, 68 surface forms (train F1 of the uncapped optimum: 0.605). |
| Fires at 14,944 of 36,786,566 positions.""" |
| |
| """prev SAE description: occurrences of the word "anchor" and its variations""" |
| |
| fire = { |
| |
| ' anchor', ' anchored', ' anchors', |
| |
| ' flagship', |
| |
| 'anch', ' Anch', ' anch', |
| |
| ' premise', |
| |
| 'pin', |
| |
| ' spear', ' Spear', ' spears', |
| |
| ' accent', ' accents', |
| |
| ' Atl', 'Atl', 'atl', |
| |
| ' shore', |
| |
| ' grounded', ' grounding', |
| |
| ' stabil', ' stabilization', ' stabilized', |
| |
| ' beacon', ' Beacon', |
| |
| ' Kimmel', |
| |
| ' pillar', ' pillars', ' Pillar', |
| |
| ' bast', |
| |
| ' evangel', |
| |
| ' staple', |
| |
| ' parasites', ' parasitic', |
| |
| ' satire', |
| |
| ' orth', ' Orth', |
| |
| ' centerpiece', |
| |
| ' buoy', |
| |
| ' Mant', ' mant', |
| |
| ' puppet', ' Puppet', |
| |
| 'aden', ' Aden', |
| |
| ' Sheen', |
| |
| ' outpost', |
| |
| ' containment', 'container', |
| |
| ' cornerstone', |
| |
| 'abin', |
| |
| ' upl', |
| |
| ' Kath', |
| |
| ' collateral', |
| |
| ' mast', |
| |
| ' medi', |
| |
| ' parasite', |
| |
| ' glitter', |
| |
| ' hooks', |
| |
| ' undersc', |
| |
| ' affirm', |
| |
| ' foothold', |
| |
| ' catalyst', |
| |
| ' Canary', |
| |
| ' indoctr', |
| |
| ' Lana', |
| |
| ' sabot', |
| |
| ' subst', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'anchor-family', 1 |
|
|
| def prog_L08F21974(input_tokens): |
| """15 token families, 65 surface forms (train F1 of the uncapped optimum: 0.142). |
| Fires at 382,037 of 36,786,566 positions. |
| Fires at 1.04% 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: issues related to system performance and connection errors""" |
| |
| fire = { |
| |
| '.', |
| |
| ' for', |
| |
| ' your', ' Your', 'Your', |
| |
| ' after', ' After', 'after', |
| |
| ' users', ' user', ' Users', 'Users', ' User', 'user', 'User', 'users', |
| |
| ' some', ' Some', 'some', ' SOME', |
| |
| 'block', ' block', ' blocked', ' blocking', ' blocks', 'blocking', ' blocker', ' blockers', 'blocks', ' Blocks', |
| |
| ' because', ' Because', 'Because', 'because', |
| |
| ' system', ' systems', ' System', 'system', 'System', ' SYSTEM', |
| |
| ' may', ' MAY', |
| |
| ' work', ' working', ' works', ' Work', ' Works', 'Works', ' WORK', |
| |
| ' customers', ' customer', ' custom', ' Customers', ' Custom', ' customized', |
| |
| ' Un', ' un', 'Un', |
| |
| ' update', ' Update', 'update', |
| |
| ' service', ' Service', 'service', 'Service', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '.-family', 1 |
|
|
| def prog_L08F21975(input_tokens): |
| """1 token families, 1 surface forms (train F1 of the uncapped optimum: 0.124). |
| Fires at 97,533 of 36,786,566 positions.""" |
| |
| """prev SAE description: instances of numerical values or ratings""" |
| |
| fire = { |
| |
| '\n', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '\\n-family', 1 |
|
|
| def prog_L08F21976(input_tokens): |
| """11 token families, 15 surface forms (train F1 of the uncapped optimum: 0.075). |
| Fires at 81,325 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to software release information across various platforms""" |
| |
| fire = { |
| |
| ' and', |
| |
| ' or', |
| |
| ' devices', |
| |
| ' &', |
| |
| ' tablets', ' tablet', ' Tablet', |
| |
| ' Xbox', 'Xbox', |
| |
| ' PS', |
| |
| ' version', |
| |
| ' users', |
| |
| ' Android', 'Android', |
| |
| 'based', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'and-family', 1 |
|
|
| def prog_L08F21977(input_tokens): |
| """32 token families, 94 surface forms (train F1 of the uncapped optimum: 0.135). |
| Fires at 337,035 of 36,786,566 positions. |
| Fires at 0.92% 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: expressions of inability or obstacles in achieving goals""" |
| |
| fire = { |
| |
| ' to', |
| |
| 't', "'t", |
| |
| ' I', |
| |
| ' not', 'Notable', |
| |
| ' just', ' Just', 'Just', 'just', |
| |
| ' want', ' wanted', ' wants', ' wanting', |
| |
| ' would', ' Would', 'would', 'Would', |
| |
| ' asked', ' ask', ' asking', ' asks', ' Ask', |
| |
| ' get', ' gets', |
| |
| "'re", 're', |
| |
| ' What', 'What', |
| |
| ' do', ' Do', 'Do', |
| |
| ' going', |
| |
| ' don', |
| |
| ' start', ' started', ' starting', 'Starting', ' Starting', ' Start', 'Start', 'started', ' Started', |
| |
| ' need', ' needed', ' NEED', |
| |
| ' looking', ' look', ' looked', ' Looking', ' Look', |
| |
| ' trying', ' try', ' Try', ' Trying', 'Try', 'try', |
| |
| ' got', 'got', |
| |
| ' did', ' Did', 'did', |
| |
| ' took', |
| |
| ' might', ' Might', 'might', |
| |
| ' take', |
| |
| ' didn', ' Didn', 'didn', |
| |
| ' Here', 'Here', 'here', |
| |
| ' You', 'You', 'you', |
| |
| ' set', 'SET', |
| |
| ' taking', ' takes', ' Taking', 'taking', 'Taking', |
| |
| ' should', |
| |
| ' find', ' finding', ' finds', ' Finding', |
| |
| ' put', ' puts', |
| |
| ' why', ' Why', 'Why', 'why', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'to-family', 1 |
|
|
| def prog_L08F21978(input_tokens): |
| """42 token families, 56 surface forms (train F1 of the uncapped optimum: 0.316). |
| Fires at 127,094 of 36,786,566 positions.""" |
| |
| """prev SAE description: verbs related to actions and achievements""" |
| |
| fire = { |
| |
| ' made', ' Made', |
| |
| ' took', |
| |
| ' got', ' Got', 'got', |
| |
| ' did', ' DID', |
| |
| ' came', ' Came', |
| |
| ' went', ' Went', 'went', |
| |
| ' won', ' Won', |
| |
| ' wrote', 'wrote', |
| |
| ' lost', |
| |
| ' started', |
| |
| ' spent', |
| |
| ' bought', ' Bought', |
| |
| ' gave', |
| |
| ' sent', |
| |
| ' happened', |
| |
| ' created', |
| |
| ' received', |
| |
| ' signed', |
| |
| ' voted', |
| |
| ' fell', |
| |
| ' arrived', |
| |
| ' moved', |
| |
| ' ran', |
| |
| ' passed', ' Passed', |
| |
| ' tried', |
| |
| ' broke', |
| |
| ' missed', |
| |
| ' raised', |
| |
| ' chose', |
| |
| ' learned', ' Learned', |
| |
| ' pulled', |
| |
| ' knew', |
| |
| ' changed', ' Changed', 'changed', |
| |
| ' dropped', |
| |
| ' grew', |
| |
| ' scored', |
| |
| ' promised', |
| |
| ' picked', |
| |
| ' managed', |
| |
| ' kept', |
| |
| ' suffered', |
| |
| ' purchased', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'made-family', 1 |
|
|
| def prog_L08F21979(input_tokens): |
| """1 token families, 5 surface forms (train F1 of the uncapped optimum: 0.555). |
| Fires at 42,725 of 36,786,566 positions.""" |
| |
| """prev SAE description: occurrences of the word "first." """ |
| |
| fire = { |
| |
| ' first', ' First', 'First', 'first', ' FIRST', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'first-family', 1 |
|
|
| def prog_L08F21980(input_tokens): |
| """8 token families, 19 surface forms (train F1 of the uncapped optimum: 0.112). |
| Fires at 123,354 of 36,786,566 positions.""" |
| |
| """prev SAE description: technical programming languages and their syntax""" |
| |
| fire = { |
| |
| ' �', |
| |
| 'h', ' h', |
| |
| 'k', ' k', |
| |
| 'b', ' b', |
| |
| 'n', ' n', 'én', |
| |
| 'AP', 'ap', ' ap', ' Ap', 'Ap', |
| |
| 'f', ' f', |
| |
| 'v', ' v', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '�-family', 1 |
|
|
| def prog_L08F21981(input_tokens): |
| """6 token families, 7 surface forms (train F1 of the uncapped optimum: 0.062). |
| Fires at 39,531 of 36,786,566 positions.""" |
| |
| """prev SAE description: query-based phrasing or questions regarding knowledge and understanding""" |
| |
| fire = { |
| |
| ' you', ' YOU', |
| |
| ' do', |
| |
| ' go', |
| |
| ' trying', |
| |
| ' begin', |
| |
| ' possibly', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'you-family', 1 |
|
|
| def prog_L08F21982(input_tokens): |
| """4 token families, 5 surface forms (train F1 of the uncapped optimum: 0.077). |
| Fires at 18,974 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases indicating locations or proximity to various areas""" |
| |
| fire = { |
| |
| ' beyond', ' Beyond', |
| |
| ' above', |
| |
| ' surrounding', |
| |
| ' abroad', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'beyond-family', 1 |
|
|
| def prog_L08F21983(input_tokens): |
| """11 token families, 17 surface forms (train F1 of the uncapped optimum: 0.616). |
| Fires at 146,546 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: terms related to geographical features and settings""" |
| |
| fire = { |
| |
| ' teenager', ' teenagers', |
| |
| ' legislature', ' Legislature', |
| |
| ' lately', |
| |
| ' congressman', ' Congressman', |
| |
| ' midfielder', |
| |
| ' courtesy', |
| |
| ' politician', ' politicians', |
| |
| ' WikiLeaks', ' Wikileaks', |
| |
| ' cryptocurrency', |
| |
| 'Introduction', ' Introduction', |
| |
| ' businessman', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'teenag-family', 1 |
|
|
| def prog_L08F21984(input_tokens): |
| """31 token families, 38 surface forms (train F1 of the uncapped optimum: 0.659). |
| Fires at 7,133 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to multi-year contracts or agreements""" |
| |
| fire = { |
| |
| ' multi', ' Multi', 'Multi', 'multi', |
| |
| ' mult', 'Mult', ' multic', 'mult', |
| |
| ' mega', |
| |
| 'illion', |
| |
| ' jugg', |
| |
| ' triple', |
| |
| ' Anglo', |
| |
| ' multim', |
| |
| ' quadru', |
| |
| ' multit', |
| |
| ' multif', |
| |
| ' socio', |
| |
| 'umbo', |
| |
| ' binge', |
| |
| ' multip', ' multiply', |
| |
| ' multitude', |
| |
| ' oft', |
| |
| 'ULT', |
| |
| ' Quad', |
| |
| ' omnip', |
| |
| ' multimedia', |
| |
| ' juggling', |
| |
| ' uber', |
| |
| ' quasi', |
| |
| ' intra', |
| |
| ' Integ', |
| |
| ' pepp', |
| |
| ' unic', |
| |
| 'single', |
| |
| ' Guatem', |
| |
| 'complex', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'multi-family', 1 |
|
|
| def prog_L08F21985(input_tokens): |
| """43 token families, 54 surface forms (train F1 of the uncapped optimum: 0.602). |
| Fires at 14,441 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to negative experiences or impacts""" |
| |
| fire = { |
| |
| ' negative', 'negative', ' Negative', ' negatives', |
| |
| ' positive', |
| |
| 'oid', |
| |
| ' ugly', |
| |
| ' depressing', ' depressive', ' depressed', |
| |
| 'verse', 'versely', |
| |
| ' scathing', |
| |
| ' disciplinary', |
| |
| ' damning', |
| |
| ' adverse', ' adversely', |
| |
| ' sour', ' Sour', |
| |
| ' unwanted', |
| |
| ' minus', 'minus', |
| |
| ' midterm', |
| |
| ' nude', |
| |
| ' downward', |
| |
| ' unhappy', |
| |
| ' negatively', |
| |
| ' subjective', |
| |
| ' nasty', |
| |
| 'rible', |
| |
| ' Ironically', 'Ironically', |
| |
| ' inaccurate', |
| |
| ' loser', ' losers', |
| |
| ' dispar', |
| |
| ' detrimental', |
| |
| ' downside', |
| |
| ' downgrade', |
| |
| ' derogatory', |
| |
| ' salty', |
| |
| ' traumatic', |
| |
| 'Neg', |
| |
| ' gloomy', |
| |
| ' pessimistic', |
| |
| ' negativity', |
| |
| ' Worse', |
| |
| ' unfavorable', |
| |
| ' disapproval', |
| |
| ' sensory', |
| |
| ' derivative', |
| |
| ' disappro', |
| |
| ' undesirable', |
| |
| ' detriment', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'negat-family', 1 |
|
|
| def prog_L08F21986(input_tokens): |
| """20 token families, 27 surface forms (train F1 of the uncapped optimum: 0.294). |
| Fires at 53,003 of 36,786,566 positions.""" |
| |
| """prev SAE description: sports outcomes and references to victories and defeats""" |
| |
| fire = { |
| |
| ' win', ' wins', |
| |
| ' victory', ' Victory', |
| |
| ' loss', 'loss', ' Loss', |
| |
| ' defeat', ' defeats', |
| |
| ' match', |
| |
| ' performance', |
| |
| ' opener', 'opening', |
| |
| ' draw', |
| |
| ' clash', |
| |
| ' winner', 'winner', |
| |
| ' sidelines', |
| |
| ' upset', |
| |
| 'final', |
| |
| ' friendly', |
| |
| ' tie', |
| |
| ' overtime', |
| |
| ' matchup', |
| |
| ' comeback', |
| |
| ' losses', |
| |
| ' victories', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'win-family', 1 |
|
|
| def prog_L08F21987(input_tokens): |
| """33 token families, 46 surface forms (train F1 of the uncapped optimum: 0.365). |
| Fires at 45,091 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to sports conferences and divisions""" |
| |
| fire = { |
| |
| ' League', ' league', |
| |
| ' division', ' Division', 'division', |
| |
| ' Conference', 'conference', |
| |
| ' playoffs', ' playoff', ' Playoff', ' Playoffs', |
| |
| ' Championship', |
| |
| ' Cup', ' Cups', |
| |
| ' Series', |
| |
| ' SEC', 'SEC', |
| |
| ' schedule', |
| |
| ' AFC', |
| |
| ' Eastern', |
| |
| ' FA', |
| |
| ' NFC', |
| |
| ' NCAA', |
| |
| ' Ten', |
| |
| ' ACC', 'ACC', |
| |
| ' standings', |
| |
| ' Stanley', |
| |
| 'HOU', |
| |
| 'CS', |
| |
| ' calendar', |
| |
| ' promotion', |
| |
| 'FC', |
| |
| ' AL', |
| |
| ' Wizards', |
| |
| ' prem', 'Prem', |
| |
| 'AST', |
| |
| ' Premiership', |
| |
| ' Pac', |
| |
| ' bracket', ' brackets', |
| |
| ' NL', 'NL', |
| |
| ' CONC', |
| |
| ' Cubs', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'league-family', 1 |
|
|
| def prog_L08F21988(input_tokens): |
| """2 token families, 2 surface forms (train F1 of the uncapped optimum: 0.055). |
| Fires at 39,606 of 36,786,566 positions.""" |
| |
| """prev SAE description: instances of strong affirmations or definitive statements""" |
| |
| fire = { |
| |
| '\n', |
| |
| ' Salt', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '\\n-family', 1 |
|
|
| def prog_L08F21989(input_tokens): |
| """6 token families, 9 surface forms (train F1 of the uncapped optimum: 0.583). |
| Fires at 11,061 of 36,786,566 positions.""" |
| |
| """prev SAE description: mentions of the concept of "role" in various contexts""" |
| |
| fire = { |
| |
| ' role', ' Role', 'Role', 'role', |
| |
| ' involvement', |
| |
| ' roles', |
| |
| ' participation', |
| |
| ' responsibilities', |
| |
| ' implicated', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'role-family', 1 |
|
|
| def prog_L08F21990(input_tokens): |
| """4 token families, 5 surface forms (train F1 of the uncapped optimum: 0.094). |
| Fires at 126,431 of 36,786,566 positions.""" |
| |
| """prev SAE description: quotes or speech from characters""" |
| |
| fire = { |
| |
| ' —', |
| |
| ' --', |
| |
| ' says', |
| |
| ',"', ' ,"', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return '—-family', 1 |
|
|
| def prog_L08F21991(input_tokens): |
| """16 token families, 25 surface forms (train F1 of the uncapped optimum: 0.109). |
| Fires at 193,995 of 36,786,566 positions. |
| Fires at 0.53% 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: topics that have garnered significant attention or coverage""" |
| |
| fire = { |
| |
| ' a', |
| |
| ' its', |
| |
| ' more', |
| |
| ' new', |
| |
| ' some', |
| |
| ' much', ' MUCH', |
| |
| ' little', 'little', |
| |
| ' attention', ' Attention', |
| |
| ' received', ' receiving', |
| |
| ' another', |
| |
| ' lot', |
| |
| ' big', |
| |
| ' same', ' Same', |
| |
| ' major', 'major', |
| |
| ' gained', ' gaining', ' gain', ' gains', |
| |
| ' second', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'a-family', 1 |
|
|
| def prog_L08F21992(input_tokens): |
| """2 token families, 8 surface forms (train F1 of the uncapped optimum: 0.708). |
| Fires at 19,357 of 36,786,566 positions.""" |
| |
| """prev SAE description: phrases that indicate something is "best known" or notable for a specific reason""" |
| |
| fire = { |
| |
| ' best', ' Best', 'Best', 'best', ' BEST', |
| |
| ' worst', 'worst', ' Worst', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'best-family', 1 |
|
|
| def prog_L08F21993(input_tokens): |
| """9 token families, 15 surface forms (train F1 of the uncapped optimum: 0.671). |
| Fires at 7,144 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to a specific topic or subject within various contexts""" |
| |
| fire = { |
| |
| ' subject', ' subjects', ' Subject', 'Subject', 'subject', ' Subjects', |
| |
| ' topic', |
| |
| ' exempt', 'exempt', |
| |
| 'JECT', |
| |
| ' Prophet', |
| |
| ' commencement', |
| |
| 'object', |
| |
| ' Perse', |
| |
| ' propriet', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'subject-family', 1 |
|
|
| def prog_L08F21994(input_tokens): |
| """31 token families, 37 surface forms (train F1 of the uncapped optimum: 0.629). |
| Fires at 12,433 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the continent of Asia and related discussions""" |
| |
| fire = { |
| |
| ' Asia', 'asia', 'Asia', |
| |
| ' Asian', ' Asians', 'Asian', |
| |
| ' Africa', |
| |
| ' overseas', |
| |
| ' Southeast', |
| |
| ' Americas', |
| |
| ' Electronics', |
| |
| ' Manila', |
| |
| ' Armenia', |
| |
| 'achi', |
| |
| 'adia', |
| |
| ' bush', |
| |
| ' Buddhism', |
| |
| ' Austral', 'Austral', |
| |
| ' Euras', |
| |
| 'rica', |
| |
| ' Arabian', |
| |
| 'orthern', |
| |
| ' Indo', ' indo', |
| |
| ' Lia', |
| |
| ' Anime', |
| |
| 'orea', |
| |
| ' Ae', |
| |
| ' Oriental', |
| |
| 'xia', |
| |
| 'NEY', |
| |
| 'thus', |
| |
| ' Caucasus', |
| |
| ' Peoples', |
| |
| ' aeros', |
| |
| ' Malays', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'asia-family', 1 |
|
|
| def prog_L08F21995(input_tokens): |
| """2 token families, 3 surface forms (train F1 of the uncapped optimum: 0.036). |
| Fires at 25,418 of 36,786,566 positions.""" |
| |
| """prev SAE description: author names and publication details in academic references""" |
| |
| fire = { |
| |
| ',', ' ,', |
| |
| ' et', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return ',-family', 1 |
|
|
| def prog_L08F21996(input_tokens): |
| """50 token families, 53 surface forms (train F1 of the uncapped optimum: 0.657). |
| Fires at 35,075 of 36,786,566 positions.""" |
| |
| """prev SAE description: mentions of the name "Lloyd" in various contexts""" |
| |
| fire = { |
| |
| ' Lane', |
| |
| 'li', |
| |
| ' Baker', ' baker', |
| |
| 'worth', |
| |
| 'INGTON', 'ington', |
| |
| ' Floyd', |
| |
| 'aki', |
| |
| ' Elliott', |
| |
| ' Ellis', |
| |
| ' Murdoch', |
| |
| ' Goldman', |
| |
| ' Moss', |
| |
| ' Villa', |
| |
| ' Self', 'Self', |
| |
| ' Lloyd', |
| |
| ' Banks', |
| |
| ' Hernandez', |
| |
| 'iana', |
| |
| ' Gonzalez', |
| |
| ' Tayyip', |
| |
| ' Shah', |
| |
| ' Davies', |
| |
| ' Reader', |
| |
| ' Friedman', |
| |
| ' sandwich', |
| |
| ' Saf', |
| |
| ' Messi', |
| |
| ' Nic', |
| |
| ' palm', |
| |
| ' Durant', |
| |
| ' Churchill', |
| |
| ' Allan', |
| |
| ' Mariners', |
| |
| ' Elliot', |
| |
| ' Chandler', |
| |
| ' Boyd', |
| |
| ' Fitzgerald', |
| |
| ' Vick', |
| |
| ' Lennon', |
| |
| ' Dwight', |
| |
| ' Longh', |
| |
| ' Diana', |
| |
| ' Randall', |
| |
| ' Mavericks', |
| |
| ' Doyle', |
| |
| 'elman', |
| |
| ' Milton', |
| |
| ' Lionel', |
| |
| ' Zion', |
| |
| ' Walton', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'lane-family', 1 |
|
|
| def prog_L08F21997(input_tokens): |
| """13 token families, 33 surface forms (train F1 of the uncapped optimum: 0.072). |
| Fires at 153,739 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: references to specific media, brands, or notable events""" |
| |
| fire = { |
| |
| 's', |
| |
| ' this', ' THIS', |
| |
| ' game', ' Game', ' GAME', |
| |
| ' these', |
| |
| ' system', ' systems', ' System', |
| |
| ' campaign', ' Campaign', 'campaign', |
| |
| ' book', ' Book', 'book', ' Books', ' BOOK', |
| |
| ' team', |
| |
| ' post', ' Post', ' postings', ' POST', |
| |
| ' series', ' Series', ' Serial', |
| |
| ' life', ' Life', |
| |
| ' Bitcoin', ' bitcoin', ' Bitcoins', |
| |
| ' blog', ' Blog', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 's-family', 1 |
|
|
| def prog_L08F21998(input_tokens): |
| """1 token families, 5 surface forms (train F1 of the uncapped optimum: 0.499). |
| Fires at 14,577 of 36,786,566 positions.""" |
| |
| """prev SAE description: references to the concept of "state," particularly in various contexts""" |
| |
| fire = { |
| |
| ' state', 'state', 'State', ' STATE', 'STATE', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'state-family', 1 |
|
|
| def prog_L08F21999(input_tokens): |
| """5 token families, 13 surface forms (train F1 of the uncapped optimum: 0.115). |
| Fires at 23,085 of 36,786,566 positions.""" |
| |
| """prev SAE description: terms related to subscription add-ons and features""" |
| |
| fire = { |
| |
| ' add', ' Add', 'add', ' adds', 'Add', 'Adds', ' ADD', 'adding', 'ADD', |
| |
| ' DLC', |
| |
| 'IN', |
| |
| 'ons', |
| |
| 'ison', |
| } |
| token = input_tokens[-1] |
| if token not in fire: |
| return "no match", 0 |
| return 'add-family', 1 |
|
|
