purvansh01 commited on
Commit
474cf3f
·
verified ·
1 Parent(s): 5fae6cb

Add 15 tuned XGBoost models (native .ubj), model card, feature code, manifests

Browse files
.gitattributes CHANGED
@@ -33,3 +33,18 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ weights/gm12878/seed0.ubj filter=lfs diff=lfs merge=lfs -text
37
+ weights/gm12878/seed1.ubj filter=lfs diff=lfs merge=lfs -text
38
+ weights/gm12878/seed2.ubj filter=lfs diff=lfs merge=lfs -text
39
+ weights/gm12878/seed3.ubj filter=lfs diff=lfs merge=lfs -text
40
+ weights/gm12878/seed4.ubj filter=lfs diff=lfs merge=lfs -text
41
+ weights/k562/seed0.ubj filter=lfs diff=lfs merge=lfs -text
42
+ weights/k562/seed1.ubj filter=lfs diff=lfs merge=lfs -text
43
+ weights/k562/seed2.ubj filter=lfs diff=lfs merge=lfs -text
44
+ weights/k562/seed3.ubj filter=lfs diff=lfs merge=lfs -text
45
+ weights/k562/seed4.ubj filter=lfs diff=lfs merge=lfs -text
46
+ weights/tcell/seed0.ubj filter=lfs diff=lfs merge=lfs -text
47
+ weights/tcell/seed1.ubj filter=lfs diff=lfs merge=lfs -text
48
+ weights/tcell/seed2.ubj filter=lfs diff=lfs merge=lfs -text
49
+ weights/tcell/seed3.ubj filter=lfs diff=lfs merge=lfs -text
50
+ weights/tcell/seed4.ubj filter=lfs diff=lfs merge=lfs -text
MANIFEST.sha256 ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 3ce1faeec8366c8072538de7a8c08d31aa91ed3b55d909ef6a06884c7a172eae ./README.md
2
+ d1a434de867b20cb28b60a946aa60d8347d79dd5dc7e59ac43fe40235a8e225c ./manifest.json
3
+ 13f322cf067bedde8276b7aef8d4abfe144b58231dff9bf3b29675174d0a06b4 ./weights/gm12878/seed0.ubj
4
+ 88ab0d42d350cde0b6bfa94946486e55868a7684b332e47e80b3596390d04560 ./weights/gm12878/seed1.ubj
5
+ 9d363259df1c63a594e1b0cebac39acc75728c691c734aa87616dcd21670bb80 ./weights/gm12878/seed2.ubj
6
+ 1041a5e92e94bfd6657c3e2a1ef4d93100c0c830bcfc2880c319ec53cb6a3f92 ./weights/gm12878/seed3.ubj
7
+ 2b0c40701970b6408f2c63ad7b764cf6b8e121bce6a7f91f66749adeee8647f7 ./weights/gm12878/seed4.ubj
8
+ 4a219400927f238c7f8413f40f66230ad0603c30e4fd4e23af7e53938894b835 ./weights/k562/seed0.ubj
9
+ 7d96b8ccabd70e7b11e83b690fed0230e5e9936c3d6e2f62d024a27ec2b6e751 ./weights/k562/seed1.ubj
10
+ 4437e20b2cf2c68faf4a73fde01133346d6444699d736a46f58e6cc1380d434b ./weights/k562/seed2.ubj
11
+ e64ac6274f305464e5a2062a721cc4f08d50c8bd91239c0eb428667b85e60489 ./weights/k562/seed3.ubj
12
+ a6982dc9b6b329360d09bb3116ed8ff58bc619f9d57d9d197c0e62f49e51dbac ./weights/k562/seed4.ubj
13
+ 21a2fc2a0af208b0276e4d004cbf995700bf5a8677edb4877adec15d703ed745 ./weights/tcell/seed0.ubj
14
+ 9f7228f6111499e024e8676dc5cb468e19bd44e3f136fcc09e468ba60c43dd4a ./weights/tcell/seed1.ubj
15
+ 270958d5d2801f09d3cfbc5686accba0403d6b0c0a0697d1b34265876e462ec4 ./weights/tcell/seed2.ubj
16
+ 81203dfd753cd408fa6aeb67bf83f73ba9ae54aab5c5703f0f74ad347c87b7f9 ./weights/tcell/seed3.ubj
17
+ ead8c8f51c5fd8fcf0fb820799499fedd9c695699ad18efd191a164524e32478 ./weights/tcell/seed4.ubj
18
+ 5298bc0bb86e84441be634d21239b7715b73497b6e9999084e7f0b52d894b641 ./xgb_features.py
README.md ADDED
@@ -0,0 +1,66 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ library_name: xgboost
4
+ pipeline_tag: tabular-classification
5
+ tags: [genomics, regulatory-elements, ENCODE, xgboost, kmer, dna, baseline]
6
+ datasets: [PxALabs/encode-regulatory-elements]
7
+ ---
8
+ # IRG XGBoost baselines (notebook 09)
9
+
10
+ Fifteen gradient-boosted classifiers (3 biosamples x 5 seeds) that predict whether a 1,000 bp window is an ENCODE candidate
11
+ cis-regulatory element (cCRE: PLS, pELS or dELS) of one biosample, against GC-matched genomic background, from **6-mer
12
+ frequencies** (4,096 features). Trained on `PxALabs/encode-regulatory-elements` v2.1, `*_gc_matched` tables. Part of
13
+ github.com/PxA-Labs/interpretable-regulatory-genomics. **No pretrained or language models are involved.** These are the tuned
14
+ reference baseline that CNNs and any later model must be compared against.
15
+
16
+ ## Files
17
+ - `weights/<biosample>/seed<N>.ubj` - XGBoost native binary models (`tcell`, `k562`, `gm12878`; seeds 0-4); no pickle
18
+ - `xgb_features.py` - the feature function (6-mer frequencies; windows containing N are skipped)
19
+ - `manifest.json` - per file: SHA-256, size, selected hyperparameters, best iteration, validation and test AUROC; `MANIFEST.sha256`
20
+
21
+ ## Usage
22
+ ```python
23
+ import xgboost as xgb
24
+ from xgb_features import kmer_features
25
+ clf = xgb.XGBClassifier(); clf.load_model("weights/k562/seed0.ubj")
26
+ p = clf.predict_proba(kmer_features(["ACGT" * 250], k=6))[:, 1] # probability of being a K562 cCRE
27
+ ```
28
+ Use all five seeds and report the spread; do not pick a seed on test data.
29
+
30
+ ## Training
31
+ Random search over 24 configurations on the **validation** chromosomes only (seed 0, early stopping on validation AUC, up to
32
+ 1,500 trees, patience 50). Search space: max_depth {4, 6, 8}, learning_rate {0.05, 0.1}, subsample {0.6, 0.8, 1.0},
33
+ colsample_bytree {0.3, 0.5, 0.8}, min_child_weight {1, 5, 10}, reg_lambda {1, 5, 20}; configuration 0 is the earlier default.
34
+ Selected: tcell (depth 6, lr 0.05, subsample 0.6, colsample 0.3, min_child_weight 10, lambda 1); k562 (depth 6, lr 0.05, subsample 1.0,
35
+ colsample 0.3, min_child_weight 5, lambda 20); gm12878 (depth 8, lr 0.05, subsample 1.0, colsample 0.3, min_child_weight 1, lambda 20).
36
+ Split: train chr1-14, X, Y / validation chr15-18 / test chr19-22. Test chromosomes were never used for selection.
37
+ Source: Kaggle kernel `purvanshj009/irg-09-xgb-tuned` v2, git `ff83f8f`, XGBoost on GPU.
38
+
39
+ ## Evaluation (test chromosomes, mean over 5 seeds; SD in brackets)
40
+ "Matched" = test subset matched on GC and CpG observed/expected, equal positives and negatives per bin (a sub-population; not comparable with the full-task AUROC).
41
+
42
+ | biosample | test AUROC | matched AUROC | untuned default (same code) |
43
+ |:--|--:|--:|--:|
44
+ | tcell | 0.8894 (0.0003) | 0.7431 (0.0012) | 0.8845 / 0.7347 |
45
+ | k562 | 0.8695 (0.0004) | 0.7686 (0.0011) | 0.8636 / 0.7614 |
46
+ | gm12878 | 0.8429 (0.0008) | 0.7526 (0.0011) | 0.8350 / 0.7452 |
47
+
48
+ Against the tuned CNNs of `PxALabs/irg-cnn-baselines`, these models are level on T cell and K562 and slightly behind on GM12878
49
+ (ShallowCNN +0.004 full, +0.012 matched). Full analysis and caveats: `docs/results/xgb_tuned.md` in the GitHub repository.
50
+
51
+ ## Intended use and limitations
52
+ - Reference baselines for sequence-based cCRE classification. Not for clinical use or variant-effect prediction.
53
+ - Each model is specific to the biosample it was trained on; cross-biosample use is not covered here.
54
+ - The negatives are GC-matched genomic windows outside the biosample's cCREs, not validated non-regulatory elements.
55
+ - **The search was truncated.** The selected configurations sit on the edge of the search space (lowest learning rate and colsample tried;
56
+ highest lambda in two biosamples), and the best iteration was 1,369-1,499 of the 1,500-tree cap, so these models are probably still
57
+ slightly under-tuned.
58
+ - Only k=6 was tuned. The 24-configuration search used one seed, and validation differences between configurations were small (about 0.005-0.007).
59
+ - Composition shortcuts (GC, CpG) are controlled only in the matched evaluation; repeat content is not controlled in training.
60
+ - The untuned-default comparison models were not saved.
61
+ - Data note: the project's earlier tables named `k562_*` were T-cell elements; these models use the corrected v2.1 tables.
62
+
63
+ ## Provenance and integrity
64
+ Every file was reloaded and its best iteration checked against the run log. Seed-0 models for each biosample and seed 3 for GM12878
65
+ were re-scored on the real test rows and reproduced the AUROC recorded during the run to four decimals (0.8895, 0.8698, 0.8419, 0.8436).
66
+ The other 11 were checked for loading and best iteration only.
manifest.json ADDED
@@ -0,0 +1,293 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "source_kernel": "purvanshj009/irg-09-xgb-tuned",
3
+ "kernel_version": "2",
4
+ "git_sha": "ff83f8f",
5
+ "features": "6-mer frequencies, 4096 columns, tools/hf/xgb_features.py",
6
+ "models": [
7
+ {
8
+ "file": "weights/gm12878/seed0.ubj",
9
+ "sha256": "13f322cf067bedde8276b7aef8d4abfe144b58231dff9bf3b29675174d0a06b4",
10
+ "biosample": "gm12878",
11
+ "seed": 0,
12
+ "bytes": 16726754,
13
+ "best_iteration": 1499,
14
+ "auroc_val": 0.8483,
15
+ "auroc_test": 0.8419,
16
+ "auroc_gc_cpg_matched": 0.751,
17
+ "params": {
18
+ "max_depth": 8,
19
+ "learning_rate": 0.05,
20
+ "subsample": 1.0,
21
+ "colsample_bytree": 0.3,
22
+ "min_child_weight": 1,
23
+ "reg_lambda": 20.0
24
+ }
25
+ },
26
+ {
27
+ "file": "weights/gm12878/seed1.ubj",
28
+ "sha256": "88ab0d42d350cde0b6bfa94946486e55868a7684b332e47e80b3596390d04560",
29
+ "biosample": "gm12878",
30
+ "seed": 1,
31
+ "bytes": 16792852,
32
+ "best_iteration": 1482,
33
+ "auroc_val": 0.8485,
34
+ "auroc_test": 0.8424,
35
+ "auroc_gc_cpg_matched": 0.7523,
36
+ "params": {
37
+ "max_depth": 8,
38
+ "learning_rate": 0.05,
39
+ "subsample": 1.0,
40
+ "colsample_bytree": 0.3,
41
+ "min_child_weight": 1,
42
+ "reg_lambda": 20.0
43
+ }
44
+ },
45
+ {
46
+ "file": "weights/gm12878/seed2.ubj",
47
+ "sha256": "9d363259df1c63a594e1b0cebac39acc75728c691c734aa87616dcd21670bb80",
48
+ "biosample": "gm12878",
49
+ "seed": 2,
50
+ "bytes": 15666515,
51
+ "best_iteration": 1354,
52
+ "auroc_val": 0.848,
53
+ "auroc_test": 0.8432,
54
+ "auroc_gc_cpg_matched": 0.7526,
55
+ "params": {
56
+ "max_depth": 8,
57
+ "learning_rate": 0.05,
58
+ "subsample": 1.0,
59
+ "colsample_bytree": 0.3,
60
+ "min_child_weight": 1,
61
+ "reg_lambda": 20.0
62
+ }
63
+ },
64
+ {
65
+ "file": "weights/gm12878/seed3.ubj",
66
+ "sha256": "1041a5e92e94bfd6657c3e2a1ef4d93100c0c830bcfc2880c319ec53cb6a3f92",
67
+ "biosample": "gm12878",
68
+ "seed": 3,
69
+ "bytes": 16765989,
70
+ "best_iteration": 1498,
71
+ "auroc_val": 0.8489,
72
+ "auroc_test": 0.8436,
73
+ "auroc_gc_cpg_matched": 0.7533,
74
+ "params": {
75
+ "max_depth": 8,
76
+ "learning_rate": 0.05,
77
+ "subsample": 1.0,
78
+ "colsample_bytree": 0.3,
79
+ "min_child_weight": 1,
80
+ "reg_lambda": 20.0
81
+ }
82
+ },
83
+ {
84
+ "file": "weights/gm12878/seed4.ubj",
85
+ "sha256": "2b0c40701970b6408f2c63ad7b764cf6b8e121bce6a7f91f66749adeee8647f7",
86
+ "biosample": "gm12878",
87
+ "seed": 4,
88
+ "bytes": 16802440,
89
+ "best_iteration": 1496,
90
+ "auroc_val": 0.8489,
91
+ "auroc_test": 0.8436,
92
+ "auroc_gc_cpg_matched": 0.7537,
93
+ "params": {
94
+ "max_depth": 8,
95
+ "learning_rate": 0.05,
96
+ "subsample": 1.0,
97
+ "colsample_bytree": 0.3,
98
+ "min_child_weight": 1,
99
+ "reg_lambda": 20.0
100
+ }
101
+ },
102
+ {
103
+ "file": "weights/k562/seed0.ubj",
104
+ "sha256": "4a219400927f238c7f8413f40f66230ad0603c30e4fd4e23af7e53938894b835",
105
+ "biosample": "k562",
106
+ "seed": 0,
107
+ "bytes": 5315689,
108
+ "best_iteration": 1493,
109
+ "auroc_val": 0.8775,
110
+ "auroc_test": 0.8698,
111
+ "auroc_gc_cpg_matched": 0.7698,
112
+ "params": {
113
+ "max_depth": 6,
114
+ "learning_rate": 0.05,
115
+ "subsample": 1.0,
116
+ "colsample_bytree": 0.3,
117
+ "min_child_weight": 5,
118
+ "reg_lambda": 20.0
119
+ }
120
+ },
121
+ {
122
+ "file": "weights/k562/seed1.ubj",
123
+ "sha256": "7d96b8ccabd70e7b11e83b690fed0230e5e9936c3d6e2f62d024a27ec2b6e751",
124
+ "biosample": "k562",
125
+ "seed": 1,
126
+ "bytes": 5310506,
127
+ "best_iteration": 1492,
128
+ "auroc_val": 0.8771,
129
+ "auroc_test": 0.8697,
130
+ "auroc_gc_cpg_matched": 0.7683,
131
+ "params": {
132
+ "max_depth": 6,
133
+ "learning_rate": 0.05,
134
+ "subsample": 1.0,
135
+ "colsample_bytree": 0.3,
136
+ "min_child_weight": 5,
137
+ "reg_lambda": 20.0
138
+ }
139
+ },
140
+ {
141
+ "file": "weights/k562/seed2.ubj",
142
+ "sha256": "4437e20b2cf2c68faf4a73fde01133346d6444699d736a46f58e6cc1380d434b",
143
+ "biosample": "k562",
144
+ "seed": 2,
145
+ "bytes": 5299434,
146
+ "best_iteration": 1493,
147
+ "auroc_val": 0.8772,
148
+ "auroc_test": 0.8694,
149
+ "auroc_gc_cpg_matched": 0.7687,
150
+ "params": {
151
+ "max_depth": 6,
152
+ "learning_rate": 0.05,
153
+ "subsample": 1.0,
154
+ "colsample_bytree": 0.3,
155
+ "min_child_weight": 5,
156
+ "reg_lambda": 20.0
157
+ }
158
+ },
159
+ {
160
+ "file": "weights/k562/seed3.ubj",
161
+ "sha256": "e64ac6274f305464e5a2062a721cc4f08d50c8bd91239c0eb428667b85e60489",
162
+ "biosample": "k562",
163
+ "seed": 3,
164
+ "bytes": 5293102,
165
+ "best_iteration": 1498,
166
+ "auroc_val": 0.8762,
167
+ "auroc_test": 0.8699,
168
+ "auroc_gc_cpg_matched": 0.7691,
169
+ "params": {
170
+ "max_depth": 6,
171
+ "learning_rate": 0.05,
172
+ "subsample": 1.0,
173
+ "colsample_bytree": 0.3,
174
+ "min_child_weight": 5,
175
+ "reg_lambda": 20.0
176
+ }
177
+ },
178
+ {
179
+ "file": "weights/k562/seed4.ubj",
180
+ "sha256": "a6982dc9b6b329360d09bb3116ed8ff58bc619f9d57d9d197c0e62f49e51dbac",
181
+ "biosample": "k562",
182
+ "seed": 4,
183
+ "bytes": 5295410,
184
+ "best_iteration": 1452,
185
+ "auroc_val": 0.878,
186
+ "auroc_test": 0.8689,
187
+ "auroc_gc_cpg_matched": 0.7669,
188
+ "params": {
189
+ "max_depth": 6,
190
+ "learning_rate": 0.05,
191
+ "subsample": 1.0,
192
+ "colsample_bytree": 0.3,
193
+ "min_child_weight": 5,
194
+ "reg_lambda": 20.0
195
+ }
196
+ },
197
+ {
198
+ "file": "weights/tcell/seed0.ubj",
199
+ "sha256": "21a2fc2a0af208b0276e4d004cbf995700bf5a8677edb4877adec15d703ed745",
200
+ "biosample": "tcell",
201
+ "seed": 0,
202
+ "bytes": 4198815,
203
+ "best_iteration": 1482,
204
+ "auroc_val": 0.8981,
205
+ "auroc_test": 0.8895,
206
+ "auroc_gc_cpg_matched": 0.7438,
207
+ "params": {
208
+ "max_depth": 6,
209
+ "learning_rate": 0.05,
210
+ "subsample": 0.6,
211
+ "colsample_bytree": 0.3,
212
+ "min_child_weight": 10,
213
+ "reg_lambda": 1.0
214
+ }
215
+ },
216
+ {
217
+ "file": "weights/tcell/seed1.ubj",
218
+ "sha256": "9f7228f6111499e024e8676dc5cb468e19bd44e3f136fcc09e468ba60c43dd4a",
219
+ "biosample": "tcell",
220
+ "seed": 1,
221
+ "bytes": 4188348,
222
+ "best_iteration": 1464,
223
+ "auroc_val": 0.8974,
224
+ "auroc_test": 0.8892,
225
+ "auroc_gc_cpg_matched": 0.742,
226
+ "params": {
227
+ "max_depth": 6,
228
+ "learning_rate": 0.05,
229
+ "subsample": 0.6,
230
+ "colsample_bytree": 0.3,
231
+ "min_child_weight": 10,
232
+ "reg_lambda": 1.0
233
+ }
234
+ },
235
+ {
236
+ "file": "weights/tcell/seed2.ubj",
237
+ "sha256": "270958d5d2801f09d3cfbc5686accba0403d6b0c0a0697d1b34265876e462ec4",
238
+ "biosample": "tcell",
239
+ "seed": 2,
240
+ "bytes": 3561878,
241
+ "best_iteration": 1191,
242
+ "auroc_val": 0.8962,
243
+ "auroc_test": 0.8895,
244
+ "auroc_gc_cpg_matched": 0.7419,
245
+ "params": {
246
+ "max_depth": 6,
247
+ "learning_rate": 0.05,
248
+ "subsample": 0.6,
249
+ "colsample_bytree": 0.3,
250
+ "min_child_weight": 10,
251
+ "reg_lambda": 1.0
252
+ }
253
+ },
254
+ {
255
+ "file": "weights/tcell/seed3.ubj",
256
+ "sha256": "81203dfd753cd408fa6aeb67bf83f73ba9ae54aab5c5703f0f74ad347c87b7f9",
257
+ "biosample": "tcell",
258
+ "seed": 3,
259
+ "bytes": 4084693,
260
+ "best_iteration": 1401,
261
+ "auroc_val": 0.8967,
262
+ "auroc_test": 0.8892,
263
+ "auroc_gc_cpg_matched": 0.7427,
264
+ "params": {
265
+ "max_depth": 6,
266
+ "learning_rate": 0.05,
267
+ "subsample": 0.6,
268
+ "colsample_bytree": 0.3,
269
+ "min_child_weight": 10,
270
+ "reg_lambda": 1.0
271
+ }
272
+ },
273
+ {
274
+ "file": "weights/tcell/seed4.ubj",
275
+ "sha256": "ead8c8f51c5fd8fcf0fb820799499fedd9c695699ad18efd191a164524e32478",
276
+ "biosample": "tcell",
277
+ "seed": 4,
278
+ "bytes": 3846613,
279
+ "best_iteration": 1307,
280
+ "auroc_val": 0.8955,
281
+ "auroc_test": 0.8898,
282
+ "auroc_gc_cpg_matched": 0.7448,
283
+ "params": {
284
+ "max_depth": 6,
285
+ "learning_rate": 0.05,
286
+ "subsample": 0.6,
287
+ "colsample_bytree": 0.3,
288
+ "min_child_weight": 10,
289
+ "reg_lambda": 1.0
290
+ }
291
+ }
292
+ ]
293
+ }
weights/gm12878/seed0.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:13f322cf067bedde8276b7aef8d4abfe144b58231dff9bf3b29675174d0a06b4
3
+ size 16726754
weights/gm12878/seed1.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:88ab0d42d350cde0b6bfa94946486e55868a7684b332e47e80b3596390d04560
3
+ size 16792852
weights/gm12878/seed2.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9d363259df1c63a594e1b0cebac39acc75728c691c734aa87616dcd21670bb80
3
+ size 15666515
weights/gm12878/seed3.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1041a5e92e94bfd6657c3e2a1ef4d93100c0c830bcfc2880c319ec53cb6a3f92
3
+ size 16765989
weights/gm12878/seed4.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b0c40701970b6408f2c63ad7b764cf6b8e121bce6a7f91f66749adeee8647f7
3
+ size 16802440
weights/k562/seed0.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4a219400927f238c7f8413f40f66230ad0603c30e4fd4e23af7e53938894b835
3
+ size 5315689
weights/k562/seed1.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7d96b8ccabd70e7b11e83b690fed0230e5e9936c3d6e2f62d024a27ec2b6e751
3
+ size 5310506
weights/k562/seed2.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4437e20b2cf2c68faf4a73fde01133346d6444699d736a46f58e6cc1380d434b
3
+ size 5299434
weights/k562/seed3.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e64ac6274f305464e5a2062a721cc4f08d50c8bd91239c0eb428667b85e60489
3
+ size 5293102
weights/k562/seed4.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a6982dc9b6b329360d09bb3116ed8ff58bc619f9d57d9d197c0e62f49e51dbac
3
+ size 5295410
weights/tcell/seed0.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:21a2fc2a0af208b0276e4d004cbf995700bf5a8677edb4877adec15d703ed745
3
+ size 4198815
weights/tcell/seed1.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9f7228f6111499e024e8676dc5cb468e19bd44e3f136fcc09e468ba60c43dd4a
3
+ size 4188348
weights/tcell/seed2.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:270958d5d2801f09d3cfbc5686accba0403d6b0c0a0697d1b34265876e462ec4
3
+ size 3561878
weights/tcell/seed3.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:81203dfd753cd408fa6aeb67bf83f73ba9ae54aab5c5703f0f74ad347c87b7f9
3
+ size 4084693
weights/tcell/seed4.ubj ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ead8c8f51c5fd8fcf0fb820799499fedd9c695699ad18efd191a164524e32478
3
+ size 3846613
xgb_features.py ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """k-mer frequency features used by the IRG XGBoost baselines (identical to kaggle/kernels/09-xgb-tuned).
2
+
3
+ from xgb_features import kmer_features
4
+ X = kmer_features(["ACGT" * 250], k=6) # (n, 4096) float32, each row sums to 1 (windows containing N are skipped)
5
+ """
6
+ from __future__ import annotations
7
+
8
+ import numpy as np
9
+
10
+ _LUT = np.full(256, -1, dtype=np.int64)
11
+ for _i, _c in enumerate("ACGT"):
12
+ _LUT[ord(_c)] = _i
13
+ _LUT[ord(_c.lower())] = _i
14
+
15
+
16
+ def kmer_features(seqs: list[str], k: int = 6) -> np.ndarray:
17
+ X = np.zeros((len(seqs), 4**k), dtype=np.float32)
18
+ for r, s in enumerate(seqs):
19
+ a = _LUT[np.frombuffer(s.encode(), dtype=np.uint8)]
20
+ n = len(a) - k + 1
21
+ idx, ok = np.zeros(n, dtype=np.int64), np.ones(n, dtype=bool)
22
+ for j in range(k):
23
+ seg = a[j : j + n]
24
+ ok &= seg >= 0
25
+ idx = idx * 4 + np.where(seg >= 0, seg, 0)
26
+ c = np.bincount(idx[ok], minlength=4**k).astype(np.float32)
27
+ X[r] = c / max(c.sum(), 1.0)
28
+ return X