-
Notifications
You must be signed in to change notification settings - Fork 25
/
identify_associations.py
304 lines (263 loc) · 12.3 KB
/
identify_associations.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
__all__ = ["identify_associations"]
from functools import reduce
from pathlib import Path
from typing import Literal, cast
import hydra
import numpy as np
import pandas as pd
import torch
from omegaconf import OmegaConf
from torch.utils.data import SequentialSampler
from move.conf.schema import (
IdentifyAssociationsBayesConfig,
IdentifyAssociationsConfig,
IdentifyAssociationsTTestConfig,
MOVEConfig,
)
from move.core.logging import get_logger
from move.core.typing import IntArray, FloatArray
from move.data import io
from move.data.dataloaders import MOVEDataset, make_dataloader
from move.data.perturbations import perturb_data
from move.data.preprocessing import one_hot_encode_single
from move.models.vae import VAE
TaskType = Literal["bayes", "ttest"]
def _get_task_type(
task_config: IdentifyAssociationsConfig,
) -> TaskType:
task_type = OmegaConf.get_type(task_config)
if task_type is IdentifyAssociationsBayesConfig:
return "bayes"
if task_type is IdentifyAssociationsTTestConfig:
return "ttest"
raise ValueError("Unsupported type of task!")
def _validate_task_config(
task_config: IdentifyAssociationsConfig, task_type: TaskType
) -> None:
if not (0.0 <= task_config.sig_threshold <= 1.0):
raise ValueError("Significance threshold must be within [0, 1].")
if task_type == "ttest":
task_config = cast(IdentifyAssociationsTTestConfig, task_config)
if len(task_config.num_latent) != 4:
raise ValueError("4 latent space dimensions required.")
def identify_associations(config: MOVEConfig):
"""Trains multiple models to identify associations between the dataset
of interest and the continuous datasets."""
logger = get_logger(__name__)
task_config = cast(IdentifyAssociationsConfig, config.task)
task_type = _get_task_type(task_config)
logger.info(f"Beginning task: identify associations ({task_type})")
_validate_task_config(task_config, task_type)
interim_path = Path(config.data.interim_data_path)
models_path = interim_path / "models"
if task_config.save_refits:
models_path.mkdir(exist_ok=True)
output_path = Path(config.data.processed_data_path) / "identify_associations"
output_path.mkdir(exist_ok=True, parents=True)
# Read original data and create perturbed datasets
logger.info(f"Perturbing dataset: '{task_config.target_dataset}'")
cat_list, cat_names, con_list, con_names = io.load_preprocessed_data(
interim_path,
config.data.categorical_names,
config.data.continuous_names,
)
mappings = io.load_mappings(interim_path / "mappings.json")
target_mapping = mappings[task_config.target_dataset]
target_value = one_hot_encode_single(target_mapping, task_config.target_value)
logger.debug(
f"Target value: {task_config.target_value} => {target_value.astype(int)[0]}"
)
train_mask, train_dataloader = make_dataloader(
cat_list,
con_list,
shuffle=True,
batch_size=task_config.batch_size,
drop_last=True,
)
logger.debug(f"Masked training samples: {np.sum(~train_mask)}/{train_mask.size}")
con_shapes = [con.shape[1] for con in con_list]
dataloaders = perturb_data(
cat_list,
con_list,
config.data.categorical_names,
task_config.target_dataset,
target_value,
)
baseline_dataloader = dataloaders[-1]
baseline_dataset = cast(MOVEDataset, baseline_dataloader.dataset)
num_perturbed = len(dataloaders) - 1 # P
num_samples = len(cast(SequentialSampler, baseline_dataloader.sampler)) # N
num_continuous = sum(con_shapes) # C
logger.debug(f"# perturbed features: {num_perturbed}")
logger.debug(f"# continuous features: {num_continuous}")
assert baseline_dataset.con_all is not None
orig_con = baseline_dataset.con_all
nan_mask = (orig_con == 0).numpy() # NaN values encoded as 0s
logger.debug(f"# NaN values: {np.sum(nan_mask)}/{orig_con.numel()}")
target_dataset_idx = config.data.categorical_names.index(task_config.target_dataset)
target_dataset = cat_list[target_dataset_idx]
feature_mask = np.all(target_dataset == target_value, axis=2) # 2D: N x P
feature_mask |= np.sum(target_dataset, axis=2) == 0
def _bayes_approach(
task_config: IdentifyAssociationsBayesConfig,
) -> tuple[IntArray, FloatArray]:
assert task_config.model is not None
# Train models
logger.info("Training models")
mean_diff = np.zeros((num_perturbed, num_samples, num_continuous))
normalizer = 1 / task_config.num_refits
for j in range(task_config.num_refits):
# Initialize model
model: VAE = hydra.utils.instantiate(
task_config.model,
continuous_shapes=baseline_dataset.con_shapes,
categorical_shapes=baseline_dataset.cat_shapes,
)
if j == 0:
logger.debug(f"Model: {model}")
# Train/reload model
model_path = models_path / f"model_{task_config.model.num_latent}_{j}.pt"
if model_path.exists():
logger.debug(f"Re-loading refit {j + 1}/{task_config.num_refits}")
model.load_state_dict(torch.load(model_path))
else:
logger.debug(f"Training refit {j + 1}/{task_config.num_refits}")
hydra.utils.call(
task_config.training_loop,
model=model,
train_dataloader=train_dataloader,
)
if task_config.save_refits:
torch.save(model.state_dict(), model_path)
model.eval()
# Calculate baseline reconstruction
_, baseline_recon = model.reconstruct(baseline_dataloader)
# Calculate perturb reconstruction => keep track of mean difference
for i in range(num_perturbed):
_, perturb_recon = model.reconstruct(dataloaders[i])
diff = perturb_recon - baseline_recon # 2D: N x C
mean_diff[i, :, :] += diff * normalizer
# Calculate Bayes factors
logger.info("Identifying significant features")
bayes_k = np.empty((num_perturbed, num_continuous))
for i in range(num_perturbed):
mask = feature_mask[:, [i]] | nan_mask # 2D: N x C
diff = np.ma.masked_array(mean_diff[i, :, :], mask=mask) # 2D: N x C
prob = np.ma.compressed(np.mean(diff > 1e-8, axis=0)) # 1D: C
bayes_k[i, :] = np.abs(np.log(prob + 1e-8) - np.log(1 - prob + 1e-8))
# Calculate Bayes probabilities
bayes_p = np.exp(bayes_k) / (1 + np.exp(bayes_k)) # 2D: N x C
sort_ids = np.argsort(bayes_k, axis=None)[::-1] # 1D: N x C
prob = np.take(bayes_p, sort_ids) # 1D: N x C
logger.debug(f"Bayes proba range: [{prob[-1]:.3f} {prob[0]:.3f}]")
# Calculate FDR
fdr = np.cumsum(1 - prob) / np.arange(1, prob.size + 1) # 1D
idx = np.argmin(np.abs(fdr - task_config.sig_threshold))
logger.debug(f"FDR range: [{fdr[0]:.3f} {fdr[-1]:.3f}]")
return sort_ids[:idx], prob[:idx]
def _ttest_approach(
task_config: IdentifyAssociationsTTestConfig,
) -> tuple[IntArray, FloatArray]:
from scipy.stats import ttest_rel
# Train models
logger.info("Training models")
pvalues = np.empty(
(
len(task_config.num_latent),
task_config.num_refits,
num_perturbed,
num_continuous,
)
)
for k, num_latent in enumerate(task_config.num_latent):
for j in range(task_config.num_refits):
# Initialize model
model: VAE = hydra.utils.instantiate(
task_config.model,
continuous_shapes=baseline_dataset.con_shapes,
categorical_shapes=baseline_dataset.cat_shapes,
num_latent=num_latent,
)
if j == 0:
logger.debug(f"Model: {model}")
# Train model
model_path = models_path / f"model_{num_latent}_{j}.pt"
if model_path.exists():
logger.debug(f"Re-loading refit {j + 1}/{task_config.num_refits}")
model.load_state_dict(torch.load(model_path))
else:
logger.debug(f"Training refit {j + 1}/{task_config.num_refits}")
hydra.utils.call(
task_config.training_loop,
model=model,
train_dataloader=train_dataloader,
)
if task_config.save_refits:
torch.save(model.state_dict(), model_path)
model.eval()
# Get baseline reconstruction and baseline difference
_, baseline_recon = model.reconstruct(baseline_dataloader)
baseline_diff = np.empty((10, num_samples, num_continuous))
for i in range(10):
_, recon = model.reconstruct(baseline_dataloader)
baseline_diff[i, :, :] = recon - baseline_recon
baseline_diff = np.mean(baseline_diff, axis=0) # 2D: N x C
baseline_diff = np.where(nan_mask, np.nan, baseline_diff)
# T-test between baseline and perturb difference
for i in range(num_perturbed):
_, perturb_recon = model.reconstruct(dataloaders[i])
perturb_diff = perturb_recon - baseline_recon
mask = feature_mask[:, [i]] | nan_mask
_, pvalues[k, j, i, :] = ttest_rel(
a=np.where(mask, np.nan, perturb_diff),
b=np.where(mask, np.nan, baseline_diff),
axis=0,
nan_policy="omit",
)
# Correct p-values (Bonferroni)
pvalues = np.minimum(pvalues * num_continuous, 1.0)
np.save(interim_path / "pvals.npy", pvalues)
# Find significant hits
overlap_thres = task_config.num_refits // 2
reject = pvalues <= task_config.sig_threshold # 4D: L x R x P x C
overlap = reject.sum(axis=1) >= overlap_thres # 3D: L x P x C
sig_ids = overlap.sum(axis=0) >= 3 # 2D: P x C
sig_ids = np.flatnonzero(sig_ids) # 1D
# Report median p-value
masked_pvalues = np.ma.masked_array(pvalues, mask=~reject) # 4D
masked_pvalues = np.ma.median(masked_pvalues, axis=1) # 3D
masked_pvalues = np.ma.median(masked_pvalues, axis=0) # 2D
sig_pvalues = np.ma.compressed(np.take(masked_pvalues, sig_ids)) # 1D
return sig_ids, sig_pvalues
if task_type == "bayes":
task_config = cast(IdentifyAssociationsBayesConfig, task_config)
sig_ids, *extra_cols = _bayes_approach(task_config)
extra_colnames = ["proba"]
else:
task_config = cast(IdentifyAssociationsTTestConfig, task_config)
sig_ids, *extra_cols = _ttest_approach(task_config)
extra_colnames = ["p_value"]
# Prepare results
logger.info(f"Significant hits found: {sig_ids.size}")
if sig_ids.size > 0:
sig_ids = np.vstack((sig_ids // num_continuous, sig_ids % num_continuous)).T
logger.info("Writing results")
results = pd.DataFrame(sig_ids, columns=["feature_a_id", "feature_b_id"])
results.sort_values("feature_a_id", inplace=True)
a_df = pd.DataFrame(dict(feature_a_name=cat_names[target_dataset_idx]))
a_df.index.name = "feature_a_id"
a_df.reset_index(inplace=True)
con_names = reduce(list.__add__, con_names)
b_df = pd.DataFrame(dict(feature_b_name=con_names))
b_df.index.name = "feature_b_id"
b_df.reset_index(inplace=True)
results = results.merge(a_df, on="feature_a_id").merge(b_df, on="feature_b_id")
results["feature_b_dataset"] = pd.cut(
results["feature_b_id"],
bins=np.cumsum([0] + con_shapes),
right=False,
labels=config.data.continuous_names,
)
for col, colname in zip(extra_cols, extra_colnames):
results[colname] = col
results.to_csv(output_path / "results_sig_assoc.tsv", sep="\t", index=False)