Spaces:
Runtime error
Runtime error
consolidated cmdline args
Browse files
msma.py
CHANGED
@@ -2,7 +2,7 @@ import datetime
|
|
2 |
import json
|
3 |
import os
|
4 |
import pickle
|
5 |
-
from functools import partial
|
6 |
from pickle import dump, load
|
7 |
from typing import Literal
|
8 |
|
@@ -135,50 +135,6 @@ def quantile_scorer(gmm, X, y=None):
|
|
135 |
return np.quantile(gmm.score_samples(X), 0.1)
|
136 |
|
137 |
|
138 |
-
def train_gmm(score_path, outdir, grid_search=False):
|
139 |
-
X = torch.load(score_path)
|
140 |
-
|
141 |
-
gm = GaussianMixture(
|
142 |
-
n_components=7, init_params="kmeans", covariance_type="full", max_iter=100000
|
143 |
-
)
|
144 |
-
clf = Pipeline([("scaler", StandardScaler()), ("GMM", gm)])
|
145 |
-
|
146 |
-
if grid_search:
|
147 |
-
param_grid = dict(
|
148 |
-
GMM__n_components=range(2, 11, 1),
|
149 |
-
)
|
150 |
-
|
151 |
-
grid = GridSearchCV(
|
152 |
-
estimator=clf,
|
153 |
-
param_grid=param_grid,
|
154 |
-
cv=5,
|
155 |
-
n_jobs=2,
|
156 |
-
verbose=1,
|
157 |
-
scoring=quantile_scorer,
|
158 |
-
)
|
159 |
-
|
160 |
-
grid_result = grid.fit(X)
|
161 |
-
|
162 |
-
print("Best: %f using %s" % (grid_result.best_score_, grid_result.best_params_))
|
163 |
-
print("-----" * 15)
|
164 |
-
means = grid_result.cv_results_["mean_test_score"]
|
165 |
-
stds = grid_result.cv_results_["std_test_score"]
|
166 |
-
params = grid_result.cv_results_["params"]
|
167 |
-
for mean, stdev, param in zip(means, stds, params):
|
168 |
-
print("%f (%f) with: %r" % (mean, stdev, param))
|
169 |
-
clf = grid.best_estimator_
|
170 |
-
|
171 |
-
clf.fit(X)
|
172 |
-
inlier_nll = -clf.score_samples(X)
|
173 |
-
|
174 |
-
os.makedirs(outdir, exist_ok=True)
|
175 |
-
with open(f"{outdir}/refscores.npz", "wb") as f:
|
176 |
-
np.savez_compressed(f, inlier_nll)
|
177 |
-
|
178 |
-
with open(f"{outdir}/gmm.pkl", "wb") as f:
|
179 |
-
dump(clf, f, protocol=5)
|
180 |
-
|
181 |
-
|
182 |
def compute_gmm_likelihood(x_score, gmmdir):
|
183 |
with open(f"{gmmdir}/gmm.pkl", "rb") as f:
|
184 |
clf = load(f)
|
@@ -237,8 +193,9 @@ def cmdline():
|
|
237 |
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
238 |
|
239 |
|
240 |
-
|
241 |
-
@
|
|
|
242 |
"--preset",
|
243 |
help="Configuration preset",
|
244 |
metavar="STR",
|
@@ -246,20 +203,73 @@ def cmdline():
|
|
246 |
default="edm2-img64-s-fid",
|
247 |
show_default=True,
|
248 |
)
|
249 |
-
@click.option(
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
|
254 |
-
|
255 |
-
)
|
256 |
-
@click.option(
|
257 |
-
|
258 |
-
|
259 |
-
|
260 |
-
|
261 |
-
|
262 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
263 |
def cache_score_norms(preset, dataset_path, outdir):
|
264 |
device = DEVICE
|
265 |
dsobj = ImageFolderDataset(path=dataset_path, resolution=64)
|
@@ -290,28 +300,6 @@ def cache_score_norms(preset, dataset_path, outdir):
|
|
290 |
|
291 |
|
292 |
@cmdline.command(name="train-flow")
|
293 |
-
@click.option(
|
294 |
-
"--dataset_path",
|
295 |
-
help="Path to the dataset",
|
296 |
-
metavar="ZIP|DIR",
|
297 |
-
type=str,
|
298 |
-
default=None,
|
299 |
-
)
|
300 |
-
@click.option(
|
301 |
-
"--outdir",
|
302 |
-
help="Where to load/save the results",
|
303 |
-
metavar="DIR",
|
304 |
-
type=str,
|
305 |
-
required=True,
|
306 |
-
)
|
307 |
-
@click.option(
|
308 |
-
"--preset",
|
309 |
-
help="Configuration preset",
|
310 |
-
metavar="STR",
|
311 |
-
type=str,
|
312 |
-
default="edm2-img64-s-fid",
|
313 |
-
show_default=True,
|
314 |
-
)
|
315 |
@click.option(
|
316 |
"--epochs",
|
317 |
help="Number of epochs",
|
@@ -328,6 +316,7 @@ def cache_score_norms(preset, dataset_path, outdir):
|
|
328 |
default=4,
|
329 |
show_default=True,
|
330 |
)
|
|
|
331 |
def train_flow(dataset_path, preset, outdir, epochs, **flow_kwargs):
|
332 |
print("using device:", DEVICE)
|
333 |
device = DEVICE
|
|
|
2 |
import json
|
3 |
import os
|
4 |
import pickle
|
5 |
+
from functools import partial, wraps
|
6 |
from pickle import dump, load
|
7 |
from typing import Literal
|
8 |
|
|
|
135 |
return np.quantile(gmm.score_samples(X), 0.1)
|
136 |
|
137 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
138 |
def compute_gmm_likelihood(x_score, gmmdir):
|
139 |
with open(f"{gmmdir}/gmm.pkl", "rb") as f:
|
140 |
clf = load(f)
|
|
|
193 |
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
194 |
|
195 |
|
196 |
+
def common_args(func):
|
197 |
+
@wraps(func)
|
198 |
+
@click.option(
|
199 |
"--preset",
|
200 |
help="Configuration preset",
|
201 |
metavar="STR",
|
|
|
203 |
default="edm2-img64-s-fid",
|
204 |
show_default=True,
|
205 |
)
|
206 |
+
@click.option(
|
207 |
+
"--dataset_path",
|
208 |
+
help="Path to the dataset",
|
209 |
+
metavar="ZIP|DIR",
|
210 |
+
type=str,
|
211 |
+
default=None,
|
212 |
+
)
|
213 |
+
@click.option(
|
214 |
+
"--outdir",
|
215 |
+
help="Where to load/save the results",
|
216 |
+
metavar="DIR",
|
217 |
+
type=str,
|
218 |
+
required=True,
|
219 |
+
)
|
220 |
+
def wrapper(*args, **kwargs):
|
221 |
+
return func(*args, **kwargs)
|
222 |
+
|
223 |
+
return wrapper
|
224 |
+
|
225 |
+
@cmdline.command('train-gmm')
|
226 |
+
@common_args
|
227 |
+
def train_gmm(score_path, outdir, grid_search=False):
|
228 |
+
X = torch.load(score_path)
|
229 |
+
|
230 |
+
gm = GaussianMixture(
|
231 |
+
n_components=7, init_params="kmeans", covariance_type="full", max_iter=100000
|
232 |
+
)
|
233 |
+
clf = Pipeline([("scaler", StandardScaler()), ("GMM", gm)])
|
234 |
+
|
235 |
+
if grid_search:
|
236 |
+
param_grid = dict(
|
237 |
+
GMM__n_components=range(2, 11, 1),
|
238 |
+
)
|
239 |
+
|
240 |
+
grid = GridSearchCV(
|
241 |
+
estimator=clf,
|
242 |
+
param_grid=param_grid,
|
243 |
+
cv=5,
|
244 |
+
n_jobs=2,
|
245 |
+
verbose=1,
|
246 |
+
scoring=quantile_scorer,
|
247 |
+
)
|
248 |
+
|
249 |
+
grid_result = grid.fit(X)
|
250 |
+
|
251 |
+
print("Best: %f using %s" % (grid_result.best_score_, grid_result.best_params_))
|
252 |
+
print("-----" * 15)
|
253 |
+
means = grid_result.cv_results_["mean_test_score"]
|
254 |
+
stds = grid_result.cv_results_["std_test_score"]
|
255 |
+
params = grid_result.cv_results_["params"]
|
256 |
+
for mean, stdev, param in zip(means, stds, params):
|
257 |
+
print("%f (%f) with: %r" % (mean, stdev, param))
|
258 |
+
clf = grid.best_estimator_
|
259 |
+
|
260 |
+
clf.fit(X)
|
261 |
+
inlier_nll = -clf.score_samples(X)
|
262 |
+
|
263 |
+
os.makedirs(outdir, exist_ok=True)
|
264 |
+
with open(f"{outdir}/refscores.npz", "wb") as f:
|
265 |
+
np.savez_compressed(f, inlier_nll)
|
266 |
+
|
267 |
+
with open(f"{outdir}/gmm.pkl", "wb") as f:
|
268 |
+
dump(clf, f, protocol=5)
|
269 |
+
|
270 |
+
|
271 |
+
@cmdline.command(name="cache-scores")
|
272 |
+
@common_args
|
273 |
def cache_score_norms(preset, dataset_path, outdir):
|
274 |
device = DEVICE
|
275 |
dsobj = ImageFolderDataset(path=dataset_path, resolution=64)
|
|
|
300 |
|
301 |
|
302 |
@cmdline.command(name="train-flow")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
303 |
@click.option(
|
304 |
"--epochs",
|
305 |
help="Number of epochs",
|
|
|
316 |
default=4,
|
317 |
show_default=True,
|
318 |
)
|
319 |
+
@common_args
|
320 |
def train_flow(dataset_path, preset, outdir, epochs, **flow_kwargs):
|
321 |
print("using device:", DEVICE)
|
322 |
device = DEVICE
|