-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathclustering_experiments.py
183 lines (160 loc) · 8.87 KB
/
clustering_experiments.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
import argparse
import os
import itertools
import pickle
import json
import zlib
import pandas as pd
import tensorflow as tf
from callbacks import PerformanceMonitor
from datasets import load_tensorflow_dataset
from models import build_encoder, build_decoder, select_model
from priors import select_prior
from utils import sample_data_indices, clustering_performance
# script arguments
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', type=str, default='mnist')
parser.add_argument('--latent_dim', type=int, default=10)
parser.add_argument('--learning_rate', type=float, default=1e-4)
parser.add_argument('--max_epochs', type=int, default=10000)
parser.add_argument('--mode', type=str)
parser.add_argument('--num_trials', type=int, default=10)
parser.add_argument('--replace', action='store_true', default=False, help='force saved model replacement')
parser.add_argument('--seed', type=int, default=1)
args = parser.parse_args()
# check arguments
assert args.mode in {'tuning', 'testing'}
# enable GPU determinism
tf.config.experimental.enable_op_determinism()
# make experiment directory
exp_path = os.path.join('experiments', 'clustering', str(args.seed), args.dataset, 'dim_{:d}'.format(args.latent_dim))
os.makedirs(exp_path, exist_ok=True)
# load dataset
x_train, labels_train, x_valid, labels_valid = load_tensorflow_dataset(args.dataset)
num_classes = int(tf.shape(tf.unique(labels_train)[0])[0])
# tuning mode test cases
if args.mode == 'tuning':
test_cases = [
dict(model='EmpiricalBayesVariationalAutoencoder',
model_kwargs=dict(batch_size=bs, learning_rate=args.learning_rate),
prior='VampPriorMixture',
prior_kwargs=dict(num_clusters=100, inference='MAP-DP', prior_learning_ratio=plr))
for bs, plr in list(itertools.product([128, 256, 512, 1024], [1.0]))]
# testing mode test cases
elif args.mode == 'testing':
try:
df = pd.read_pickle(os.path.join(exp_path, 'tuning_performance.pkl'))
df = df.groupby(level=['batch_size', 'learning_rate', 'prior_learning_ratio']).mean()
bs, lr, plr = df['NMI (unmerged)'].idxmax()
batch_size, learning_rate, prior_learning_ratio = int(bs), float(lr), float(plr)
except FileNotFoundError:
raise FileNotFoundError('please run script in tuning mode first')
# models, priors, and configurations to test
test_cases = [
dict(model='VariationalAutoencoder',
model_kwargs=dict(batch_size=batch_size, learning_rate=learning_rate),
prior='VampPrior',
prior_kwargs=dict(num_clusters=100, inference=None, prior_learning_ratio=None)),
dict(model='VariationalDeepEmbedding',
model_kwargs=dict(batch_size=batch_size, learning_rate=learning_rate),
prior='GaussianMixture',
prior_kwargs=dict(num_clusters=num_classes, inference=None, prior_learning_ratio=None)),
# dict(model='EmpiricalBayesVariationalAutoencoder',
# model_kwargs=dict(batch_size=batch_size, learning_rate=learning_rate),
# prior='GaussianMixture',
# prior_kwargs=dict(num_clusters=num_classes, inference='MLE', prior_learning_ratio=prior_learning_ratio)),
dict(model='EmpiricalBayesVariationalAutoencoder',
model_kwargs=dict(batch_size=batch_size, learning_rate=learning_rate),
prior='GaussianMixture',
prior_kwargs=dict(num_clusters=100, inference='MAP', prior_learning_ratio=prior_learning_ratio)),
dict(model='EmpiricalBayesVariationalAutoencoder',
model_kwargs=dict(batch_size=batch_size, learning_rate=learning_rate),
prior='GaussianMixture',
prior_kwargs=dict(num_clusters=100, inference='MAP-DP', prior_learning_ratio=prior_learning_ratio)),
# dict(model='EmpiricalBayesVariationalAutoencoder',
# model_kwargs=dict(batch_size=batch_size, learning_rate=learning_rate),
# prior='VampPriorMixture',
# prior_kwargs=dict(num_clusters=num_classes, inference='MLE', prior_learning_ratio=prior_learning_ratio)),
dict(model='EmpiricalBayesVariationalAutoencoder',
model_kwargs=dict(batch_size=batch_size, learning_rate=learning_rate),
prior='VampPriorMixture',
prior_kwargs=dict(num_clusters=100, inference='MAP', prior_learning_ratio=prior_learning_ratio)),
dict(model='EmpiricalBayesVariationalAutoencoder',
model_kwargs=dict(batch_size=batch_size, learning_rate=learning_rate),
prior='VampPriorMixture',
prior_kwargs=dict(num_clusters=100, inference='MAP-DP', prior_learning_ratio=prior_learning_ratio)),
]
else:
raise NotImplementedError
# loop over the trials and test cases
performance = pd.DataFrame()
elbo_surgery = pd.DataFrame()
for trial in range(1, 1 + args.num_trials):
for i, test_case in enumerate(test_cases):
print('*** Trial {:d}/{:d} | Case {:d}/{:d} ***'.format(trial, args.num_trials, i + 1, len(test_cases)))
# a deterministic but seemingly random transformation of the experiment seed into a trial seed
trial_seed = int(zlib.crc32(str(trial * (args.seed or 1)).encode())) % (2 ** 32 - 1)
tf.keras.utils.set_random_seed(trial_seed)
# VampPrior pseudo-input initialization
u_init = tf.gather(x_train, sample_data_indices(x_train, test_case['prior_kwargs']['num_clusters']))
# select prior
prior_lr = test_case['prior_kwargs'].get('prior_learning_ratio')
prior_lr = None if prior_lr is None else test_case['model_kwargs']['learning_rate'] * prior_lr
latent_prior = select_prior(test_case['prior'], **test_case['prior_kwargs'], **dict(
latent_dim=args.latent_dim,
learning_rate=prior_lr,
u=tf.Variable(u_init, name='u'),
))
# construct and compile model
model = select_model(test_case['model'])(
encoder=build_encoder(dim_x=x_train.shape.as_list()[1:]),
decoder=build_decoder(latent_dim=args.latent_dim, dim_x=x_train.shape.as_list()[1:]),
prior=latent_prior,
**test_case['model_kwargs'],
)
model.compile(optimizer=tf.keras.optimizers.Adam(test_case['model_kwargs']['learning_rate']))
# model save path
# test_case = copy.deepcopy(test_case)
# if hasattr(test_case['learning_rate'], 'get_config'):
# test_case['learning_rate'] = test_case['learning_rate'].get_config()
save_path = os.path.join(exp_path, 'trial_{:d}'.format(trial), test_case['model'], test_case['prior'])
save_path = os.path.join(save_path, 'config_{:d}'.format(zlib.crc32(json.dumps(test_case).encode('utf-8'))))
# if we are set to resume and the model directory already contains a saved model, load it
if not bool(args.replace) and os.path.exists(os.path.join(save_path, 'checkpoint')):
print('Loading existing model.')
checkpoint = tf.train.Checkpoint(model)
checkpoint.restore(os.path.join(save_path, 'best_checkpoint')).expect_partial()
# otherwise, fit and save the model
else:
hist = model.fit(
x=dict(x=x_train, label=labels_train),
validation_data=dict(x=x_valid, label=labels_valid),
batch_size=test_case['model_kwargs']['batch_size'],
epochs=args.max_epochs,
verbose=False,
callbacks=[PerformanceMonitor(monitor='val_nmi', patience=100)] + model.callbacks(x=x_train))
model.save_weights(os.path.join(save_path, 'best_checkpoint'))
with open(os.path.join(save_path, 'history.pkl'), 'wb') as f:
pickle.dump(hist.history, f)
# configuration index
index = pd.MultiIndex.from_frame(pd.DataFrame({
**{'model': [test_case['model']]},
**test_case['model_kwargs'],
**{'prior': [test_case['prior']]},
**test_case['prior_kwargs'],
**{'save_path': save_path}
}))
# save clustering performance
tf.keras.utils.set_random_seed(trial_seed)
all_data = dict(x=tf.concat([x_train, x_valid], axis=0), label=tf.concat([labels_train, labels_valid], axis=0))
latent_samples = model.predict(all_data, verbose=False)
cluster_probs = model.cluster_probabilities(latent_samples)
performance = pd.concat([performance, clustering_performance(cluster_probs, all_data['label'], index)])
performance.to_pickle(os.path.join(exp_path, args.mode + '_performance.pkl'))
# save model performance
if args.mode == 'testing':
tf.keras.utils.set_random_seed(trial_seed)
elbo_surgery = pd.concat([elbo_surgery, pd.DataFrame(model.elbo_surgery(dict(x=x_valid)), index=index)])
elbo_surgery.to_pickle(os.path.join(exp_path, 'elbo_surgery.pkl'))
# clear model from memory
tf.keras.backend.clear_session()