-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpaper_evaluation.py
215 lines (184 loc) · 8.65 KB
/
paper_evaluation.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
import pandas as pd
import numpy as np
import tensorflow as tf
from keras import models, metrics, callbacks, optimizers, utils, losses
import architectures
import hydra
from omegaconf import DictConfig, OmegaConf
import lib.utility_layer_test as utility_layer_test
from sklearn.model_selection import train_test_split
import logging
from sklearn.preprocessing import LabelEncoder
import wandb
from sklearn.utils import class_weight
import os
from config import ConfigFullstack
from constants import LABEL, FLOW_STATS_COLUMNS
import utilities
from data_processing import get_ppi_matrices, get_dataset
logging.basicConfig(level=logging.INFO)
LABEL = "LABEL"
# for reproducibility
utils.set_random_seed(0xdeadbeef)
@hydra.main(version_base=None, config_path="conf")
def main(cfg : DictConfig) -> None:
cfg_dict = OmegaConf.to_container(cfg, resolve=True)
config = ConfigFullstack(**cfg_dict)
os.environ["WANDB_INIT_TIMEOUT"] = "300"
os.environ["WANDB_HTTP_TIMEOUT"] = "300"
os.environ["WANDB__SERVICE_WAIT"] = "300"
run = wandb.init(
project=f"{config.tag}",
dir=os.path.expanduser("~"),
config=utilities.flatten_dict(cfg_dict)
)
config = dict(cfg_dict)
config.update(wandb.config)
config = ConfigFullstack(**config)
logging.info(f"Config: {config}")
timestamp = run.start_time
experiment_nickname = f"{run.name}.{timestamp}"
dataset = get_dataset(config)
class_count = dataset["LABEL"].nunique()
labels = dataset[LABEL].values
# encode labels to int
labels_encoder = LabelEncoder()
labels_encoded = labels_encoder.fit_transform(labels)
label_names = labels_encoder.classes_
wandb.config.update({
"class_count": class_count,
"experiment_nickname": experiment_nickname,
"label_names": label_names
})
wandb.log({"misc/config": wandb.Html("<pre>" + "\n".join(f"{k}: {v}" for k,v in vars(config).items()) + "</pre>")}, step=0)
logging.info(f"Class count: {class_count}")
logging.info(f"Label names: {label_names}")
# Split dataset into train, validation and test sets
train_indices, test_indices = train_test_split(dataset.index, test_size=0.3, stratify=labels_encoded)
train_indices, val_indices = train_test_split(train_indices, test_size=0.15, stratify=labels_encoded[train_indices])
value_count_train = pd.Series(labels_encoded[train_indices]).value_counts()
value_count_val = pd.Series(labels_encoded[val_indices]).value_counts()
value_count_test = pd.Series(labels_encoded[test_indices]).value_counts()
value_count_full = pd.Series(labels_encoded).value_counts()
logging.info(f"Full set value counts: {value_count_full}")
logging.info(f"Train set value counts: {value_count_train}")
logging.info(f"Validation set value counts: {value_count_val}")
logging.info(f"Test set value counts: {value_count_test}")
####################
# Prepare PPI data #
####################
train_ppi, val_ppi, test_ppi = get_ppi_matrices(dataset, train_indices, val_indices, test_indices)
##########################
# Prepare FlowStats data #
##########################
train_flow_stats = dataset.loc[train_indices, FLOW_STATS_COLUMNS]
val_flow_stats = dataset.loc[val_indices, FLOW_STATS_COLUMNS]
test_flow_stats = dataset.loc[test_indices, FLOW_STATS_COLUMNS]
# deleting dataset to free up memory
del dataset
# class weights
logging.info("Calculating class weights")
class_weights_calculate = class_weight.compute_class_weight(
class_weight='balanced',
classes=np.unique(labels),
y=labels
)
class_weights = {i: class_weights_calculate[i] for i in range(len(class_weights_calculate))}
logging.info(f"Class weights: {class_weights}")
################
# DST training #
################
if len(config.pretrained) > 0:
dst_model = models.load_model(f"{config.model_folder}/{config.pretrained}")
else:
logging.info("DST model train: Training DST model")
early_stopping = callbacks.EarlyStopping(
monitor='val_loss',
patience=50,
restore_best_weights=True,
verbose=1,
min_delta=0.0001
)
dst_model = architectures.DstModel_FullStack_EarlyFusion(
prototypes=config.prototypes,
class_num=class_count,
distance_metric=config.distance_metric,
ppi_reprezentation_size=config.ppi_reprezentation_size,
flowstats_reprezentation_size=config.flowstats_reprezentation_size,
processed_reprezentation_size=config.processed_reprezentation_size,
extra_layer=config.extra_layer,
extra_layer_size=config.extra_layer_size,
prototype_dropout=config.prototype_dropout,
flowstats_dropout=config.flowstats_dropout,
ppi_dropout=config.ppi_dropout,
prototype_hidden_size=config.prototype_hidden_size
)
if config.optimizer == 'adam':
logging.info("DST model train: Using Adam optimizer")
optimizer = optimizers.Adam(learning_rate=config.learning_rate)
elif config.optimizer == 'sgd':
logging.info("DST model train: Using SGD optimizer")
optimizer = optimizers.SGD(learning_rate=config.learning_rate)
else: # Default
logging.info("DST model train: Using Adam optimizer (DEFAULTED)")
optimizer = optimizers.Adam(learning_rate=config.learning_rate)
if config.loss == 'crossentropy':
logging.info("DST model train: Using Categorical Crossentropy loss")
loss = losses.CategoricalCrossentropy()
elif config.loss == 'focal':
logging.info("DST model train: Using Focal Crossentropy loss")
loss = losses.CategoricalFocalCrossentropy()
elif config.loss == 'f1':
logging.info("DST model train: Using F1 loss")
loss = architectures.f1_loss
else: # Default
logging.info("DST model train: Using Categorical Crossentropy loss (DEFAULTED)")
loss = losses.CategoricalCrossentropy()
logging.info("DST model train: Compiling model")
dst_model.compile(
optimizer=optimizer,
loss=loss,
metrics=['accuracy', metrics.F1Score(average="macro")]
)
logging.info("DST model train: Fitting model")
dst_model.fit(
(train_ppi, train_flow_stats),
utils.to_categorical(labels_encoded[train_indices]),
epochs=config.epochs,
batch_size=config.batch_size,
validation_data=((val_ppi, val_flow_stats), utils.to_categorical(labels_encoded[val_indices])),
shuffle=True,
verbose=1,
callbacks=[utilities.WandbCustomCallback(model_name="dst_model"), early_stopping],
class_weight=class_weights
)
try:
logging.info("DST model train: Saving model")
dst_model.save(f"{config.model_folder}/{experiment_nickname}.dst_model.keras")
except Exception as e:
logging.error(f"Error saving model: {e}")
# Testing pignistic transformation
logging.info("DST model test: Testing pignistic transformation")
dst_model_pignistic_pred = dst_model.predict((test_ppi, test_flow_stats), batch_size=config.batch_size)
utilities.wandb_classification_report(labels_encoded[test_indices], dst_model_pignistic_pred.argmax(axis=1), label_names, "dst_model_pignistic")
# Testing DST model
logging.info("DST model test: Testing DST model set-value inference")
combined_masses = dst_model.get_combined_mass((test_ppi, test_flow_stats))
result_table = []
for tol_i in [0,1,2,3,4]:
utility_matrix, act_set = utilities.generate_utility_matrix(num_class=class_count, tol_i=tol_i)
for nu in [0, 0.2, 0.4, 0.5, 0.6, 0.8, 1]:
dm = utility_layer_test.DM_test(num_class=class_count, num_set=len(act_set), nu=nu)
dm.set_weights(tf.reshape(utility_matrix, [1, len(act_set), class_count]))
decisions = dm(combined_masses)
dm_test_output_max = tf.argmax(decisions, axis=-1).numpy()
imprecise_results = [act_set[act] for act in dm_test_output_max]
result = utilities.set_value_evaluation(labels_encoded[test_indices], imprecise_results, dm_test_output_max, utility_matrix, act_set, labels_encoder.classes_)
result["tolerance"] = tol_i
result["nu"] = nu
result_table.append(result)
result_table = pd.DataFrame(result_table)
wandb.log({"Evaluation": wandb.Table(dataframe=result_table)})
run.finish()
if __name__ == "__main__":
main()