-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathensemble.py
155 lines (121 loc) · 6.06 KB
/
ensemble.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
import os
import argparse
import glob
import tensorflow as tf
import numpy as np
import pandas as pd
from sklearn.metrics import confusion_matrix, classification_report, f1_score
import utils
import models
from dataloader import DataLoader
from train import *
from settings import PROJECT_ROOT
from params import param_batch as param
log = utils.get_logger("Ensemble", None)
def ensemble(args):
"""Validate Ensemble model with full validation set.... (Hmm)
"""
max_models = utils.get_arg(args, "max_models")
criterion = utils.get_arg(args, "criterion")
method = utils.get_arg(args, "method")
filename = utils.get_arg(args, "filename")
log.info("Starting ensemble of {} models with existing ckpt from {}".format(max_models, filename))
log.info("Criterion: {}, Method: {}".format(criterion, method))
pass_empty = 0
valid_dataloader = DataLoader(total_validation=True)
for i in range(max_models):
param_dict, pass_empty = get_best_hyperparams(filename, criterion, i, pass_empty)
valid_dataloader.reset_args(param_dict)
tf.reset_default_graph()
model = utils.find_class_by_name([models], param_dict['model'])(param_dict)
model.build_graph(is_training=tf.constant(False, dtype=tf.bool))
session = tf.Session(config=tf.ConfigProto(
gpu_options=tf.GPUOptions(allow_growth=True),
log_device_placement=False,
allow_soft_placement=True)
)
restore_session(session, param_dict)
_, _, f1 , y_logit, y_true = valid_full(0, model, valid_dataloader, session, param_dict)
session.close()
assert y_logit.shape == (3, 87*200)
assert y_true.shape == (3, 87*200)
y_pred = np.greater(y_logit, 0).astype(int)
print("(Full validation) F1 by jogyo.... : {}".format(utils.calculate_average_F1_score(y_pred.tolist(), y_true.tolist())))
print("(Full validation) F1 by sklearn : {}".format(f1_score(list(y_true.reshape([-1])), list(y_pred.reshape([-1])))))
print("(Part validation) table valid_f1: {}".format(param_dict['valid_f1']))
y_logit = y_logit.reshape([1, -1])
y_true = y_true.reshape([1, -1])
if i == 0:
y_logits = y_logit
y_trues = y_true.astype(int)
else:
y_logits = np.concatenate((y_logits, y_logit), axis=0)
assert np.equal(y_trues, y_true).sum() == np.prod(y_true.shape), "Must return same true values"
if method == "average":
y_logits = np.mean(y_logits, axis=0)
y_preds = np.greater(y_logits, 0).astype(int)
y_preds.reshape([1, -1])
elif method == "vote":
y_logits = np.greater(y_logits, 0).astype(int)
one_counts = np.count_nonzero(y_logits, axis=0)
zero_counts = max_models - one_counts
# Warning: Predicts Tie as 1 if (max_models % 2 == 0)
y_preds = np.greater_equal(one_counts, zero_counts).astype(int)
y_preds = y_preds.reshape([1, -1])
else:
raise ValueError("Invalid criterion type: {}".format(criterion))
y_preds_list = y_preds.reshape([-1]).tolist()
y_trues_list = y_trues.reshape([-1]).tolist()
print("#############ENSEMBLE RESULT##############")
for line in classification_report(y_trues_list, y_preds_list).split("\n"):
print(line)
print(confusion_matrix(y_trues_list, y_preds_list))
print("(Full validation) (jogyo) F1_score: {}".format(utils.calculate_average_F1_score(y_preds.tolist(), y_trues.tolist())))
print("(Full validation) (sklearn) F1_score: {}".format(f1_score(y_trues_list, y_preds_list)))
def get_best_hyperparams(filename, criterion, i, pass_empty):
"""Returns hyperparameters of ith best result in result.txt where ckpt exists
Args:
criterion (str): valid_f1 or valid_acc
i (int): ith best hyperparam to retrieve
Returns:
best_params (dict)
"""
df_result = pd.read_csv(os.path.join(PROJECT_ROOT, filename), header=0, delimiter="\t")
df_result = df_result[df_result.no_save_ckpt == False].sort_values(by=criterion, ascending=False)
while True:
best_params = df_result.iloc[i+pass_empty].to_dict()
ckpt_path = os.path.join(PROJECT_ROOT, best_params['train_dir'], best_params['tag_label'], best_params['unique_key'], "*{}*".format(best_params['unique_key']))
ckpt_list = glob.glob(ckpt_path)
if len(ckpt_list) >= 3:
break
else:
log.warning("Result with unique key({}) doesn't have ckpt files in the directory: {}. "
"It has been trained to save ckpt. Proceeding to next best model...".format(best_params['unique_key'], ckpt_list[0]))
pass_empty += 1
return best_params, pass_empty
def restore_session(session, param_dict):
saver = tf.train.Saver(tf.global_variables(), max_to_keep=1)
session.run(tf.global_variables_initializer())
session.run(tf.local_variables_initializer())
ckpt_path = os.path.join(PROJECT_ROOT, param_dict['train_dir'], param_dict['tag_label'], param_dict['unique_key']) #, "*{}*".format(param_dict['unique_key']))
#ckpt_list = glob.glob(ckpt_path)
#data_path = [ckpt for ckpt in ckpt_list if '.data' in ckpt][0]
try:
if os.path.isdir(ckpt_path):
old_checkpoint_path = ckpt_path
ckpt_path = tf.train.latest_checkpoint(ckpt_path)
log.info("Update checkpoint_path: {} -> {}".format(
old_checkpoint_path, ckpt_path)
)
log.info("Restoring from {}".format(ckpt_path))
saver.restore(session, ckpt_path)
except:
raise Exception("Something is wrong with ckpt.")
if __name__ == "__main__":
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--max_models", default=3, type=int)
parser.add_argument("--criterion", default="valid_f1", type=str, choices=["valid_f1", "valid_acc"])
parser.add_argument("--method", default="vote", type=str, choices=["average", "vote"])
parser.add_argument("--filename", default="log/batch_log/result_experiment.txt", type=str)
args = parser.parse_args()
ensemble(args)