-
Notifications
You must be signed in to change notification settings - Fork 7
/
replicate_ensemble_results.py
160 lines (134 loc) · 6.49 KB
/
replicate_ensemble_results.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
# This script easily replicates ensemble results from the paper
import os
import sys
import numpy as np
import torch
import torch.nn.functional as F
from sklearn import metrics
from utils.utils import setup_logging
# These predictions are included in TOP-N ensemble
# it was found via method find_best_ensemble_greedy from ensembling/ensemble_helper.py
TOP_N_ensemble = [
"result_F1_0.57948_L_0.6698856112670224_2019-01-28_08:24_pcknot5.npy",
"result_F1_0.57759_L_0.703442574330578_2019-01-28_00:15_pcbirger.npy",
"result_F1_0.57623_L_0.6621931040825227_2019-01-28_00:32_pcknot5.npy",
"result_F1_0.57526_L_0.6638631148319039_2019-01-27_08:12_pcknot4.npy",
"result_F1_0.57423_L_0.7102468566180802_2019-01-28_17:03_pcknot5.npy",
"result_F1_0.57371_L_0.6669414722463592_2019-01-27_00:46_pcknot5.npy",
"result_F1_0.56750_L_0.6898565446440823_2019-01-26_20:31_pcknot4.npy",
"result_F1_0.56656_L_0.699664715034862_2019-01-27_15:57_pcbirger.npy",
"result_F1_0.56460_L_0.724339671515812_2019-01-28_15:53_pcbirger.npy",
"result_F1_0.56433_L_0.663498227135592_2019-01-28_13:27_pcknot2.npy",
"result_F1_0.56313_L_0.689033422880176_2019-01-26_20:39_pcknot4.npy",
"result_F1_0.56069_L_0.670826427442727_2019-01-27_02:10_pcknot4.npy",
"result_F1_0.55930_L_0.6865916204641289_2019-01-27_16:14_pcbirger.npy",
"result_F1_0.55580_L_0.7056901221467318_2019-01-26_20:24_pcknot4.npy",
"result_F1_0.55509_L_0.7102856230281916_2019-01-28_00:06_pcbirger.npy",
"result_F1_0.55504_L_0.6975949840002625_2019-01-27_23:51_pcbirger.npy",
"result_F1_0.55092_L_0.6955123813847969_2019-01-28_12:34_pcknot4.npy"
]
# These predictions are not included in EXC-N ensemble
# it was found via method remove_worst_k from ensembling/ensemble_helper.py
EXC_N_ensemble = [
"result_F1_0.56266_L_0.749426320401043",
"result_F1_0.55143_L_0.712958636656001",
"result_F1_0.55581_L_0.695809597395501",
"result_F1_0.56191_L_0.661831214200615",
"result_F1_0.57948_L_0.669885611267022",
"result_F1_0.57292_L_0.705603626619452"
]
def evaluate(data="test", ensemble_type="TOP_N", strategy="avg_softmaxes"):
path = f"predictions/numpy_final_all_{'VAL' if data == 'validation' else 'TEST'}"
files = sorted(os.listdir(path))
prefix = "val_" if data == 'validation' else "test_"
valid = [f for f in files if f.startswith(prefix) and f.endswith("npy")]
if ensemble_type == "TOP_N":
valid_ensemble_subset = [f"{prefix}{s}" for s in list(TOP_N_ensemble)]
result_files = [f for f in valid if "result" in f and f in valid_ensemble_subset]
else: # EXC_N
def has_worst_substr(f):
for w in EXC_N_ensemble:
if w in f: return True
return False
result_files = [f for f in valid if "result" in f and not has_worst_substr(f)]
print("Ensemble is build from following files:")
print(result_files)
print(f"{len(result_files)} files total")
label_file = [f for f in valid if "labels" in f][0]
labels = np.load(os.path.join(path, label_file))
result_matrices = [np.load(os.path.join(path, result_file)) for result_file in
result_files]
results = np.array(result_matrices)
tweet_ids = open(f"saved/ensemble/numpy_result/{prefix}ids.txt", "r").read().split()
if strategy == "average_logits":
feats = np.average(results, 0)
results = torch.Tensor(feats)
elif strategy == "sum_softmaxes": # summing has same effect as averaging
results = torch.Tensor(results)
# Models x batch x classes
results = F.softmax(results, -1)
results = torch.sum(results, 0)
elif strategy == "weighted_softmax_sum":
results = torch.Tensor(results)
# Models x batch x classes
results = F.softmax(results, -1)
for k in range(results.shape[0]): results[k] = results[k] * weights[k]
results = torch.sum(results, 0)
elif strategy == "avg_softmaxes":
# Models x batch x classes
results = torch.Tensor(results)
results = F.softmax(results, -1)
results = torch.mean(results, 0)
else:
return
labels = torch.Tensor(labels).long()
softmaxed_results = results if strategy == "avg_softmaxes" else F.softmax(results, -1)
maxpreds, argmaxpreds = torch.max(softmaxed_results, dim=1)
total_preds = list(argmaxpreds.cpu().numpy())
total_labels = list(labels.cpu().numpy())
correct_vec = argmaxpreds == labels
total_correct = torch.sum(correct_vec).item()
acc = total_correct / results.shape[0]
F1 = metrics.f1_score(total_labels, total_preds, average="macro")
F1_cls = metrics.f1_score(total_labels, total_preds, average=None)
return acc, F1, tuple(F1_cls)
def print_results(acc, F1, per_class_F1):
print(f"Acc: {acc}\n F1: {F1}\n C1_F1: "
f"{per_class_F1[0]}\n C2_F1: {per_class_F1[1]}\n C3_F1: {per_class_F1[2]}\n C4_F1: {per_class_F1[3]}")
if __name__ == "__main__":
setup_logging(os.path.basename(sys.argv[0]).split(".")[0],
logpath="logs/",
config_path="configurations/logging.yml")
# EXC_N
print("-"*50)
data, ensemble_type, strategy = "validation", "EXC_N", "avg_softmaxes"
print(f"data: {data}, ensemble type: {ensemble_type}, strategy type: {strategy}")
results = evaluate(data, ensemble_type, strategy)
print_results(*results)
print("-"*50)
data = "test"
print(f"data: {data}, ensemble type: {ensemble_type}, strategy type: {strategy}")
results = evaluate(data, ensemble_type, strategy)
print_results(*results)
# TOP_N
print("-"*50)
data, ensemble_type, strategy = "validation", "TOP_N", "avg_softmaxes"
print(f"data: {data}, ensemble type: {ensemble_type}, strategy type: {strategy}")
results = evaluate(data, ensemble_type, strategy)
print_results(*results)
print("-"*50)
data = "test"
print(f"data: {data}, ensemble type: {ensemble_type}, strategy type: {strategy}")
results = evaluate(data, ensemble_type, strategy)
print_results(*results)
# TOP_N_S (BEST RESULTS ON TEST DATA)
print("-"*50)
data, ensemble_type, strategy = "validation", "TOP_N", "average_logits"
print(f"data: {data}, ensemble type: {ensemble_type}, strategy type: {strategy}")
results = evaluate(data, ensemble_type, strategy)
print_results(*results)
print("-"*50)
data = "test"
print(f"data: {data}, ensemble type: {ensemble_type}, strategy type: {strategy}")
results = evaluate(data, ensemble_type, strategy)
print_results(*results)