stats.py
5.96 KB
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
import argparse
import os
import core.data
import math
import numpy as np
import scipy.stats
import pickle
import matplotlib.pyplot as plt
import matplotlib.colors as mcolors
from utils import SubCommandRunner
from cycler import cycler
def pred_distribution(predictions: str, labels: str, labelencoder: str, outdir: str):
predictions = core.data.read_id_values(args.predictions, float)
labels = core.data.read_labels(args.labels)
le = None
with open(args.labelencoder, "rb") as f:
le = pickle.load(f)
stats = {}
print("PREDICTIONS ---------------------------")
for id_, predictions_ in predictions.items():
label = labels[id_][0]
if label not in stats:
stats[label] = {
"nb_utt": 1,
"predictions": np.expand_dims(predictions_, axis=0)
}
else:
stats[label]["nb_utt"] = stats[label]["nb_utt"] + 1
stats[label]["predictions"] = np.append(stats[label]["predictions"], np.expand_dims(predictions_, axis=0), axis=0)
print("CALCULATING ---------------------------")
colors = [
"darkorange",
"red",
"blue"
]
custom_cycler = (cycler(color=list(mcolors.TABLEAU_COLORS)) *
cycler(linestyle=['-', '--', '-.']))
for label, stats_ in stats.items():
plt.gca().set_prop_cycle(custom_cycler)
stats_mean = np.mean(stats_["predictions"], axis=0)
stats_std = np.std(stats_["predictions"], axis=0)
#print(label)
#print(stats_mean)
#print(stats_std)
kwargs = dict(alpha=0.5)
for i in range(stats_["predictions"].shape[1]):
label_str = le.inverse_transform([i])[0]
#plt.hist(stats_["predictions"][:, i], bins=10, label=label_str, **kwargs)
mu = stats_mean[i]
variance = stats_std[i] * stats_std[i]
sigma = stats_std[i]
# math.sqrt(variance)
print(f"{i}: mu {mu}, var {variance}, sigma {sigma}")
#x_values = np.arange(-1, 5, 0.1)
#y_values = scipy.stats.norm(mu, variance)
#y = scipy.stats.norm.pdf(x,mean,std)
#plt.plot(x_values, y_values.pdf(x_values,))
#x, step = np.linspace(mu - 3*sigma, mu + 3*sigma, 1000, retstep=True)
x = np.linspace(0, 1, 1000)
#x = np.linspace(mu - 3*sigma, mu + 3*sigma, 1000)
#x, step = np.linspace(0, 1, 1000, retstep=True)
P = scipy.stats.norm.cdf(x, mu, sigma)
#print(step)
plt.plot(x, P, label=label_str, **kwargs)
#plt.savefig("simple_gaussian.pdf")
plt.legend()
plt.savefig(os.path.join(args.outdir, f"{label}_prediction_cdf.pdf"))
plt.clf()
print("Decisions")
def pred_distribution_wt_sel(predictions: str, labels: str, labelencoder: str, outdir: str):
keys_preds, matrix_preds = core.data.read_features_with_matrix(predictions)
n = 3
print(matrix_preds.shape)
for j in range(matrix_preds.shape[1]):
indices = (-matrix_preds[:, j]).argsort()[:n]
print(f"INDICE: {j}")
print("indices")
print(indices)
print("Best values")
print(matrix_preds[indices, j])
print("All dimensions of best values")
print(matrix_preds[indices])
# Select the n best for each column
pass
def utt2dur(utt2dur: str, labels: str):
if labels == None:
pass
else:
pass
durations = []
with open(utt2dur, "r") as f:
for line in f:
splited = line.replace("\n", "").split(" ")
durations.append(float(splited[1]))
durations = np.asarray(durations, dtype=float)
print(durations.shape)
mean = np.mean(durations)
std = np.std(durations)
print(f"mean: {mean}")
print(f"std: {std}")
if __name__ == "__main__":
# Parser
parser = argparse.ArgumentParser(description="Statistics")
subparsers = parser.add_subparsers(title="actions")
# pred-distribution
parser_pred_dist = subparsers.add_parser("pred-distribution", help="plot distributions of prediction through labels")
parser_pred_dist.add_argument("--predictions", type=str, help="prediction file", required=True)
parser_pred_dist.add_argument("--labels", type=str, help="label file", required=True)
parser_pred_dist.add_argument("--labelencoder", type=str, help="label encode pickle file", required=True)
parser_pred_dist.add_argument("--outdir", type=str, help="output file", required=True)
parser_pred_dist.set_defaults(which="pred_distribution")
# pred-distribution-with-selection
parser_pred_dist_wt_sel = subparsers.add_parser("pred-distribution-with-selection", help="plot distributions of prediction through labels with a selection of the n best records by column/class prediction.")
parser_pred_dist_wt_sel.add_argument("--predictions", type=str, help="prediction file", required=True)
parser_pred_dist_wt_sel.add_argument("--labels", type=str, help="label file", required=True)
parser_pred_dist_wt_sel.add_argument("--labelencoder", type=str, help="label encode pickle file", required=True)
parser_pred_dist_wt_sel.add_argument("--outdir", type=str, help="output file", required=True)
parser_pred_dist_wt_sel.set_defaults(which="pred_distribution_with_selection")
# duration-stats
parser_utt2dur = subparsers.add_parser("utt2dur", help="distribution of utt2dur")
parser_utt2dur.add_argument("--utt2dur", type=str, help="utt2dur file", required=True)
parser_utt2dur.add_argument("--labels", type=str, default=None, help="labels file")
parser_utt2dur.set_defaults(which="utt2dur")
# Parse
args = parser.parse_args()
# Run commands
runner = SubCommandRunner({
"pred_distribution": pred_distribution,
"pred_distribution_with_selection": pred_distribution_wt_sel,
"utt2dur": utt2dur
})
runner.run(args.which, args.__dict__, remove="which")