-
Notifications
You must be signed in to change notification settings - Fork 0
/
validate.py
192 lines (156 loc) · 5.77 KB
/
validate.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
import argparse
import glob
from constants import *
import librosa
import time
import sys
import pyaudio
import numpy as np
import librosa
from tensorflow import keras
from extract import extract_features
import os
import matplotlib.pyplot as plt
class EvaluationResult:
def __init__(self, total_time, confidences):
self.total_time = total_time
self.confidences = confidences
class TimeRange:
def __init__(self, start_time, end_time):
start_segments = start_time.strip().split(":")
end_segments = end_time.strip().split(":")
assert len(start_segments) == 2
assert len(end_segments) == 2
self.start_seconds = int(start_segments[0]) * 60 + int(start_segments[1])
self.end_seconds = int(end_segments[0]) * 60 + int(end_segments[1])
def run_model(model_path, audio_data, sample_rate, confidence_limit):
model = keras.models.load_model(model_path)
len_per_sample = sample_rate * SAMPLE_LENGTH_SEC
showering_confidences = []
total_showering_time = 0
for i in range(0, len(audio_data), len_per_sample):
sample = audio_data[i : i + len_per_sample]
if len(sample) != len_per_sample:
continue
features = extract_features(sample, sample_rate).reshape(
-1, N_MELS, FEATURE_COUNT, 1
)
network_output = model.predict(
features, verbose=0, use_multiprocessing=True, batch_size=150
)
class_confidences = np.mean(network_output, axis=0)
showering_confidences.append(class_confidences[0])
duration = SAMPLE_LENGTH_SEC
if class_confidences[0] > confidence_limit:
total_showering_time = total_showering_time + SAMPLE_LENGTH_SEC
return EvaluationResult(total_showering_time, showering_confidences)
def calculate_real_duration(time_ranges):
duration = 0
for trange in time_ranges:
duration = duration + (trange.end_seconds - trange.start_seconds)
return duration
def plot_evaluation_results(
eval_results, time_ranges, confidence_limit, binarize=False
):
if len(eval_results) == 0:
return
x = np.arange(
0,
len(eval_results[list(eval_results.keys())[0]].confidences) * SAMPLE_LENGTH_SEC,
SAMPLE_LENGTH_SEC,
)
fig, (conf_plot, time_plot) = plt.subplots(2)
for key, result in eval_results.items():
y = result.confidences
if binarize:
y = [1.0 if e >= confidence_limit else 0.0 for e in y]
conf_plot.plot(x, y, label=f"{key}: {result.total_time}")
# Confidence limit line
conf_plot.axhline(y=confidence_limit, color="r", linestyle="-")
# Time range of actual showering
colors = ["g", "b", "y"]
for count, time_range in enumerate(time_ranges):
conf_plot.axvline(
x=time_range.start_seconds,
color=colors[count % len(colors)],
linestyle="dashed",
)
conf_plot.axvline(
x=time_range.end_seconds,
color=colors[count % len(colors)],
linestyle="dashed",
)
conf_plot.set_title("Confidence Test Audio")
conf_plot.set_xlabel("time [s]")
conf_plot.set_ylabel("confidence [%]")
leg = conf_plot.legend(loc="upper right")
lined = {} # Will map legend lines to original lines.
for legline, origline in zip(leg.get_lines(), conf_plot.get_lines()):
legline.set_picker(True) # Enable picking on the legend line.
lined[legline] = origline
def on_pick(event):
# On the pick event, find the original line corresponding to the legend
# proxy line, and toggle its visibility.
legline = event.artist
origline = lined[legline]
visible = not origline.get_visible()
origline.set_visible(visible)
# Change the alpha on the line in the legend so we can see what lines
# have been toggled.
legline.set_alpha(1.0 if visible else 0.2)
fig.canvas.draw()
fig.canvas.mpl_connect('pick_event', on_pick)
# time plot
names = []
durations = []
for key, result in eval_results.items():
durations.append(result.total_time)
names.append(key)
time_plot.bar(names, durations)
if len(time_ranges) > 0:
time_plot.axhline(
y=calculate_real_duration(time_ranges), color="r", linestyle="-"
)
plt.show()
if __name__ == "__main__":
parser = argparse.ArgumentParser(
prog="Validator for dolphin model",
description="Pits different models against a sample and checks the accuracy",
)
parser.add_argument(
"-t",
"--timestamps",
type=str,
help='Timestamp of the showering segments in the format of "00:00-01:20,01:30-01:54"',
)
parser.add_argument(
"-l",
"--limit",
type=float,
default=0.5,
help="Minimum confidence value to count as showering",
)
parser.add_argument("-b", "--binarize", action="store_true")
parser.add_argument("songfile")
parser.add_argument(
"model",
nargs="+",
type=str,
help="Path to a single model to evaluate accurancy for",
)
args = parser.parse_args()
if len(args.model) == 0:
print("Please provide a model")
sys.exit(1)
time_ranges = []
if args.timestamps:
for segment in args.timestamps.strip().split(","):
times = segment.strip().split("-")
time_ranges.append(TimeRange(times[0], times[1]))
audio_data, sample_rate = librosa.load(args.songfile, sr=RATE)
evaluation_results = dict()
for model in args.model:
evaluation_results[os.path.basename(model)] = run_model(
model, audio_data, sample_rate, args.limit
)
plot_evaluation_results(evaluation_results, time_ranges, args.limit, args.binarize)