forked from mlcommons/GaNDLF
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgandlf_collectStats
275 lines (220 loc) · 10.2 KB
/
gandlf_collectStats
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
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
#!usr/bin/env python
# -*- coding: utf-8 -*-
import os
import sys
import argparse
import subprocess
import pathlib
import ast
from pathlib import Path
from datetime import date
import numpy as np
import pandas as pd
from io import StringIO
import seaborn as sns
import matplotlib.pyplot as plt
def main():
copyrightMessage = (
"Contact: [email protected]\n\n"
+ "This program is NOT FDA/CE approved and NOT intended for clinical use.\nCopyright (c) "
+ str(date.today().year)
+ " University of Pennsylvania. All rights reserved."
)
parser = argparse.ArgumentParser(
prog="GANDLF_CollectStats",
formatter_class=argparse.RawTextHelpFormatter,
description="Collect statistics from different testing/validation combinations from output directory.\n\n"
+ copyrightMessage,
)
parser.add_argument(
"-m",
"--modeldir",
metavar="",
type=str,
help="Input directory which contains testing and validation models",
)
parser.add_argument(
"-o",
"--outputdir",
metavar="",
type=str,
help="Output directory to save stats and plot",
)
parser.add_argument(
"-c",
"--combinedplots",
metavar="",
default=False,
type=ast.literal_eval,
help="Overlays training and validation plots for both accuracy and loss (classification only).",
)
args = parser.parse_args()
inputDir = os.path.normpath(args.modeldir)
outputDir = os.path.normpath(args.outputdir)
Path(outputDir).mkdir(parents=True, exist_ok=True)
outputFile = os.path.join(outputDir, "data.csv") # data file name
outputPlot = os.path.join(outputDir, "plot.png") # plot file
combinedPlots = args.combinedplots
trainingLogs = os.path.join(inputDir, "logs_training.csv")
validationLogs = os.path.join(inputDir, "logs_validation.csv")
testingLogs = os.path.join(inputDir, "logs_testing.csv")
if os.path.exists(testingLogs):
testingLogsCSV = pd.read_csv(testingLogs)
if len(testingLogsCSV) == 0: # check for classification task
print("Classification task detected, generating accuracy and loss plots.")
if combinedPlots: # check whether user wants training + validation overlayed plots
df_training = pd.read_csv(trainingLogs)
df_validation = pd.read_csv(validationLogs)
epochs = len(df_training)
fig, axes = plt.subplots(nrows=1, ncols=2) # set plot properties
plt.subplots_adjust(
wspace=0.5, hspace=0.5
) # ensure spacing between plots
splot = sns.lineplot(
data=df_training,
x="epoch_no",
y="train_balanced_accuracy",
ax=axes[0],
) # plot training accuracy data
splot = sns.lineplot(
data=df_validation,
x="epoch_no",
y="valid_balanced_accuracy",
ax=axes[0],
) # plot validation accuracy data
splot.set(
xlim=(0, epochs - 1)
) # set limits for x-axis for proper visualization
splot.set(ylim=(0, 1)) # set limits for y-axis for proper visualization
splot.set(
xlabel="Epoch", ylabel="Accuracy", title="Accuracy Plot"
) # add labels and title to plot
axes[0].legend(labels=["Training", "Validation"]) # add legend to plot
splot = sns.lineplot(
data=df_training, x="epoch_no", y="train_loss", ax=axes[1]
) # plot training loss data
splot = sns.lineplot(
data=df_validation, x="epoch_no", y="valid_loss", ax=axes[1]
) # plot validation loss data
splot.set(
xlim=(0, epochs - 1)
) # set limits for x-axis for proper visualization
splot.set(
xlabel="Epoch", ylabel="Loss", title="Loss Plot"
) # add labels and title to plot
axes[1].legend(labels=["Training", "Validation"]) # add legend to plot
plt.savefig(outputPlot, dpi=600) # save plot
print("Plots saved successfully.")
else:
df_training = pd.read_csv(trainingLogs)
df_validation = pd.read_csv(validationLogs)
epochs = len(df_training)
fig, axes = plt.subplots(nrows=2, ncols=2) # set plot properties
plt.subplots_adjust(wspace=0.5, hspace=0.5)
splot = sns.lineplot(
data=df_training,
x="epoch_no",
y="train_balanced_accuracy",
ax=axes[0, 0],
) # plot the data
splot.set(xlim=(0, epochs - 1))
splot.set(ylim=(0, 1)) # set limits for y-axis for proper visualization
splot.set(
xlabel="Epoch", ylabel="Accuracy", title="Training Accuracy Plot"
) # set labels
splot = sns.lineplot(
data=df_validation,
x="epoch_no",
y="valid_balanced_accuracy",
ax=axes[0, 1],
) # plot the data
splot.set(xlim=(0, epochs - 1))
splot.set(ylim=(0, 1)) # set limits for y-axis for proper visualization
splot.set(
xlabel="Epoch", ylabel="Accuracy", title="Validation Accuracy Plot"
) # set labels
splot = sns.lineplot(
data=df_training, x="epoch_no", y="train_loss", ax=axes[1, 0]
) # plot the data
splot.set(xlim=(0, epochs - 1))
splot.set(
xlabel="Epoch", ylabel="Loss", title="Training Loss Plot"
) # set labels
splot = sns.lineplot(
data=df_validation, x="epoch_no", y="valid_loss", ax=axes[1, 1]
) # plot the data
splot.set(xlim=(0, epochs - 1))
splot.set(
xlabel="Epoch", ylabel="Loss", title="Validation Loss Plot"
) # set labels
plt.savefig(outputPlot, dpi=600)
print("Plots saved successfully.")
else:
print("Segmentation task detected, generating dice and loss plots.")
final_stats = "Epoch,Train_Loss,Train_Dice,Val_Loss,Val_Dice,Testing_Loss,Testing_Dice\n" # the columns that need to be present in final output; epoch is always removed
# loop through output directory
for dirs in os.listdir(inputDir):
currentTestingDir = os.path.join(inputDir, dirs)
if os.path.isdir(currentTestingDir): # go in only if it is a directory
if "testing_" in dirs: # ensure it is part of the testing structure
for val in os.listdir(
currentTestingDir
): # loop through all validation directories
currentValidationDir = os.path.join(currentTestingDir, val)
if os.path.isdir(currentValidationDir):
filesInDir = os.listdir(
currentValidationDir
) # get all files in each directory
for i, n in enumerate(filesInDir):
# when the log has been found, collect the final numbers
if "trainingScores_log" in n:
log_file = os.path.join(currentValidationDir, n)
with open(log_file) as f:
for line in f:
pass
final_stats = final_stats + line
data_string = StringIO(final_stats)
data_full = pd.read_csv(data_string, sep=",")
del data_full["Epoch"] # no need for epoch
data_full.to_csv(outputFile, index=False) # save updated data
# perform deep copy
data_loss = data_full.copy()
data_dice = data_full.copy()
cols = [
"Train",
"Val",
"Testing",
] # set the datasets that need to be plotted
for i in cols:
del data_dice[i + "_Loss"] # keep only dice
del data_loss[i + "_Dice"] # keep only loss
data_loss.rename(
columns={i + "_Loss": i}, inplace=True
) # rename the columns
data_dice.rename(
columns={i + "_Dice": i}, inplace=True
) # rename the columns
fig, axes = plt.subplots(
nrows=1, ncols=2, constrained_layout=True
) # set plot properties
bplot = sns.boxplot(
data=data_dice, width=0.5, palette="colorblind", ax=axes[0]
) # plot the data
bplot.set(ylim=(0, 1)) # set limits for y-axis for proper visualization
bplot.set(xlabel="Dataset", ylabel="Dice", title="Dice plot") # set labels
bplot.set_xticklabels(
bplot.get_xticklabels(), rotation=15, ha="right"
) # rotate so that everything is visible
bplot = sns.boxplot(
data=data_loss, width=0.5, palette="colorblind", ax=axes[1]
) # plot the data
bplot.set(ylim=(0, 1)) # set limits for y-axis for proper visualization
bplot.set(xlabel="Dataset", ylabel="Loss", title="Loss plot") # set labels
bplot.set_xticklabels(
bplot.get_xticklabels(), rotation=15, ha="right"
) # rotate so that everything is visible
plt.savefig(outputPlot, dpi=600)
print("Plots saved successfully.")
# main function
if __name__ == "__main__":
main()