-
Notifications
You must be signed in to change notification settings - Fork 0
/
util.py
323 lines (277 loc) · 13.8 KB
/
util.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
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
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
# -*- coding: utf-8 -*-
"""
@author: F.Rosenthal, P.Schwarz
"""
import os
import matplotlib as mpl
import matplotlib.pyplot as plt
import tracemalloc
import numpy as np
import json
import pandas as pd
from itertools import product
from sklearn.metrics import ConfusionMatrixDisplay
from matplotlib.backends.backend_pdf import PdfPages
mpl.rcParams['mathtext.fontset'] = 'stix'
mpl.rcParams['font.family'] = 'STIXGeneral'
def tracking_start():
tracemalloc.stop()
# print("Tracing status: ", tracemalloc.is_tracing())
tracemalloc.start()
# print("Tracing status: ", tracemalloc.is_tracing())
def tracing_mem():
first_size, first_peak = tracemalloc.get_traced_memory()
peak = first_peak / (1024 * 1024)
print(f"Peak Size: {peak} MB")
def format_time(seconds):
minutes = seconds // 60
hours = int(minutes // 60)
seconds = int(seconds % 60)
minutes = int(minutes % 60)
return "{:02d}:{:02d}:{:02d}".format(hours, minutes, seconds)
def plot_spectrogram(t, f, Sxx):
"""
Plots spectrograms as output by scipy.
txf should be the dimensions of Sxx.
Parameters
----------
t : Time vector.
f : Frequency vector.
Sxx : Array containing the spectrogram levels.
Returns
-------
None.
"""
os.makedirs(os.path.join(os.getcwd(), "plots"), exist_ok=True)
fig = plt.figure()
plt.pcolormesh(t, f, Sxx, shading='gouraud')
plt.ylabel('Frequency [Hz]')
plt.xlabel('Time [sec]')
plt.show()
plt.close(fig)
def plot_graph(data, title, xlabel='Test Samples', ylabel='Error Rate', ylim=None, ground=False):
"""
Plots given data as line plot.
:param data: Array containing data points for plotting
:param title: Title of the plot
:param xlabel: Name of horiziontal axis
:param ylabel: Name of the vertical axis
:param ylim: Value bounds for displaying values
:param ground: If true, vertical axis will start at zero
"""
os.makedirs(os.path.join(os.getcwd(), "plots"), exist_ok=True)
n = np.arange(0, np.size(data))
fig = plt.figure(dpi=300)
plt.plot(n, data)
plt.xlabel(xlabel)
plt.ylabel(ylabel)
if ground:
plt.ylim(bottom=0)
if ylim is not None:
plt.ylim(ylim)
plt.title(title)
plt.savefig(f"plots/{title}.png")
plt.close(fig)
def plot_multiline_graph(data, title):
"""
Plots given data as multiple line diagrams.
:param data: 2D-Array containg the data to plot
:param title: Title of the diagram
"""
os.makedirs(os.path.join(os.getcwd(), "plots"), exist_ok=True)
keys = list(data.keys())
n = np.arange(0, np.size(data[keys[0]]))
fig = plt.figure(dpi=300)
for k in data.keys():
plt.plot(n, data[k], label=f"k={k} er={data[k][-1]}") # FR: got KeyError from data[k][-1]
plt.legend()
plt.xlabel("Test samples")
plt.ylabel("Error Rate")
plt.ylim([0, 1])
plt.title(title)
plt.savefig(f"plots/error_rates_{title}.png")
plt.close(fig)
def generate_random_plot_data(data):
"""
Generates random data points and fills the residual fields in the report data
:param data: report data
:return: data report, filled with random residual values
"""
import random
for id in data['results'].keys():
data['results'][id]['residuals'] = []
for i in range(10):
data['results'][id]['residuals'].append(random.randint(1, 100))
return data
def plot_confusion_matrix(data, savestring=""):
"""Wrapper function for sklearn's ConfusionMatrixDisplay.
The input dictionary will be used to plot a row-normalized confusion Matrix.
It will help exploring errors over classes. The matrix is saved as a pdf files.
Args:
data (dict): Results dictionary as generated by digit_classifer().
savestring (str, optional): Additional info to pass to the diagrams filename. Defaults to "".
"""
os.makedirs(os.path.join(os.getcwd(), "plots"), exist_ok=True)
y_true = []
y_pred = []
for k in data['results'].keys():
y_true.append(data['results'][k]['actual'])
y_pred.append(data['results'][k]['estimated'])
cm = ConfusionMatrixDisplay.from_predictions(y_true=y_true, y_pred=y_pred,
labels=list(range(10)),
normalize='true')
fig, ax = plt.subplots(dpi=300)
cm.plot(ax=ax, values_format='.2f')
ax.set_ylabel("Actual Digit")
ax.set_xlabel("Predicted Digit")
plt.savefig(f"plots/confusionmatrix_{savestring}.pdf", format="pdf", bbox_inches="tight")
plt.close(fig)
del ax
def plot_residuals(data, savestring=""):
os.makedirs(os.path.join(os.getcwd(), "plots"), exist_ok=True)
"""A function to make subplots of residuals over base digits for every actual digit class.
Plot inspired by Elden.
Resulting plot will be saved as a PNG-file.
Take this module's boxplot_residuals() to get statistically more useful plots.
Args:
data (dict): A dict where the classification results are stored.
savestring (str, optional): Additional string with metadata to pass to save-filename. Defaults to "".
"""
residuals_per_digit = {}
for i in range(10):
residuals_per_digit[str(i)] = []
for id in data['results'].keys():
k = str(data['results'][id]['actual'])
residuals_per_digit[k].append(list(data['results'][id]['residuals']))
fig, axs = plt.subplots(2, 5, figsize=(10, 5), dpi=300, sharey=True)
# plt.suptitle("Residuals vs. tested basis for every actual Digit.", fontsize=18, y=0.95)
for i, ax in enumerate(axs.flat):
for residual_list in residuals_per_digit[str(i)]:
ax.semilogy(list(range(10)), residual_list, 'k', alpha=0.04)
ax.set(xlabel='Tested Basis', ylabel='$\log_{2}(Residual)$')
ax.set_title(f"Digit Class {str(i)}")
ax.label_outer()
plt.setp(axs, xticks=list(range(10)), xticklabels=list(range(10)))
plt.tight_layout(rect=[0, 0.03, 0.9, 0.9])
plt.savefig(f"plots/residualplot_{savestring}.pdf", format="pdf", bbox_inches="tight")
plt.close(fig)
def boxplot_residuals(data, savestring=""):
"""A function to make subplots of residuals over base digits for every actual digit class.
Resulting plot will be saved as a PNG-file.
Args:
data (dict): A dict where the classification results are stored.
savestring (str, optional): Additional string with metadata to pass to save-filename. Defaults to "".
The first few lines are needed to transfor data from corresponding to the test sample
to corresponding to a hierarchy of actual digit classes instead.
This is what we are actually interested in here since we want to plot statistical information for every digit class.
The dict from json data is converted to a pandas DataFrame oriented with test samples as rows.
The dataframe has collected residuals per test as a list in every field of column 'residuals'. This has to be split.
We split the lists of residuals to new columns such that one column corresponds to a tested base.
df[list(range(10))] just creates 10 columns with digits as names.
We furthermore just index the dataframe efficiently such that we get columns as pandas series and can convert them to lists.
We want to plot the boxplots by iterating through a dict. Lets call it box_dict.
It's keys (digit_class) refer to the actual digit classes. Each value is a list of 10 arrays,
each array consisting of all residuals for one combination of digit class and tested base
We apply log2() on the array values to handle outliers.
The x-labels have to be constructed with
[""] + [str(i) for i in range(10)] + [""]
because the boxplots need the positions 0 and 11 as xticks.
"""
os.makedirs(os.path.join(os.getcwd(), "plots"), exist_ok=True)
df = pd.DataFrame.from_dict(data['results'], orient='index')
df[list(range(10))] = pd.DataFrame(df.residuals.tolist(), index=df.index)
box_dict = {digit_class: [] for digit_class in range(10)}
for (digit_class, test_base) in product(range(10), range(10)):
box_dict[digit_class].append(np.log2(df[test_base].loc[df['actual'] == digit_class].to_list()))
x_labels = [""] + [str(i) for i in range(10)] + [""]
fig, axs = plt.subplots(2, 5, figsize=(10, 5), dpi=300, sharey=True)
# plt.suptitle("Boxplot of residuals vs. tested basis for every actual digit.", fontsize=18, y=0.95)
for i, ax in enumerate(axs.flat):
ax.boxplot(box_dict[i], showfliers=False)
ax.set(xlabel='Tested Basis', ylabel='$\log_{2}(Residual)$')
ax.set_title(f"Digit Class {str(i)}")
ax.label_outer()
plt.setp(axs, xticks=list(range(12)), xticklabels=x_labels)
plt.tight_layout(rect=[0, 0.03, 0.9, 0.9])
# plt.grid(True)
plt.savefig(f"plots/residual_boxplots_{savestring}.pdf", format="pdf", bbox_inches="tight")
plt.close(fig)
del axs
def boxplot_residuals_multipage(data, savestring=""):
"""A function to make subplots of residuals over base digits for every actual digit class.
Resulting plot will be saved as a PNG-file.
Args:
data (dict): A dict where the classification results are stored.
savestring (str, optional): Additional string with metadata to pass to save-filename. Defaults to "".
The first few lines are needed to transfor data from corresponding to the test sample
to corresponding to a hierarchy of actual digit classes instead.
This is what we are actually interested in here since we want to plot statistical information for every digit class.
The dict from json data is converted to a pandas DataFrame oriented with test samples as rows.
The dataframe has collected residuals per test as a list in every field of column 'residuals'. This has to be split.
We split the lists of residuals to new columns such that one column corresponds to a tested base.
df[list(range(10))] just creates 10 columns with digits as names.
We furthermore just index the dataframe efficiently such that we get columns as pandas series and can convert them to lists.
We want to plot the boxplots by iterating through a dict. Lets call it box_dict.
It's keys (digit_class) refer to the actual digit classes. Each value is a list of 10 arrays,
each array consisting of all residuals for one combination of digit class and tested base
We apply log2() on the array values to handle outliers.
The x-labels have to be constructed with
[""] + [str(i) for i in range(10)] + [""]
because the boxplots need the positions 0 and 11 as xticks.
"""
os.makedirs(os.path.join(os.getcwd(), "plots"), exist_ok=True)
df = pd.DataFrame.from_dict(data['results'], orient='index')
df[list(range(10))] = pd.DataFrame(df.residuals.tolist(), index=df.index)
box_dict = {digit_class: [] for digit_class in range(10)}
for (digit_class, test_base) in product(range(10), range(10)):
box_dict[digit_class].append(np.log2(df[test_base].loc[df['actual'] == digit_class].to_list()))
x_labels = [""] + [str(i) for i in range(10)] + [""]
with PdfPages('multipage_pdf.pdf') as pdf:
fig, axs = plt.subplots(3, 1, figsize=(4.7, 7.5), dpi=300, sharey=True)
# plt.suptitle("Boxplot of residuals vs. tested basis for every actual digit.", fontsize=18, y=0.95)
for i, ax in zip(range(0, 3), axs.flat):
ax.boxplot(box_dict[i])
ax.set(xlabel='Tested Basis', ylabel='$\log_{10}(Residual)$')
ax.set_title(f"Actual Digit {str(i)}")
plt.setp(axs, xticks=list(range(12)), xticklabels=x_labels)
plt.tight_layout(rect=[0, 0.03, 1, 0.95])
pdf.savefig(fig) # f"plots/residual_boxplots_{savestring}.pdf", format="pdf", bbox_inches="tight")
plt.close(fig)
fig, axs = plt.subplots(3, 1, figsize=(4.7, 7.5), dpi=300, sharey=True)
# plt.suptitle("Boxplot of residuals vs. tested basis for every actual digit.", fontsize=18, y=0.95)
for i, ax in zip(range(3, 6), axs.flat):
ax.boxplot(box_dict[i])
ax.set(xlabel='Tested Basis', ylabel='$\log_{10}(Residual)$')
ax.set_title(f"Actual Digit {str(i)}")
plt.setp(axs, xticks=list(range(12)), xticklabels=x_labels)
plt.tight_layout(rect=[0, 0.03, 1, 0.95])
pdf.savefig(fig) # f"plots/residual_boxplots_{savestring}.pdf", format="pdf", bbox_inches="tight")
plt.close(fig)
fig, axs = plt.subplots(3, 1, figsize=(4.7, 7.5), dpi=300, sharey=True)
# plt.suptitle("Boxplot of residuals vs. tested basis for every actual digit.", fontsize=18, y=0.95)
for i, ax in zip(range(6, 9), axs.flat):
ax.boxplot(box_dict[i])
ax.set(xlabel='Tested Basis', ylabel='$\log_{10}(Residual)$')
ax.set_title(f"Actual Digit {str(i)}")
plt.setp(axs, xticks=list(range(12)), xticklabels=x_labels)
plt.tight_layout(rect=[0, 0.03, 1, 0.95])
pdf.savefig(fig) # f"plots/residual_boxplots_{savestring}.pdf", format="pdf", bbox_inches="tight")
plt.close(fig)
fig, ax = plt.subplots(1, 1, figsize=(4.7, 2.5), dpi=300, sharey=True)
# plt.suptitle("Boxplot of residuals vs. tested basis for every actual digit.", fontsize=18, y=0.95)
i = 9
ax.boxplot(box_dict[i])
ax.set(xlabel='Tested Basis', ylabel='$\log_{10}(Residual)$')
ax.set_title(f"Actual Digit {str(i)}")
plt.setp(axs, xticks=list(range(12)), xticklabels=x_labels)
plt.tight_layout(rect=[0, 0.03, 1, 0.95])
pdf.savefig(fig) # f"plots/residual_boxplots_{savestring}.pdf", format="pdf", bbox_inches="tight")
plt.close(fig)
if __name__ == '__main__':
path = "number_metrics_7500_1500.json"
with open(path) as json_file:
data = json.load(json_file)
plot_confusion_matrix(data)
plot_residuals(data)
boxplot_residuals(data)
print("Plots have been saved to 'plots/'.")