-
Notifications
You must be signed in to change notification settings - Fork 2
/
ModelClassificationBase.py
379 lines (276 loc) · 14.7 KB
/
ModelClassificationBase.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
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
# -*- coding: utf-8 -*-
"""
Created on Wed Oct 12 17:23:05 2022
@author: Simon Bilik
This class is used for the classification of the evaluated model
"""
import os
import pickle
import yaml
import time
import logging
import numpy as np
import matplotlib.pyplot as plt
from sklearn import svm
from sklearn.manifold import TSNE
from sklearn.decomposition import PCA
from sklearn.ensemble import IsolationForest
from sklearn.preprocessing import RobustScaler
from sklearn.covariance import EllipticEnvelope
from sklearn.neighbors import LocalOutlierFactor
from sklearn.metrics import precision_recall_curve, auc, roc_auc_score, roc_curve, confusion_matrix, ConfusionMatrixDisplay, precision_score, recall_score, f1_score
class ModelClassificationBase():
## Set the constants and paths
def __init__(
self,
modelDataPath,
experimentPath,
modelSel,
layerSel,
labelInfo,
imageDim,
modelData,
featExtName,
anomaly_algorithm_selection = ["Robust covariance", "One-Class SVM", "Isolation Forest", "Local Outlier Factor"],
visualize = True
):
# Set the base path
self.layerSel = layerSel
self.modelName = modelSel
self.modelDataPath = modelDataPath
self.experimentPath = experimentPath
self.fitPath = os.path.join(*os.path.split(modelDataPath)[:-1])
# Set the constants
self.imageDim = imageDim
self.labelInfo = labelInfo
# Set the feature extractor name and print separator
self.featExtName = featExtName
# Save the modelData variable
self.modelData = modelData
#save anomaly algorithm selection
self.anomaly_algorithm_selection = anomaly_algorithm_selection
self.visualize = visualize
self.predictedLabels = []
logging.info('Feature extraction method: ' + self.featExtName)
logging.info('------------------------------------------------------------------------------------------------')
## Get data from file
def procDataFromFile(self):
actStrs = ['Train', 'Test', 'Valid']
for actStr in actStrs:
self.actStr = actStr
# Build the paths
orgDatasetPath = os.path.join(self.experimentPath, 'Org_' + actStr + '.npz')
procDatasetPath = os.path.join(self.modelDataPath, 'Eval_' + actStr + '.npz')
# Load the NPZ files
orgDataset = np.load(orgDatasetPath)
procDataset = np.load(procDatasetPath)
if self.actStr == 'Train':
# Store the data and get the metrics
self.processedData = {'Org': orgDataset['orgData'], 'Dec': procDataset['decData'], 'Lab': orgDataset['labels']}
self.metricsTr, _ = self.computeMetrics(self.processedData)
elif self.actStr == 'Test':
# Store the data and get the metrics
self.processedData = {'Org': orgDataset['orgData'], 'Dec': procDataset['decData'], 'Lab': orgDataset['labels']}
self.metricsTs, self.labelsTs = self.computeMetrics(self.processedData)
## Get data from dictionary
def procDataFromDict(self):
# actStrs = ['Train', 'Test', 'Valid']
for actStr, data in self.modelData.items():
self.actStr = actStr
self.processedData = data
if self.actStr == 'Train':
# Store the data and get the metrics
self.metricsTr, _ = self.computeMetrics(self.processedData)
elif self.actStr == 'Test':
# Store the data and get the metrics
self.metricsTs, self.labelsTs = self.computeMetrics(self.processedData)
# Visualise the data
if self.visualize:
self.fsVisualise(self.metricsTs, self.labelsTs)
elif self.actStr == 'Valid':
# Store the data and get the metrics
self.metricsVl, self.labelsVl = self.computeMetrics(self.processedData)
elif self.actStr == 'Predict':
# Store the data and get the metrics
self.metricsPr, _ = self.computeMetrics(self.processedData)
## Compute the classification metrics
def computeMetrics(self, processedData):
pass
## Normalize data (Gaussian)
def normalize2DData(self, data):
transformer = RobustScaler().fit(data)
dataSc = transformer.transform(data)
#return (data - np.min(data)) / (np.max(data) - np.min(data))
return dataSc
## Calculate classification metrics
def getEvaluationMetrics(self, scores, name, ax = None):
# Get the ROC curve
precision, recall, _ = precision_recall_curve(self.labelsTs, scores)
prc_auc = auc(recall, precision)
roc_auc = roc_auc_score(self.labelsTs, scores)
fpr, tpr, trs = roc_curve(self.labelsTs, scores)
# Log the information
logging.info("Algorithm: " + name)
logging.info("AUC-ROC: " + f'{float(roc_auc):.2f}')
logging.info("AUC-PRE: " + f'{float(prc_auc):.2f}')
# Plot the ROC
if ax:
ax.plot(fpr, tpr)
ax.set_title(name + ', AUC: ' + f'{float(roc_auc):.2f}')
ax.set(xlabel = "False positive rate", ylabel = "True positive rate")
# Get and return optimal treshold using Youden’s J statistic
#optimal_idx = np.argmax(tpr - fpr)
#optimal_trs = trs[optimal_idx]
# Get and return optimal treshold using equal error rate (EER)
fnr = 1 - tpr
optimal_trs = trs[np.nanargmin(np.absolute((fnr - fpr)))]
return optimal_trs
## Get the OK and NOK counts together with TPR and TNR
def getConfusionMatrix(self, labelsPred, name, ax = None):
cm = confusion_matrix(self.labelsTs, labelsPred)
precision = precision_score(self.labelsTs, labelsPred)
recall = recall_score(self.labelsTs, labelsPred)
f1score = f1_score(self.labelsTs, labelsPred)
logging.info("Precision: " + f'{float(precision):.2f}')
logging.info("Recall: " + f'{float(recall):.2f}')
logging.info("F1-score: " + f'{float(f1score):.2f}')
# Get the TPR and TNR
tnr = cm[0, 0]/(cm[0, 0] + cm[0, 1])
tpr = cm[1, 1]/(cm[1, 0] + cm[1, 1])
logging.info("TPR: " + f'{float(tnr):.2f}')
logging.info("TNR: " + f'{float(tpr):.2f}')
# Get the balance ratio
tnc = 1 if cm[0, 0] == 0 else cm[0, 0]
tpc = 1 if cm[1, 1] == 0 else cm[1, 1]
bRatio = tnc/tpc if tnc <= tpc else -tpc/tnc
logging.info("Balance ratio: " + f'{float(bRatio):.2f}')
# Display confusion matrix
logging.info("Confusion matrix")
logging.info(cm)
if ax:
ConfusionMatrixDisplay.from_predictions(self.labelsTs, labelsPred, ax = ax)
## Classify the results
def dataClassify(self):
# Compare AD detection algorithms
outliers_fraction = 0.01
available_anomaly_algorithms = [
("Robust covariance", EllipticEnvelope(contamination = outliers_fraction, support_fraction = 0.9)),
("One-Class SVM", svm.OneClassSVM(nu = outliers_fraction, kernel = "rbf", gamma = 'scale')),
("Isolation Forest", IsolationForest(contamination = outliers_fraction, random_state = 42)),
("Local Outlier Factor", LocalOutlierFactor(n_neighbors = 15, contamination = outliers_fraction, novelty = True))]
anomaly_algorithms = [(name, algorithm) for name, algorithm in available_anomaly_algorithms if name in self.anomaly_algorithm_selection]
# Visualise the data
if self.visualize:
fig, axarr = plt.subplots(2, len(anomaly_algorithms))
fig.set_size_inches(16, 8)
tempTitle = "Classification results of the " + self.layerSel + '-' + self.modelName + '_' + self.labelInfo + " model using " + self.featExtName + " feature extraction"
fig.suptitle(tempTitle, fontsize=16)
#fig.subplots_adjust(left=0.02, right=0.98, bottom=0.001, top=0.96, wspace=0.05, hspace=0.01)
plotNum = 0
for name, algorithm in anomaly_algorithms:
# prepare save directory and file for the fit model
picklePath = os.path.join(self.fitPath, self.featExtName, f'{name}.pickle')
pickleDirectory = os.path.dirname(picklePath)
if not os.path.exists(pickleDirectory):
os.makedirs(pickleDirectory)
yamlPath = os.path.join(pickleDirectory, 'tresholds.yaml')
# Fit the model
if hasattr(self, 'metricsTr') and self.metricsTr is not None:
t0 = time.time()
algorithm.fit(self.metricsTr)
t1 = time.time()
logging.info('Model fitting time: ' + f'{float(t1 - t0):.2f}' + 's')
with open(picklePath, 'wb') as pickleFile:
pickle.dump(algorithm, pickleFile)
else:
if not os.path.exists(picklePath):
raise FileNotFoundError(f'Anomaly algorithm saved fit file not found for {name}!')
with open(picklePath, 'rb') as pickleFile:
algorithm_type = algorithm.__class__
algorithm = pickle.load(pickleFile)
if not isinstance(algorithm, algorithm_type):
raise ValueError(f'Loaded anomaly algorithm is not of the expected type! Expected {algorithm_type}, got {algorithm.__class__}')
# Get the scores from the validation DS, calculate ROC evaluation metric and get optimal trsh
#valScores = algorithm.decision_function(self.metricsVl)#.ravel()*(-1)
#optimal_trs = self.getEvaluationMetrics(valScores, name, axarr[0][plotNum])
# Get the scores from the test DS, calculate ROC evaluation metric and get optimal trsh
if hasattr(self, 'metricsTs') and self.metricsTs is not None:
decisionMetrics = self.metricsTs
elif hasattr(self, 'metricsPr') and self.metricsPr is not None:
decisionMetrics = self.metricsPr
else:
raise ValueError(f'No usable metrics for decision function')
# Get optimal tresholds
tstScores = algorithm.decision_function(decisionMetrics)
if hasattr(self, 'metricsTr') and self.metricsTr is not None:
if self.visualize:
optimal_trs = self.getEvaluationMetrics(tstScores, name, axarr[0][plotNum])
else:
optimal_trs = self.getEvaluationMetrics(tstScores, name)
tresholds = {}
if os.path.exists(yamlPath):
with open(yamlPath, 'r') as trsFile:
tresholds = yaml.safe_load(trsFile)
tresholds.update({name: f'{optimal_trs}'})
with open(yamlPath, 'w') as trsFile:
yaml.safe_dump(tresholds, trsFile)
else:
if not os.path.exists(yamlPath):
raise ValueError(f'Unable to load {yamlPath}')
with open(yamlPath, 'r') as trsFile:
tresholds = yaml.safe_load(trsFile)
optimal_trs = np.float64(tresholds[name])
if optimal_trs is None:
raise ValueError(f'Unable to read treshold from {yamlPath}')
# Fit the data and tag outliers
y_pred = tstScores
okIdx = np.where(tstScores >= optimal_trs)
nokIdx = np.where(tstScores < optimal_trs)
y_pred[okIdx] = 1
y_pred[nokIdx] = -1
self.predictedLabels = np.zeros_like(y_pred, dtype=bool)
self.predictedLabels[okIdx] = True
# Calculate confusion matrix
if self.visualize:
self.getConfusionMatrix(y_pred, name, axarr[1][plotNum])
plotNum +=1
if self.visualize:
fig.tight_layout()
fig.subplots_adjust(top=0.88)
fig.savefig(os.path.join(self.modelDataPath, self.layerSel + '-' + self.modelName + '_' + self.labelInfo + '_' + self.featExtName + '_ClassEval_.png'))
## Visualise the feature space
def fsVisualise(self, metrics, labels):
# Reduce the dimensionality of metrics for t-SNE transformation
if metrics.shape[0] > 50 and metrics.shape[1] > 50:
pca_red = PCA(n_components = 50)
metrics = pca_red.fit_transform(metrics)
# Perform the t-SNE and feature space visualisation
if metrics.shape[0] < 30:
# Set perplexity as a half value to the number of samples for small datasets
perp = int(metrics.shape[0]/2)
else:
perp = 30
tsne_metrics = TSNE(n_components=2, perplexity=perp, n_iter=1000, learning_rate=100, init='pca').fit_transform(metrics)
# Perform the PCA and feature space visualisation
pca = PCA(n_components=2)
pca_metrics = pca.fit_transform(metrics)
# Visualise the data
fig, axarr = plt.subplots(2)
fig.set_size_inches(8, 8)
tempTitle = "FS visualisations of the " + self.layerSel + '-' + self.modelName + '_' + self.labelInfo + " model using " + self.featExtName + " feature extraction"
fig.suptitle(tempTitle, fontsize=12)
okIdx = np.where(labels == 1)
nokIdx = np.where(labels == -1)
axarr[0].scatter(tsne_metrics[okIdx, 0], tsne_metrics[okIdx, 1], s = 4)
axarr[0].scatter(tsne_metrics[nokIdx, 0], tsne_metrics[nokIdx, 1], s = 4)
axarr[0].set(xlabel = "t-SNE 1", ylabel = "t-SNE 2")
axarr[0].set_title("t-SNE Feature space visualisation")
axarr[0].legend(["OK", "NOK"], loc='upper right')
axarr[1].scatter(pca_metrics[okIdx, 0], pca_metrics[okIdx, 1], s = 4)
axarr[1].scatter(pca_metrics[nokIdx, 0], pca_metrics[nokIdx, 1], s = 4)
axarr[1].set(xlabel = "PCA 1", ylabel = "PCA 2")
axarr[1].set_title("PCA Feature space visualisation")
axarr[1].legend(["OK", "NOK"], loc='upper right')
fig.tight_layout()
fig.subplots_adjust(top=0.88)
fig.savefig(os.path.join(self.modelDataPath, self.layerSel + '-' + self.modelName + '_' + self.labelInfo + '_' + self.featExtName + '_FeatureSpace.png'))