-
Notifications
You must be signed in to change notification settings - Fork 0
/
Model.py
289 lines (236 loc) · 11 KB
/
Model.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
import math
from matplotlib.backends.backend_pdf import PdfPages
import pickle
from sklearn.preprocessing import LabelBinarizer
from sklearn.neural_network import MLPClassifier
import pandas as pd
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
from sklearn.metrics import RocCurveDisplay
import numpy as np
from sklearn.preprocessing import OneHotEncoder
from sklearn.preprocessing import MinMaxScaler
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.metrics import roc_curve, auc
from sklearn.preprocessing import FunctionTransformer
from joblib import dump, load
from sklearn.neighbors import NearestNeighbors, VALID_METRICS_SPARSE
class Model:
def __init__(self, name,
categorical_features=[],
continous_features=[],
date_features=[],
drop_features=[],
classname='class',
):
self.logger = None
self.regression = False
self.continous_features = continous_features
self.categorical_features = categorical_features
self.date_features = date_features
self.drop_features = drop_features
self.classname = classname
self.name = name
def setLogger(self,logger):
self.logger = logger
def load_data_from(self,paths):
data=[]
for path in paths:
data.append(pd.read_csv(path))
data = pd.concat(data)
self.load_data(data=data)
def load_data(self,data=None):
if (data is None):
self.load_data_from([f'data/{self.name}.csv'])
return
self.X = data.drop([self.classname], axis=1)
self.y = data[self.classname]
self.X = self.X.loc[:, ~self.X.columns.str.contains('^Unnamed')]
def drop_columns(self,x, columns):
return x.drop(columns=self.drop_features)
def getTransformer(self):
categorical_transformer = Pipeline(steps=[
('onehot', OneHotEncoder(handle_unknown='ignore', sparse_output=False))
])
return ColumnTransformer(transformers=[
('num_preprocess', MinMaxScaler(), self.continous_features),
('cat', categorical_transformer, self.categorical_features),
# ('drop_columns', FunctionTransformer(self.drop_columns, validate=False), self.drop_features)
], verbose=False, remainder='passthrough')
def getPipline(self, model):
transformations = self.getTransformer()
return Pipeline(steps=[('preprocessor', transformations),
('classifier', model)])
def train(self):
self.clf = self.getPipline(self.getModel())
self.clf.fit(self.X_train, self.y_train)
self.saveModel()
def getFilename(self):
return f"models/{self.name}-{self.modelName}-pipe.pickle"
def saveModel(self):
pickle.dump(self.clf, open(self.getFilename(), "wb"))
def loadModel(self):
self.clf = pickle.load(open(self.getFilename(), "rb"))
def getLocalisedData(self, index, n_samples=1000, radius=1.5, globalSample=50000):
example = self.X.iloc[[index]]
samples = self.getRandomSamples(self.X, n_samples=globalSample)
relative_radius = self.getDistanceToNearestOpposite(
samples, example
)
# print(radius * relative_radius)
local = self.getNearestNeighbour(
samples, example, radius=radius * relative_radius)
# print(f"lenght : {len(local)}")
if (len(local) == 1):
print("not enough divesity trying again")
return self.getLocalisedData( index, n_samples, radius, globalSample)
if (len(local) == 0):
raise Exception(
'could not find any nearest neighbors, make radius bigger')
local = self.getRandomSamples(local, n_samples)
# print(local)
local[self.classname] = self.clf.predict(local)
return local
def getGlobalRandomSample(self, n_samples=1 ):
data = self.getRandomSamples(self.X, n_samples=n_samples)
data[self.classname] = self.clf.predict(data)
return data
def getGlobalStatisfeidSample(self, n_samples=1):
data = self.getRandomSamples(self.X, n_samples=n_samples*1000)
data[self.classname] = self.clf.predict(data)
return data.groupby(self.classname, group_keys=False).apply(
lambda x: x.sample(frac=0.001))
def getBalancedSample(self, n_samples=1):
samplesPerClass = n_samples / len(self.y.unique())
data = self.getRandomSamples(self.X, n_samples=n_samples)
data[self.classname] = self.clf.predict(data)
while True:
newsamples = self.getRandomSamples(self.X, n_samples=n_samples*1000)
newsamples[self.classname] = self.clf.predict(newsamples)
data = pd.concat([data,newsamples])
print(data[self.classname].value_counts())
if(all(count >
samplesPerClass for count in data[self.classname].value_counts())):
break
return data.groupby(self.classname, group_keys=False).apply(
lambda x: x.sample(n=math.ceil(samplesPerClass)))
def getRandomSamples(self, data, n_samples=1):
# Generate a new DataFrame with random values for numerical columns
randomized_num_df = pd.DataFrame()
for col in self.continous_features:
mean = data[col].mean()
std = data[col].std()
if (data[data[col] < 0].empty):
randomized_num_df[col] = np.abs(np.random.normal(
mean, std, n_samples))
else:
randomized_num_df[col] = np.random.normal(
mean, std, n_samples)
# Generate a new DataFrame with random values for categorical columns
randomized_cat_df = pd.DataFrame()
for col in self.categorical_features:
col_counts = data[col].value_counts()
col_values = col_counts.index.tolist()
col_weights = col_counts.values / data.shape[0]
randomized_cat_df[col] = np.random.choice(
col_values, size=n_samples, p=col_weights)
# Combine the numerical and categorical DataFrames into one DataFrame with the same index as the original DataFrame
randomized_df = pd.concat([randomized_num_df, randomized_cat_df], axis=1)
return randomized_df
def getNearestNeighbour(self, data, example, radius=0.2, hamming=False):
transformations = self.getTransformer()
transformations.fit_transform(self.X)
samples = transformations.transform(data)
example = transformations.transform(example)
if (hamming):
nn = NearestNeighbors(metric=self.hamming_distance)
else:
nn = NearestNeighbors(metric='minkowski')
nn = nn.fit(samples)
distance, indices = nn.radius_neighbors(
example, radius, sort_results=True, return_distance=True,
)
return data[data.index.isin(indices[0])]
def getDistanceToNearestOpposite(self, counterExamples, example):
if(self.regression):
ranges = self.get_quantile_ranges(self.clf.predict(
counterExamples), 4)
range_ = self.find_range_index(
ranges, self.clf.predict(example))[0]
counterExamples['range'] = self.clf.predict(
counterExamples)
counterExamples = counterExamples[counterExamples['range'] != range_]
counterExamples = counterExamples.drop(['range'], axis=1)
else:
class_ = self.clf.predict(example)[0]
counterExamples[self.classname] = self.clf.predict(counterExamples)
counterExamples = counterExamples[counterExamples[self.classname] != class_]
counterExamples = counterExamples.drop([self.classname], axis=1)
transformations = self.getTransformer()
transformations.fit_transform(self.X)
counterExamples = transformations.transform(counterExamples)
example = transformations.transform(example)
nn = NearestNeighbors(metric='minkowski')
nn = nn.fit(counterExamples)
distance, indices = nn.kneighbors(
example, 1
)
return distance[0][0]
def print(self, out):
print(out)
if (self.logger):
self.logger.info(out)
def findQuantile(self, data, number):
for key, value in data.items():
if number >= value[0] and number <= value[1]:
return key
return None
def find_range_index(self,data, numbers):
result = []
for number in numbers:
for key, value in data.items():
if number >= value[0] and number <= value[1]:
result.append(key)
break
else:
result.append(None)
return result
def get_quantile_ranges(self, df, num_quantiles=4):
# Add random noise to the input dataframe to ensure unique quantile boundaries
df_with_noise = df + np.random.normal(scale=1e-10, size=df.shape)
print(df_with_noise)
# Get the boundary ranges for the quantiles
quantile_labels = range(1, num_quantiles+1)
quantile_boundaries = pd.qcut(
df_with_noise, q=num_quantiles, labels=quantile_labels, retbins=True)[1]
# Store the boundary ranges in a dictionary
quantile_ranges = {}
for i, label in enumerate(quantile_labels):
lower_bound = math.floor(quantile_boundaries[i])
upper_bound = math.ceil(quantile_boundaries[i+1]-0.1)
quantile_ranges[label] = [lower_bound, upper_bound]
return quantile_ranges
def hamming_distance(self,x, y):
return np.sum(x != y)
def plot_mean_and_dist(self, data1, data2):
# Separate numerical and categorical columns
num_cols = data1.select_dtypes(include=[np.number]).columns.tolist()
cat_cols = data1.select_dtypes(
include=['object', 'category']).columns.tolist()
# Plot mean and distribution of numerical columns before randomization
fig, axs = plt.subplots(len(num_cols), 2, figsize=(8, 4 * len(num_cols)))
fig.suptitle('Mean and Distribution before and after randomization')
for i, col in enumerate(num_cols):
axs[i, 0].set_title(f"{col}: Before randomization")
axs[i, 0].axvline(x=data1[col].mean(), color='r', label='Mean')
axs[i, 0].hist(data1[col], bins=20, alpha=0.5, label='Distribution')
axs[i, 0].legend()
# Randomize DataFrame and plot mean and distribution of numerical column
axs[i, 1].set_title(f"{col}: After randomization")
axs[i, 1].axvline(x=data2[col].mean(),
color='r', label='Mean')
axs[i, 1].hist(data2[col], bins=20,
alpha=0.5, label='Distribution')
axs[i, 1].legend()
plt.show()