-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathTabNet.py
133 lines (94 loc) · 5.33 KB
/
TabNet.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
#!/usr/bin/env python
# coding: utf-8
import torch
import torch.nn as nn
import torch
import torch.nn as nn
import numpy as np
from sparsemax import Sparsemax
sparsemax = Sparsemax(dim=1)
# GLU
def glu(act, n_units):
act[:, :n_units] = act[:, :n_units].clone() * torch.nn.Sigmoid()(act[:, n_units:].clone())
return act
class TabNetModel(nn.Module):
def __init__(
self,
columns = 3,
num_features = 3,
feature_dims = 128,
output_dim =64,
num_decision_steps =6,
relaxation_factor = 0.5,
batch_momentum = 0.001,
virtual_batch_size = 2,
num_classes = 2,
epsilon = 0.00001
):
super().__init__()
self.columns = columns
self.num_features = num_features
self.feature_dims = feature_dims
self.output_dim = output_dim
self.num_decision_steps = num_decision_steps
self.relaxation_factor = relaxation_factor
self.batch_momentum = batch_momentum
self.virtual_batch_size = virtual_batch_size
self.num_classes = num_classes
self.epsilon = epsilon
self.feature_transform_linear1 = torch.nn.Linear(num_features, self.feature_dims * 2, bias=False)
self.BN = torch.nn.BatchNorm1d(num_features, momentum = batch_momentum)
self.BN1 = torch.nn.BatchNorm1d(self.feature_dims * 2, momentum = batch_momentum)
self.feature_transform_linear2 = torch.nn.Linear(self.feature_dims * 2, self.feature_dims * 2, bias=False)
self.feature_transform_linear3 = torch.nn.Linear(self.feature_dims * 2, self.feature_dims * 2, bias=False)
self.feature_transform_linear4 = torch.nn.Linear(self.feature_dims * 2, self.feature_dims * 2, bias=False)
self.mask_linear_layer = torch.nn.Linear(self.feature_dims * 2-output_dim, self.num_features, bias=False)
self.BN2 = torch.nn.BatchNorm1d(self.num_features, momentum = batch_momentum)
self.final_classifier_layer = torch.nn.Linear(self.output_dim, self.num_classes, bias=False)
def encoder(self, data):
batch_size = data.shape[0]
features = self.BN(data)
output_aggregated = torch.zeros([batch_size, self.output_dim])
masked_features = features
mask_values = torch.zeros([batch_size, self.num_features])
aggregated_mask_values = torch.zeros([batch_size, self.num_features])
complemantary_aggregated_mask_values =torch.ones([batch_size, self.num_features])
total_entropy = 0
for ni in range(self.num_decision_steps):
if ni==0:
transform_f1 = self.feature_transform_linear1(masked_features)
norm_transform_f1 = self.BN1(transform_f1)
transform_f2 = self.feature_transform_linear2(norm_transform_f1)
norm_transform_f2 = self.BN1(transform_f2)
else:
transform_f1 = self.feature_transform_linear1(masked_features)
norm_transform_f1 = self.BN1(transform_f1)
transform_f2 = self.feature_transform_linear2(norm_transform_f1)
norm_transform_f2 = self.BN1(transform_f2)
# GLU
transform_f2 = (glu(norm_transform_f2, self.feature_dims) +transform_f1) * np.sqrt(0.5)
transform_f3 = self.feature_transform_linear3(transform_f2)
norm_transform_f3 = self.BN1(transform_f3)
transform_f4 = self.feature_transform_linear4(norm_transform_f3)
norm_transform_f4 = self.BN1(transform_f4)
# GLU
transform_f4 = (glu(norm_transform_f4, self.feature_dims) + transform_f3) * np.sqrt(0.5)
decision_out = torch.nn.ReLU(inplace=True)(transform_f4[:, :self.output_dim])
# Decision aggregation
output_aggregated = torch.add(decision_out, output_aggregated)
scale_agg = torch.sum(decision_out, axis=1, keepdim=True) / (self.num_decision_steps - 1)
aggregated_mask_values = torch.add( aggregated_mask_values, mask_values * scale_agg)
features_for_coef = (transform_f4[:, self.output_dim:])
if ni<(self.num_decision_steps-1):
mask_linear_layer = self.mask_linear_layer(features_for_coef)
mask_linear_norm = self.BN2(mask_linear_layer)
mask_linear_norm = torch.mul(mask_linear_norm, complemantary_aggregated_mask_values)
mask_values = sparsemax(mask_linear_norm)
complemantary_aggregated_mask_values = torch.mul(complemantary_aggregated_mask_values,self.relaxation_factor - mask_values)
total_entropy = torch.add(total_entropy,torch.mean(torch.sum(-mask_values * torch.log(mask_values + self.epsilon),axis=1)) / (self.num_decision_steps - 1))
masked_features = torch.mul(mask_values , features)
return output_aggregated, total_entropy
def classify(self, output_logits):
logits = self.final_classifier_layer(output_logits)
predictions = torch.nn.Softmax(dim=1)(logits)
return logits, predictions