forked from dreamquark-ai/tabnet
-
Notifications
You must be signed in to change notification settings - Fork 0
/
tab_model.py
executable file
·154 lines (128 loc) · 4.6 KB
/
tab_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
import torch
import numpy as np
from scipy.special import softmax
from pytorch_tabnet.utils import SparsePredictDataset, PredictDataset, filter_weights
from pytorch_tabnet.abstract_model import TabModel
from pytorch_tabnet.multiclass_utils import infer_output_dim, check_output_dim
from torch.utils.data import DataLoader
import scipy
class TabNetClassifier(TabModel):
def __post_init__(self):
super(TabNetClassifier, self).__post_init__()
self._task = 'classification'
self._default_loss = torch.nn.functional.cross_entropy
self._default_metric = 'accuracy'
def weight_updater(self, weights):
"""
Updates weights dictionary according to target_mapper.
Parameters
----------
weights : bool or dict
Given weights for balancing training.
Returns
-------
bool or dict
Same bool if weights are bool, updated dict otherwise.
"""
if isinstance(weights, int):
return weights
elif isinstance(weights, dict):
return {self.target_mapper[key]: value for key, value in weights.items()}
else:
return weights
def prepare_target(self, y):
return np.vectorize(self.target_mapper.get)(y)
def compute_loss(self, y_pred, y_true):
return self.loss_fn(y_pred, y_true.long())
def update_fit_params(
self,
X_train,
y_train,
eval_set,
weights,
):
output_dim, train_labels = infer_output_dim(y_train)
for X, y in eval_set:
check_output_dim(train_labels, y)
self.output_dim = output_dim
self._default_metric = ('auc' if self.output_dim == 2 else 'accuracy')
self.classes_ = train_labels
self.target_mapper = {
class_label: index for index, class_label in enumerate(self.classes_)
}
self.preds_mapper = {
str(index): class_label for index, class_label in enumerate(self.classes_)
}
self.updated_weights = self.weight_updater(weights)
def stack_batches(self, list_y_true, list_y_score):
y_true = np.hstack(list_y_true)
y_score = np.vstack(list_y_score)
y_score = softmax(y_score, axis=1)
return y_true, y_score
def predict_func(self, outputs):
outputs = np.argmax(outputs, axis=1)
return np.vectorize(self.preds_mapper.get)(outputs.astype(str))
def predict_proba(self, X):
"""
Make predictions for classification on a batch (valid)
Parameters
----------
X : a :tensor: `torch.Tensor` or matrix: `scipy.sparse.csr_matrix`
Input data
Returns
-------
res : np.ndarray
"""
self.network.eval()
if scipy.sparse.issparse(X):
dataloader = DataLoader(
SparsePredictDataset(X),
batch_size=self.batch_size,
shuffle=False,
)
else:
dataloader = DataLoader(
PredictDataset(X),
batch_size=self.batch_size,
shuffle=False,
)
results = []
for batch_nb, data in enumerate(dataloader):
data = data.to(self.device).float()
output, M_loss = self.network(data)
predictions = torch.nn.Softmax(dim=1)(output).cpu().detach().numpy()
results.append(predictions)
res = np.vstack(results)
return res
class TabNetRegressor(TabModel):
def __post_init__(self):
super(TabNetRegressor, self).__post_init__()
self._task = 'regression'
self._default_loss = torch.nn.functional.mse_loss
self._default_metric = 'mse'
def prepare_target(self, y):
return y
def compute_loss(self, y_pred, y_true):
return self.loss_fn(y_pred, y_true)
def update_fit_params(
self,
X_train,
y_train,
eval_set,
weights
):
if len(y_train.shape) != 2:
msg = "Targets should be 2D : (n_samples, n_regression) " + \
f"but y_train.shape={y_train.shape} given.\n" + \
"Use reshape(-1, 1) for single regression."
raise ValueError(msg)
self.output_dim = y_train.shape[1]
self.preds_mapper = None
self.updated_weights = weights
filter_weights(self.updated_weights)
def predict_func(self, outputs):
return outputs
def stack_batches(self, list_y_true, list_y_score):
y_true = np.vstack(list_y_true)
y_score = np.vstack(list_y_score)
return y_true, y_score