-
Notifications
You must be signed in to change notification settings - Fork 0
/
model_lib.py
224 lines (180 loc) · 8.04 KB
/
model_lib.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
# -*- coding: utf-8 -*-
"""
Created on Sun Sep 24 22:01:12 2023
@author: ZHANG Jun
"""
import os
import json
import copy
import torch
def save_model(model, model_save_dir='agat_model'):
"""Saving PyTorch model to the disk. Save PyTorch model, including parameters and structure. See: https://pytorch.org/tutorials/beginner/basics/saveloadrun_tutorial.html
:param model: A PyTorch-based model.
:type model: PyTorch-based model.
:param model_save_dir: A directory to store the model, defaults to 'agat_model'
:type model_save_dir: str, optional
:output: A file saved to the disk under ``model_save_dir``.
:outputtype: A file.
"""
if not os.path.exists(model_save_dir):
os.makedirs(model_save_dir)
torch.save(model, os.path.join(model_save_dir, 'agat.pth'))
def load_model(model_save_dir='agat_model', device='cuda'):
"""Loading PyTorch model from the disk.
:param model_save_dir: A directory to store the model, defaults to 'agat_model'
:type model_save_dir: str, optional
:param device: Device for the loaded model, defaults to 'cuda'
:type device: str, optional
:return: A PyTorch-based model.
:rtype: PyTorch-based model.
"""
device = torch.device(device)
if device.type == 'cuda':
new_model = torch.load(os.path.join(model_save_dir, 'agat.pth'))
elif device.type == 'cpu':
new_model = torch.load(os.path.join(model_save_dir, 'agat.pth'),
map_location=torch.device(device))
new_model.eval()
new_model = new_model.to(device)
new_model.device = device
return new_model
def save_state_dict(model, state_dict_save_dir='agat_model', **kwargs):
"""Saving state dict (model weigths and other input info) to the disk. See: https://pytorch.org/tutorials/beginner/basics/saveloadrun_tutorial.html
:param model: A PyTorch-based model.
:type model: PyTorch-based model.
:param state_dict_save_dir: A directory to store the model state dict (model weigths and other input info), defaults to 'agat_model'
:type state_dict_save_dir: str, optional
:param **kwargs: More information you want to save.
:type **kwargs: kwargs
:output: A file saved to the disk under ``model_save_dir``.
:outputtype: A file
"""
if not os.path.exists(state_dict_save_dir):
os.makedirs(state_dict_save_dir)
checkpoint_dict = {**{'model_state_dict': model.state_dict()}, **kwargs}
torch.save(checkpoint_dict, os.path.join(state_dict_save_dir, 'agat_state_dict.pth'))
def load_state_dict(state_dict_save_dir='agat_model'):
"""Loading state dict (model weigths and other info) from the disk. See: https://pytorch.org/tutorials/beginner/basics/saveloadrun_tutorial.html
:param state_dict_save_dir: A directory to store the model state dict (model weigths and other info), defaults to 'agat_model'
:type state_dict_save_dir: str, optional
:return: State dict.
:rtype: TYPE
.. note::
Reconstruct a model/optimizer before using the loaded state dict.
Example::
model = PotentialModel(...)
model.load_state_dict(checkpoint['model_state_dict'])
new_model.eval()
model = model.to(device)
model.device = device
optimizer = ...
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
"""
checkpoint_dict = torch.load(os.path.join(state_dict_save_dir, 'agat_state_dict.pth'))
return checkpoint_dict
def config_parser(config):
"""Parse the input configurations/settings.
:param config: configurations
:type config: str/dict. if str, load from the json file.
:raises TypeError: DESCRIPTION
:return: TypeError('Wrong configuration type.')
:rtype: TypeError
"""
if isinstance(config, dict):
return config
elif isinstance(config, str):
with open(config, 'r') as config_f:
return json.load(config_f)
elif isinstance(config, type(None)):
return {}
else:
raise TypeError('Wrong configuration type.')
class EarlyStopping:
def __init__(self, model, logger, patience=10, model_save_dir='model_save_dir'):
"""Stop training when model performance stop improving after some steps.
:param model: AGAT model
:type model: torch.nn
:param logger: I/O file
:type logger: _io.TextIOWrapper
:param patience: Stop patience, defaults to 10
:type patience: int, optional
:param model_save_dir: A directory to save the model, defaults to 'model_save_dir'
:type model_save_dir: str, optional
"""
self.model = model
self.patience = patience
self.counter = 0
self.best_score = None
self.update = None
self.early_stop = False
self.logger = logger
self.model_save_dir = model_save_dir
if not os.path.exists(self.model_save_dir):
os.mkdir(model_save_dir)
self.save_model_info()
def step(self, score, epoch, model, optimizer):
if self.best_score is None:
self.best_score = score
self.update = True
# self.save_model(model, model_save_dir=self.model_save_dir)
# self.save_checkpoint(model, model_save_dir=self.model_save_dir)
elif score > self.best_score:
self.update = False
self.counter += 1
print(f'User log: EarlyStopping counter: {self.counter} out of {self.patience}',
file=self.logger)
if self.counter >= self.patience:
self.early_stop = True
else:
self.best_score = score
self.update = True
self.save_model(model)
save_state_dict(model, state_dict_save_dir=self.model_save_dir,
optimizer_state_dict=optimizer.state_dict(),
epoch=epoch, total_loss=score)
self.counter = 0
return self.early_stop
def save_model(self, model):
'''Saves model when validation loss decrease.'''
torch.save(model, os.path.join(self.model_save_dir, 'agat.pth'))
print(f'User info: Save model with the best score: {self.best_score}',
file=self.logger)
# def save_checkpoint(self, model):
# '''Saves model when validation loss decrease.'''
# torch.save({model.state_dict()}, os.path.join(self.model_save_dir, 'agat_state_dict.pth'))
def save_model_info(self):
info = copy.deepcopy(self.model.__dict__)
info = {k:v for k,v in info.items() if isinstance(v, (str, list, int, float))}
with open(os.path.join(self.model_save_dir, 'agat_model.json'), 'w') as f:
json.dump(info, f, indent=4)
def load_graph_build_method(path):
""" Load graph building scheme. This file is normally saved when you build your dataset.
:param path: Directory for storing ``graph_build_scheme.json`` file.
:type path: str
:return: A dict denotes how to build the graph.
:rtype: dict
"""
json_file = path
assert os.path.exists(json_file), f"{json_file} file dose not exist."
with open(json_file, 'r') as jsonf:
graph_build_scheme = json.load(jsonf)
return graph_build_scheme
def PearsonR(y_true, y_pred):
"""Calculating the Pearson coefficient.
:param y_true: The first torch.tensor.
:type y_true: torch.Tensor
:param y_pred: The second torch.tensor.
:type y_pred: torch.Tensor
:return: Pearson coefficient
:rtype: torch.Tensor
.. Note::
It looks like the `torch.jit.script` decorator is not helping in comuputing large `torch.tensor`, see `agat/test/tesor_computation_test.py` for more details.
"""
ave_y_true = torch.mean(y_true)
ave_y_pred = torch.mean(y_pred)
y_true_diff = y_true - ave_y_true
y_pred_diff = y_pred - ave_y_pred
above = torch.sum(torch.mul(y_true_diff, y_pred_diff))
below = torch.mul(torch.sqrt(torch.sum(torch.square(y_true_diff))),
torch.sqrt(torch.sum(torch.square(y_pred_diff))))
return torch.divide(above, below)