-
Notifications
You must be signed in to change notification settings - Fork 26
/
mesa.py
259 lines (206 loc) · 9.96 KB
/
mesa.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
# -*- coding: utf-8 -*-
"""
Created on Sat Feb 8 02:27:20 2020
@author: ZhiningLiu1998
mailto: [email protected] / [email protected]
"""
import os
import torch
import pandas as pd
import numpy as np
from gym import spaces
from sac_src.sac import SAC
from sac_src.replay_memory import ReplayMemory
from environment import EnsembleTrainingEnv
from utils import *
class Mesa(EnsembleTrainingEnv):
"""The ensemble imbalanced learning framework MESA.
Parameters
----------
args : arguments
See arguments.py for more information.
base_estimator : scikit-learn classifier object
The base estimator used to build ensemble classifiers.
NO need to support sample weighting.
Built-in `fit()`, `predict()`, `predict_proba()` methods are required.
n_estimators : int, optional (default=10)
The number of base estimators used to form an MESA ensemble.
Attributes
----------
args : arguments
rater : object (Rater)
Rater for evaluate classifiers performance on class imabalanced data.
See arguments.py for more information.
base_estimator_ : object (scikit-learn classifier)
The base estimator from which the ensemble is grown.
estimators_ : list of classifiers
The collection of fitted sub-estimators.
n_estimators : int
The number of base estimators used to form an MESA ensemble.
meta_sampler : object (SAC)
The meta-sampler in MESA.
env : object (EnsembleTrainingEnv)
The ensemble training environment in MESA.
memory : object (ReplayMemory)
The replay memory for Soft Actor-Critic training.
"""
def __init__(self, args, base_estimator, n_estimators=10):
super(Mesa, self).__init__(args, base_estimator)
# state-size = 2 x num_bins
state_size = int(args.num_bins*2)
action_space = spaces.Box(low=0.0, high=1.0, shape=[1], dtype=np.float32)
self.args = args
self.n_estimators = n_estimators
self.base_estimator_ = base_estimator
self.meta_sampler = SAC(state_size, action_space, self.args)
self.env = EnsembleTrainingEnv(args, base_estimator)
self.memory = ReplayMemory(self.args.replay_size)
def meta_fit(self, X_train, y_train, X_valid, y_valid, X_test=None, y_test=None):
"""Meta-training process of MESA.
Parameters
----------
X_train : array-like of shape = [n_training_samples, n_features]
The training data instances.
y_train : array-like of shape = [n_training_samples]
Labels for X_train.
X_valid : array-like of shape = [n_validation_samples, n_features]
The validation data instances.
y_valid : array-like of shape = [n_validation_samples]
Labels for X_valid.
X_test : array-like of shape = [n_training_samples, n_features], optional (default=None)
The test data instances.
y_train : array-like of shape = [n_training_samples], optional (default=None)
Labels for X_test.
Returns
----------
self : object (Mesa)
"""
# initialize replay memory and environment
self.env.load_data(X_train, y_train, X_valid, y_valid, X_test, y_test, train_ratio=self.args.train_ratio)
self.memory = memory_init_fulfill(self.args, ReplayMemory(self.args.replay_size))
self.scores = []
total_steps = self.args.update_steps + self.args.start_steps
num_steps, num_updates, num_episodes = 0, 0, 0
# start meta-training
while num_steps < total_steps:
self.env.init()
state = self.env.get_state()
done = False
# for each episode
while not done:
num_steps += 1
# take an action
if num_steps >= self.args.start_steps:
action, by = self.meta_sampler.select_action(state), 'mesa'
else:
action, by = self.meta_sampler.action_space.sample(), 'rand'
# store transition
next_state, reward, done, info = self.env.step(action[0])
reward = reward * self.args.reward_coefficient
self.memory.push(state, action, reward, next_state, float(done))
# update meta-sampler parameters
if num_steps > self.args.start_steps:
for i in range(self.args.updates_per_step):
_, _, _, _, _ = self.meta_sampler.update_parameters(
self.memory, self.args.batch_size, num_updates)
num_updates += self.args.updates_per_step
# print log to stdout
if self.args.meta_verbose is 'full':
print ('Epi.{:<4d} updates{:<4d}| {} | {} by {}'.format(num_episodes, num_updates, info, action[0], by))
if done:
num_episodes += 1
self.record_scores()
# record print mean score of latest args.meta_verbose_mean_episodes to stdout
self.verbose_mean_scores(num_episodes, num_updates, by)
return self
def record_scores(self):
"""Record the training/validation/test performance scores."""
train_score = self.env.rater.score(self.env.y_train, self.env.y_pred_train_buffer)
valid_score = self.env.rater.score(self.env.y_valid, self.env.y_pred_valid_buffer)
test_score = self.env.rater.score(self.env.y_test, self.env.y_pred_test_buffer) if self.env.flag_use_test_set else 'NULL'
self.scores.append([train_score, valid_score, test_score] if self.env.flag_use_test_set else [train_score, valid_score])
return
def verbose_mean_scores(self, num_episodes, num_updates, by):
"""Print mean score of latest n episodes to stdout.
n = args.meta_verbose_mean_episodes
Parameters
----------
num_episodes : int
The number of finished meta-training episodes.
num_updates : int
The number of finished meta-sampler updates.
by : {'rand', 'mesa'}, string
The way of selecting actions in the current episode.
"""
if self.args.meta_verbose is 'full' or (self.args.meta_verbose != 0 and num_episodes % self.args.meta_verbose == 0):
view_bound = max(-self.args.meta_verbose_mean_episodes, -len(self.scores))
recent_scores_mean = np.array(self.scores)[view_bound:].mean(axis=0)
print ('Epi.{:<4d} updates {:<4d} |last-{}-mean-{}| train {:.3f} | valid {:.3f} | test {:.3f} | by {}'.format(
num_episodes, num_updates, self.args.meta_verbose_mean_episodes, self.args.metric,
recent_scores_mean[0], recent_scores_mean[1], recent_scores_mean[2], by))
return
def fit(self, X, y, X_valid, y_valid, n_estimators=None, verbose=False):
"""Build a MESA ensemble from training set (X, y) and validation set (X_valid, y_valid).
Parameters
----------
X : array-like of shape = [n_training_samples, n_features]
The training data instances.
y : array-like of shape = [n_training_samples]
Labels for X.
X_valid : array-like of shape = [n_validation_samples, n_features]
The validation data instances.
y_valid : array-like of shape = [n_validation_samples]
Labels for X_valid.
n_estimators : int, optional (default=self.n_estimators)
The number of base estimators used to form an MESA ensemble.
verbose: bool, optional (default=False)
Whether to print progress messages to stdout.
Returns
----------
self : object (Mesa)
"""
n_estimators = self.n_estimators if n_estimators is None else n_estimators
self.load_data(X, y, X_valid, y_valid)
self.init()
self.actions_record = []
for i in range(n_estimators-1):
state = self.get_state()
action = self.meta_sampler.select_action(state)
self.actions_record.append(action[0])
_, _, _, info = self.step(action[0], verbose)
if verbose:
print ('{:<12s} | action: {} {}'.format('Mesa', action, info))
return self
def save_meta_sampler(self, directory='save_model', suffix='meta_sampler'):
"""Save trained meta-sampler to files.
Parameters
----------
directory : string, optional (default='save_model')
The directory to save files.
Create the directory if it does not exist.
suffix : string, optional (default='meta_sampler')
The actor network will be saved in {directory}/actor_{suffix}.
The critic network will be saved in {directory}/critic_{suffix}.
"""
directory_path = f'{directory}/'
if not os.path.exists(directory_path):
os.makedirs(directory_path)
actor_path = f'{directory_path}actor_{suffix}'
critic_path = f'{directory_path}critic_{suffix}'
self.meta_sampler.save_model(actor_path, critic_path)
return
def load_meta_sampler(self, directory='save_model', suffix='meta_sampler'):
"""Load trained meta-sampler from files.
Parameters
----------
directory : string, optional (default='save_model')
The directory to load files.
suffix : string, optional (default='meta_sampler')
The actor network will be loaded from {directory}/actor_{suffix}.
The critic network will be loaded from {directory}/critic_{suffix}.
"""
directory_path = f'{directory}/'
actor_path = f'{directory_path}actor_{suffix}'
critic_path = f'{directory_path}critic_{suffix}'
self.meta_sampler.load_model(actor_path, critic_path)
return self