-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrun_p1.py
251 lines (197 loc) · 9.3 KB
/
run_p1.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
from p1 import *
import numpy as np
import pandas as pd
import os
import matplotlib.pyplot as plt
from matplotlib import rc
from matplotlib import rcParams
rcParams.update({'figure.autolayout': True})
font = {'size': 15}
rc('font', **font)
#test to run
def gen_data(data_dir = "data"):
game_gen = data_gen(game=randomwalk(num_states=7), num_sets=100, trainset_size=10)
game_gen.generate_data()
game_gen.save_data(data_dir=data_dir)
return None
def run_tdSimulator(alpha=0.01, decay=0.9995, lamb=0.0, init_w=0.5, maxepisode=100000, epsilon=1e-15):
tdSimulator = tdLearner_from_simulator(alpha=alpha, decay=decay, lamb=lamb, init_w=init_w)
print "initial w is:/n", tdSimulator.w
tdSimulator.learn(maxepisode=maxepisode, epsilon=epsilon)
print tdSimulator.w
print tdSimulator.rmse
#print tdSimulator.ste
return None
def run_tdAllData(data_dir="data", alpha=0.01, lamb=0.0, init_w=0.5, maxepisode=100000, epsilon=1e-15):
tdData = tdLearner_from_data(alpha=alpha, lamb=lamb, init_w=init_w)
tdData.read_data(data_dir)
tdData.learn_all_data_repeat(maxepisode=maxepisode, epsilon=epsilon)
print np.array(tdData.w)
print tdData.rmse
#print tdData.ste
return None
def run_tdTrainsetRepeat(data_dir="data", alpha=0.01, lamb=0.0, init_w=0.5, dataset_num=0, maxepisode=30000, epsilon=1e-15):
tdTrainset = tdLearner_from_data(alpha=alpha, lamb=lamb, init_w=init_w)
tdTrainset.read_data(data_dir)
tdTrainset.learn_one_trainset_repeat(dataset_num=dataset_num, maxepisode=maxepisode, epsilon=epsilon)
print np.array(tdTrainset.w)
print tdTrainset.rmse
#print tdTrainset.ste
return None
def run_tdTrainsetRepeat_allSets(data_dir="data", alpha=0.01, lamb=0.0, init_w=0.5, maxepisode=30000, epsilon=1e-15, verbose=False):
experiment = tdLearner_from_data(alpha=alpha, lamb=lamb, init_w=init_w)
experiment.read_data(data_dir)
experiment.learn_one_trainset_repeat_allsets(maxepisode=maxepisode, epsilon=epsilon, verbose=verbose)
#print experiment.rmse
#print experiment.ste
#print np.array(experiment.w_list).mean(axis=0)
return experiment.rmse, experiment.ste
def run_experiment_1(data_dir="data", results_dir="results", lamb_range=np.arange(0., 1.05, 0.1)):
rmse_list = []
ste_list = []
for lamb in lamb_range:
print "lamb = {}".format(lamb)
rmse, ste = run_tdTrainsetRepeat_allSets(data_dir=data_dir, alpha=0.01, lamb=lamb)
rmse_list.append(rmse)
ste_list.append(ste)
print "rmse = {}\nste = {}".format(rmse, ste)
print np.array(rmse_list)
print np.array(ste_list)
result = np.array([rmse_list, ste_list]).T
result = pd.DataFrame(result, index=lamb_range, columns=["Error", "STE"])
if not os.path.exists(results_dir):
os.makedirs(results_dir)
result.to_csv(os.path.join(results_dir, "experiment1.csv"), header=True, index=True)
return None
def plot_experiment_1(results_dir, name):
csv_file = os.path.join(results_dir, "{}.csv".format(name))
fig_file = os.path.join(results_dir, "{}.png".format(name))
df = pd.read_csv(csv_file, header=0, index_col=0)
print df
#plt.errorbar(df.index, 'Error', yerr='STE', data=df)
#plt.show()
ax = df[["Error"]].plot(style=["ro-"])
ax.set_xlim(-0.1, 1.1)
ax.set_ylabel("Error")
ax.set_xlabel(r"$\lambda$")
plt.savefig(fig_file)
def run_tdTrainset(data_dir="data", alpha=0.01, dataset_num=0, lamb=0.0, init_w=0.5, verbose=False):
experiment = tdLearner_from_data(alpha=alpha, lamb=lamb, init_w=init_w)
experiment.read_data(data_dir)
experiment.learn_one_trainset(dataset_num=dataset_num, verbose=verbose)
#print experiment.rmse
#print experiment.ste
#print np.array(experiment.w_list).mean(axis=0)
return experiment.rmse, experiment.ste
def run_tdTrainset_allSets(data_dir="data", alpha=0.01, lamb=0.0, init_w=0.5, verbose=False):
experiment = tdLearner_from_data(alpha=alpha, lamb=lamb, init_w=init_w)
experiment.read_data(data_dir)
experiment.learn_one_trainset_allsets(verbose=verbose)
#print experiment.rmse
#print experiment.ste
#print np.array(experiment.w_list).mean(axis=0)
return experiment.rmse, experiment.ste
def run_experiment_2(data_dir="data", results_dir="results", alpha_range = np.arange(0., 0.65, 0.05), lamb_range=[0., 0.3, 0.8, 1.], init_w=0.5, verbose=False):
result = dict()
for lamb in lamb_range:
result[r"$\lambda$={}".format(lamb)] = []
for alpha in alpha_range:
print "lamb = {}, alpha = {}".format(lamb, alpha)
experiment = tdLearner_from_data(alpha=alpha, lamb=lamb, init_w=init_w)
experiment.read_data(data_dir)
experiment.learn_one_trainset_allsets(verbose=verbose)
#print np.array(experiment.w_list).mean(axis=0)
#print experiment.rmse, experiment.ste
result[r"$\lambda$={}".format(lamb)].append(experiment.rmse)
result = pd.DataFrame(result)
result.index = alpha_range
print result
if not os.path.exists(results_dir):
os.makedirs(results_dir)
result.to_csv(os.path.join(results_dir, "experiment2.csv"), header=True, index=True)
def plot_experiment_2(results_dir, name):
csv_file = os.path.join(results_dir, "{}.csv".format(name))
fig_file = os.path.join(results_dir, "{}.png".format(name))
fig_file1 = os.path.join(results_dir, "{}_1.png".format(name))
df = pd.read_csv(csv_file, header=0, index_col=0)
print df
ax = df.plot(style=["ro-", "bs-", "g^-", "yv-"])
ax.set_xlim(-0.04, 0.65)
ax.set_ylim(0, 0.8)
ax.set_ylabel("Error")
ax.set_xlabel(r"$\alpha$")
#plt.show()
plt.savefig(fig_file)
ax = df.plot(style=["ro-", "bs-", "g^-", "yv-"])
ax.set_xlim(-0.04, 0.65)
ax.set_ylabel("Error")
ax.set_xlabel(r"$\alpha$")
#plt.show()
plt.savefig(fig_file1)
def run_experiment_3(data_dir="data", results_dir="results", alpha_range = np.arange(0., 0.65, 0.05), lamb_range=np.arange(0., 1.05, 0.1), init_w=0.5, verbose=False):
result = dict()
for lamb in lamb_range:
result[lamb] = []
for alpha in alpha_range:
print "lamb = {}, alpha = {}".format(lamb, alpha)
experiment = tdLearner_from_data(alpha=alpha, lamb=lamb, init_w=init_w)
experiment.read_data(data_dir)
experiment.learn_one_trainset_allsets(verbose=verbose)
#print np.array(experiment.w_list).mean(axis=0)
#print experiment.rmse, experiment.ste
result[lamb].append(experiment.rmse)
result = pd.DataFrame(result)
result.index = alpha_range
print result
result2 = result.T
print result2
if not os.path.exists(results_dir):
os.makedirs(results_dir)
result2.to_csv(os.path.join(results_dir, "experiment3_1.csv"), header=True, index=True)
result2=result2.min(axis=1)
result2.columns = ["Error"]
result2.to_csv(os.path.join(results_dir, "experiment3_2.csv"), header=True, index=True)
def plot_experiment_3(results_dir, name):
csv_file = os.path.join(results_dir, "{}.csv".format(name))
fig_file = os.path.join(results_dir, "{}.png".format(name))
df = pd.read_csv(csv_file, header=0, index_col=0)
print df
df.columns = ["Error"]
print df
#plt.errorbar(df.index, 'Error', yerr='ste', data=df)
#plt.show()
ax = df.plot(style=["ro-"])
ax.set_xlim(-0.05, 1.05)
ax.set_ylabel("Error")
ax.set_xlabel(r"$\lambda$")
plt.savefig(fig_file)
def tests():
#test1: run TD(lamb) with game simulator until convergence
run_tdSimulator(alpha=0.1, decay=0.999, lamb=0.0, init_w=0.5, maxepisode=100000, epsilon=1e-15)
#test2: run TD(lamb) with all generated data (100x10 sequences) repeatedly until convergence
run_tdAllData(data_dir="data", alpha=0.01, lamb=0.2, init_w=0.5, maxepisode=100, epsilon=1e-5)
#test3: run TD(lamb) with a particular dataset repeatedly until convergence
run_tdTrainsetRepeat(data_dir="data", alpha=0.55, lamb=0.0, init_w=0.5, dataset_num=78, maxepisode=100000, epsilon=1e-15)
#test4: run TD(lamb) with a particular dataset once
rmse, ste = run_tdTrainset(data_dir="data", alpha=0.55, dataset_num=78, lamb=0.0, init_w=0.5, verbose=True)
print "rmse = {}".format(rmse)
#test5: run TD(lamb) with all dataset once then average
rmse, ste = run_tdTrainset_allSets(data_dir="data", alpha=0.55, lamb=0.0, init_w=0.5, verbose=True)
#test6: run TD(lamb) with all dataset repeatedly until convergence, then average
rmse, ste = run_tdTrainsetRepeat_allSets(data_dir="data", alpha=0.01, lamb=0.0, init_w=0.5, maxepisode=30000, epsilon=1e-15, verbose=True)
print "rmse = {}\nste = {}".format(rmse, ste)
if __name__ == '__main__':
data_dir = "data"
results_dir = "results"
generate_new_data = False #change this to True if new data is needed
#generate data from simulation
if generate_new_data: gen_data(data_dir)
#experiment 1
run_experiment_1(data_dir=data_dir, results_dir=results_dir)
plot_experiment_1(results_dir=results_dir, name="experiment1")
#experiment 2
run_experiment_2(data_dir=data_dir, results_dir=results_dir)
plot_experiment_2(results_dir=results_dir, name="experiment2")
#experiment 3
run_experiment_3(data_dir=data_dir, results_dir=results_dir)
plot_experiment_3(results_dir=results_dir, name="experiment3_2")