-
Notifications
You must be signed in to change notification settings - Fork 3
/
plot_simulation_convergence_num_p.py
187 lines (174 loc) · 5.83 KB
/
plot_simulation_convergence_num_p.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
import sys
import os
import argparse
import pickle
import numpy as np
import matplotlib
matplotlib.use('Agg')
from matplotlib import pyplot as plt
import seaborn as sns
import pandas as pd
from sklearn.linear_model import LinearRegression
import statsmodels.api as sm
import csv
from plot_simulation_general import read_method_result
from common import process_params, make_params
def parse_args(args):
parser = argparse.ArgumentParser(
"""
Plot for the simulation convergence check: what if num features grows
""")
parser.add_argument(
'--result-folder',
type=str,
default="simulation_convergence_num_p")
parser.add_argument(
'--spinn-template',
type=str,
default="_output/seed_%d/num_train_200/num_p_%d/fitted_spinn.pkl")
parser.add_argument(
'--file-template',
type=str,
default="_output/seed_%d/num_train_200/num_p_%d/fitted_spinn.csv")
parser.add_argument(
'--num-ps',
type=str,
default=make_params([25, 50, 100, 200, 400]))
parser.add_argument(
'--lasso-ratio',
type=float,
default=0.1)
parser.add_argument(
'--max-relevant-idx',
type=int,
default=6)
parser.add_argument(
'--seeds',
type=str,
default=make_params(range(1,21)))
parser.add_argument(
'--out-plot',
type=str,
default="_output/plot_simulation_mse.png")
parser.add_argument(
'--out-weight-plot',
type=str,
default="_output/plot_simulation_weights.png")
parser.set_defaults()
args = parser.parse_args(args)
args.seeds = process_params(args.seeds, int)
args.num_ps = process_params(args.num_ps, int)
return args
def plot_mse(args):
all_results = {
"mse": [],
"num_p": [],
"seed": []}
for seed in args.seeds:
for num_p in args.num_ps:
res_file = os.path.join(args.result_folder,
args.file_template % (
seed,
num_p))
try:
results = read_method_result(res_file, "spinn")
except Exception:
continue
for _, mse, _ in results:
all_results["mse"].append(float(mse))
all_results["seed"].append(seed)
all_results["num_p"].append(int(num_p))
results_df = pd.DataFrame(all_results)
#print("RESULTS")
#print(results_df)
results_df["log_num_p"] = np.log(results_df["num_p"])
#X = results_df["log_num_p"].reshape(-1,1)
X = results_df[["num_p", "log_num_p"]]#.reshape(-1,1)
y = results_df["mse"]
ols = sm.OLS(y, sm.add_constant(X))
ols_results = ols.fit()
print(ols_results.summary())
print("covariance", ols_results.cov_HC0)
print('Parameters: ', ols_results.params)
print('R2: ', ols_results.rsquared)
plt.clf()
sns_plt = sns.regplot(
x="log_num_p",
y="mse",
data=results_df,
x_jitter=0.05,
scatter_kws={'alpha':0.3},
lowess=True)
plt.xticks(size=14)
plt.yticks(size=14)
plt.xlabel("Log p", size=16)
plt.ylabel("Excess loss", size=16)
plt.tight_layout()
plt.savefig(os.path.join(args.result_folder, args.out_plot))
def plot_weights(args):
all_results = {
"relevant": [],
"weight": [],
"num_p": [],
"seed": []}
for seed in args.seeds:
for num_p in args.num_ps:
res_file = os.path.join(args.result_folder,
args.spinn_template % (
seed,
num_p))
try:
with open(res_file, "rb") as f:
spinn_res = pickle.load(f)
except Exception:
print(res_file)
continue
num_p = spinn_res["model_params"].coefs[0].shape[0]
norm_nonzero_inputs = [
(args.lasso_ratio * np.linalg.norm(spinn_res["model_params"].coefs[0][i,:], ord=1)
+ np.linalg.norm(spinn_res["model_params"].coefs[0][i,:], ord=2))
for i in range(num_p)
]
results = [
[True, np.sum(norm_nonzero_inputs[:args.max_relevant_idx])],
[False, np.sum(norm_nonzero_inputs[args.max_relevant_idx:])],
]
for relevant, mean_weight in results:
all_results["relevant"].append(relevant)
all_results["weight"].append(mean_weight)
all_results["seed"].append(seed)
all_results["num_p"].append(int(num_p))
results_df = pd.DataFrame(all_results)
#print("RESULTS")
#print(results_df)
results_df["sqrt_log_num_p"] = np.sqrt(np.log(results_df["num_p"]))
results_df["log_num_p"] = np.log(results_df["num_p"])
irrev_results_df = results_df.loc[np.logical_not(results_df["relevant"]),:]
X = irrev_results_df[["num_p", "log_num_p"]]#.reshape(-1,1)
y = irrev_results_df["weight"]
ols = sm.OLS(y, sm.add_constant(X))
ols_results = ols.fit()
print(ols_results.summary())
print("covariance", ols_results.cov_HC0)
print('Parameters: ', ols_results.params)
print('R2: ', ols_results.rsquared)
plt.clf()
sns_plt = sns.regplot(
x="sqrt_log_num_p",
y="weight",
data=irrev_results_df,
x_jitter=0.01,
scatter_kws={'alpha':0.3},
lowess=True)
plt.xticks(size=14)
plt.yticks(size=14)
plt.xlabel("Sqrt(Log p)", size=16)
plt.ylabel("SGL of irrelevant weights", size=16)
plt.tight_layout()
plt.savefig(os.path.join(args.result_folder, args.out_weight_plot))
def main(args=sys.argv[1:]):
args = parse_args(args)
plot_mse(args)
plot_weights(args)
if __name__ == "__main__":
main()