-
Notifications
You must be signed in to change notification settings - Fork 1
/
experiment.py
312 lines (221 loc) · 12.5 KB
/
experiment.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
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from sklearn_quantile import RandomForestQuantileRegressor, SampleRandomForestQuantileRegressor, KNeighborsQuantileRegressor
from sklearn.ensemble import RandomForestRegressor, RandomForestClassifier
from sklearn.base import clone
from sklearn.metrics import f1_score, balanced_accuracy_score
from scipy.stats import norm, uniform, beta, iqr, mode
import math
from functools import partial
from sklearn.model_selection import RandomizedSearchCV
from sklearn.linear_model import QuantileRegressor, LinearRegression
import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim.lr_scheduler import StepLR
import copy
from scipy.special import expit, logit
from lightgbm import LGBMRegressor
from scipy.spatial.distance import cdist
from helper import generate_data, interval_score_loss, randomized_conformal_cutoffs, select_column_per_row
from uacqr import uacqr
class experiment():
def __init__(self, fixed_model_params=None, n=None, p=None, T=None, X = None, y = None, cond_exp=None, noise_sd_fn=None, x_dist=None,
S=25, random_state=42, var_name = 'min_samples_leaf', var_list = [1,5], metric='interval_score_loss',
fast_uacqr=True, B=100, model_type='rfqr', uacqrs_agg='std', empirical_data_fraction=0.01,
oracle_g=None, oracle_t=None, inject_noise=False, randomized_conformal=False,
extraneous_quantiles=[0.1, 0.25, 0.4, 0.5, 0.6, 0.75,0.9],
uacqrs_bagging=False, sorting_column=0, max_normalization=False,
local_metric=False, oqr_metrics=False, file_name=None):
self.var_name = var_name
self.metric = metric
self.S = S
self.fast_uacqr = fast_uacqr
self.sorting_column = sorting_column
self.local_metric = local_metric
if self.metric == 'conditional_coverage':
self.local_metric = True
if file_name:
self.results_df = pd.read_pickle(file_name)
if len(self.results_df.columns) == 5:
# self.metric = 'conditional_coverage'
self.local_metric = True
self.S = self.results_df.index.value_counts().mode()[0]
return
self.var_name = self.results_df.columns[2]
self.S = int(self.results_df['alpha'].count() / self.results_df.nunique()[self.var_name])
return
col_names = ['alpha', 'cqr_score_threshold', 'uacqrs_score_threshold', 'uacqrp_test_coverage',
'cqr_test_coverage', 'uacqrs_test_coverage', 'median_test_coverage', 'uacqrp_average_length_test',
'cqr_average_length_test', 'uacqrs_average_length_test', 'cqrr_average_length_test',
'median_average_length_test', 'uacqrp_test_len_std',
'cqr_test_len_std', 'median_test_len_std', 'uacqrp_interval_score_loss', 'cqr_interval_score_loss',
'uacqrs_interval_score_loss', 'cqrr_interval_score_loss','median_interval_score_loss',
'base_average_length_test','base_test_coverage','cqrr_test_coverage',
'base_oqr_corr', 'uacqrp_oqr_corr','uacqrs_oqr_corr','cqr_oqr_corr','cqrr_oqr_corr',
'base_oqr_hsic', 'uacqrp_oqr_hsic','uacqrs_oqr_hsic','cqr_oqr_hsic','cqrr_oqr_hsic',
'base_oqr_wsc', 'uacqrp_oqr_wsc','uacqrs_oqr_wsc','cqr_oqr_wsc','cqrr_oqr_wsc',var_name]
df = pd.DataFrame(columns=col_names)
t=0
df_list = []
for var in var_list:
sim_model_params = fixed_model_params.copy()
sim_model_params[var_name] = var
if var_name == 'p':
p = var
del sim_model_params['p']
for s in range(random_state, random_state+S):
np.random.seed(s)
q_lower = 5
q_upper = 95
alpha = 0.1
if X is not None:
x_train, y_train, x_calib, y_calib, x_test, y_test = self.empirical_data_draw(X, y, empirical_data_fraction, s,
max_normalization)
elif cond_exp is not None:
x_train, y_train, x_calib, y_calib, x_test, y_test = self.simulate_data(cond_exp, noise_sd_fn, x_dist, n, p, T)
self.p=p
else:
raise Exception("Need to either provide existing data or specify a DGP")
uacqr_results = uacqr(model_params=sim_model_params,
bootstrapping_for_uacqrp=not(fast_uacqr), B=B, random_state=s, model_type=model_type, uacqrs_agg=uacqrs_agg,
oracle_g=oracle_g, randomized_conformal=randomized_conformal, extraneous_quantiles=extraneous_quantiles,
uacqrs_bagging=uacqrs_bagging)
uacqr_results.fit(x_train, y_train)
if oracle_g:
uacqr_results.calibrate(x_calib, y_calib, inject_noise=inject_noise, cond_exp=cond_exp, noise_sd_fn=noise_sd_fn)
else:
uacqr_results.calibrate(x_calib, y_calib, inject_noise=inject_noise)
uacqr_results.evaluate(x_test, y_test, oqr_metrics=oqr_metrics)
if self.local_metric:
uacqr_results.calculate_conditional_coverage(cond_exp, noise_sd_fn, plot=False, sorting_column=sorting_column,
metric=metric)
df_list.append(uacqr_results.conditional_coverage)
else:
single_run_results = vars(uacqr_results)
df.loc[t] = {k: v for k, v in single_run_results.items() if isinstance(v,(float,np.floating))}
df.loc[t,var_name] = var
t+= 1
if t % 5 ==0:
print(t)
if self.local_metric:
self.results_df = pd.concat(df_list)
else:
df = df.melt(id_vars=var_name,ignore_index=False)
df[['method','metric']] = df.variable.str.split('_',n=1, expand=True)
df.drop('variable',axis=1, inplace=True)
df.reset_index(inplace=True)
df = df.pivot(index=['index','metric', var_name],columns='method', values='value').reset_index()
df.rename(columns={"cqr":"CQR","uacqrs":"UACQR-S","uacqrp":"UACQR-P", "cqrr":"CQR-r"}, inplace=True)
self.results_df = df
self.last_run = uacqr_results
self.results_df['CQR'] = pd.to_numeric(self.results_df['CQR'])
self.results_df['UACQR-S'] = pd.to_numeric(self.results_df['UACQR-S'])
self.results_df['UACQR-P'] = pd.to_numeric(self.results_df['UACQR-P'])
self.results_df['CQR-r'] = pd.to_numeric(self.results_df['CQR-r'])
def empirical_data_draw(self, X, y, empirical_data_fraction, s, max_normalization=False):
train_fraction = 0.4
test_fraction = 1/3
calib_fraction = 1
if not isinstance(X, pd.DataFrame):
X = pd.DataFrame(X)
y = pd.Series(y.reshape(-1))
X_drawn = X.sample(frac=empirical_data_fraction, random_state=s)
x_train = X_drawn.sample(frac=train_fraction, random_state=s)
x_test = X_drawn.drop(x_train.index).sample(frac=test_fraction, random_state=s)
x_calib = X_drawn.drop(x_train.index).drop(x_test.index).sample(frac=calib_fraction, random_state=s)
y_train = y.loc[x_train.index]
if max_normalization:
scaling = np.mean(np.abs(y_train))
else:
scaling = 1
y_train = y_train / scaling
y_test = y.loc[x_test.index] / scaling
y_calib = y.loc[x_calib.index] / scaling
return x_train,y_train,x_calib,y_calib,x_test,y_test
def simulate_data(self, cond_exp, noise_sd_fn, x_dist, n, p, T):
n0 = int(n/2)
n1 = n-n0-1
data = generate_data(n, p, cond_exp, noise_sd_fn, x_dist)
x = data[0]
# for r in range(x.shape[0]):
# if x[r,0] < 0:
# x[r,1:] = x[r,0]
y = data[1]
if len(x.shape)==1:
x = x.reshape(-1,1)
x_train = x[:n0]
y_train = y[:n0]
x_calib = x[n0:n0+n1]
y_calib = y[n0:n0+n1]
test = generate_data(T, p, cond_exp, noise_sd_fn, x_dist)
x_test = test[0]
# for r in range(x.shape[0]):
# if x_even[r,0] < 0:
# x_even[r,1:] = x_even[r,0]
y_test = test[1]
return x_train,y_train,x_calib,y_calib,x_test,y_test
def plot(self, metric='average_length_test', title_prefix=None, log_x=False, log_y=False, calc_only=False,
xlabel_conditional_coverage = "$X$", custom_title=None, ax=None, bigger_font=None, xlabel=None, ylabel=None):
if self.local_metric:
self.var_name = self.sorting_column
df = self.results_df.copy()
else:
df = self.results_df.loc[self.results_df.metric == metric, ['UACQR-P', 'UACQR-S','CQR', 'CQR-r',self.var_name]]
mean_results = df.groupby(self.var_name).mean()
sem_results = df.groupby(self.var_name).sem()
self.mean_results = mean_results
self.sem_results = sem_results
if calc_only:
return
if self.local_metric:
self.var_name = xlabel_conditional_coverage
if bigger_font:
plt.rcParams['font.size'] = bigger_font
else:
plt.style.use(['default'])
if not(ax):
fig, ax = plt.subplots()
if title_prefix:
mean_results.plot(title=title_prefix+': '+ metric +' vs '+self.var_name+' ('+str(self.S)+' iters)',
color=['tab:orange', 'tab:red','tab:green', 'tab:brown','tab:purple'], ax=ax)
elif custom_title:
mean_results.plot(title=custom_title,
color=['tab:orange', 'tab:red','tab:green', 'tab:brown','tab:purple'], ax=ax)
else:
method_name = 'Fast UACQR-P' if self.fast_uacqr else 'Regular UACQR-P'
mean_results.plot(title = method_name+': '+ metric +' vs '+self.var_name+' ('+str(self.S)+' iters)',
color=['tab:orange', 'tab:red', 'tab:green', 'tab:brown','tab:purple'], ax=ax)
ax.fill_between(sem_results.index, mean_results['UACQR-P'] - 1.96 * sem_results['UACQR-P'],
mean_results['UACQR-P']+ 1.96 * sem_results['UACQR-P'],
color = 'tab:orange', alpha = .1)
ax.fill_between(sem_results.index, mean_results['CQR'] - 1.96 * sem_results['CQR'],
mean_results['CQR'] + 1.96 * sem_results['CQR'],
color = 'tab:green', alpha = .1)
ax.fill_between(sem_results.index, mean_results['UACQR-S'] - 1.96 * sem_results['UACQR-S'],
mean_results['UACQR-S'] + 1.96 * sem_results['UACQR-S'],
color = 'tab:red', alpha = .1)
ax.fill_between(sem_results.index, mean_results['CQR-r'] - 1.96 * sem_results['CQR-r'],
mean_results['CQR-r'] + 1.96 * sem_results['CQR-r'],
color = 'tab:brown', alpha = .1)
if self.local_metric:
if metric == 'conditional_coverage':
ax.axhline(y = 0.9, color = 'tab:blue', linestyle='--', zorder=0)
ax.set_ylabel('Conditional Coverage')
ax.fill_between(sem_results.index, mean_results['Base Estimator'] - 1.96 * sem_results['Base Estimator'],
mean_results['Base Estimator'] + 1.96 * sem_results['Base Estimator'],
color = 'tab:purple', alpha = .1)
ax.set_xlabel(self.var_name)
if xlabel:
ax.set_xlabel(xlabel)
if ylabel:
ax.sex_ylabel(ylabel)
if log_x:
ax.xscale('log', base=2)
if log_y:
ax.yscale('log', base=2)
def save(self, file_name):
if not file_name.endswith('.pkl'):
file_name += '.pkl'
self.results_df.to_pickle(file_name)