-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_SR_discrim.py
156 lines (121 loc) · 6.81 KB
/
run_SR_discrim.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
import argparse
import numpy as np
from helpers.Classifier import *
import torch
import os
import sys
from sklearn.metrics import roc_auc_score
import argparse
# Initialize parser
parser = argparse.ArgumentParser()
# Adding optional argument
parser.add_argument("-cu", "--cuda_slot", help = "cuda_slot")
parser.add_argument("-n", "--classifier_runs", help = "classifier_runs", default=10)
parser.add_argument("-i","--indir",help="home folder",default="/global/cfs/cdirs/m3246/rmastand/bkg_extrap/redo/")
parser.add_argument("-s","--signal",default=None,help="signal fraction",)
parser.add_argument("-c","--config",help="BC config file",default="configs/bc_discrim.yml")
parser.add_argument("-g","--gen_seed",help="Random seed for signal injections",default=1)
parser.add_argument("-full_sup",action='store_true',help="Run fully supervised case")
parser.add_argument("-ideal",action='store_true',help="Run idealized classifier")
parser.add_argument("-reweight",action='store_true',help="Run Reweight method")
parser.add_argument("-generate",action='store_true',help="Run Generate method")
parser.add_argument("-morph",action='store_true',help="Run Morph method")
# Read arguments from command line
args = parser.parse_args()
os.environ["CUDA_VISIBLE_DEVICES"]= str(args.cuda_slot)
def regularize_weights(w_arr, sigma = 3.0):
w_copy = np.copy(w_arr)
mean_w = np.mean(w_copy)
std_w = np.std(w_copy)
w_copy[w_copy > (sigma*std_w + mean_w)] = 0
return w_copy
def run_eval(set_1, set_2, code, save_dir, classifier_params, device, w_1 = None, w_2 = None, run_test = False, test_B = None, test_S = None, crop_weights = True):
if w_1 is None:
w_1 = np.array([1.]*set_1.shape[0])
if w_2 is None:
w_2 = np.array([1.]*set_2.shape[0])
if crop_weights:
w_1 = regularize_weights(w_1)
w_2 = regularize_weights(w_2)
input_x_train = np.concatenate([set_1, set_2], axis=0)
input_y_train = np.concatenate([np.zeros(set_1.shape[0]).reshape(-1,1), np.ones(set_2.shape[0]).reshape(-1,1)], axis=0)
input_w_train = np.concatenate([w_1, w_2], axis=0).reshape(-1, 1)
print(f"Working on {code}...")
print(" X train, y train, w train:", input_x_train.shape, input_y_train.shape, input_w_train.shape)
if run_test:
input_x_test = np.concatenate([test_B, test_S], axis=0)
input_y_test = np.concatenate([np.zeros(test_B.shape[0]).reshape(-1,1), np.ones(test_S.shape[0]).reshape(-1,1)], axis=0)
print(" X test, y test:", input_x_test.shape, input_y_test.shape)
for i in range(int(args.classifier_runs)):
print(f"Classifier run {i+1} of {args.classifier_runs}.")
local_id = f"{code}_run{i}"
# train classifier
NN = Classifier(n_inputs=5, layers=classifier_params["layers"], learning_rate=classifier_params["learning_rate"], device=device, scale_data=False)
NN.train(input_x_train, input_y_train, weights=input_w_train, save_model=True, model_name = f"model_{local_id}" , n_epochs=classifier_params["n_epochs"], seed = i, outdir=save_dir, plot_loss=False)
if run_test:
scores = NN.evaluation(input_x_test)
auc = roc_auc_score(input_y_test, scores)
if auc < 0.5: auc = 1.0 - auc
print(f" AUC: {auc}")
print()
def main():
# selecting appropriate device
CUDA = torch.cuda.is_available()
print("cuda available:", CUDA)
device = torch.device("cuda" if CUDA else "cpu")
static_data_dir = f"{args.indir}/data/"
seeded_data_dir = f"{args.indir}/data/seed{args.gen_seed}/"
samples_dir = f"{args.indir}/samples/seed{args.gen_seed}/"
eval_dir = f"{args.indir}/evaluation/seed{args.gen_seed}/"
os.makedirs(eval_dir, exist_ok=True)
# Load in the classifier params
with open(args.config, 'r') as stream:
params = yaml.safe_load(stream)
n_context = 2
# load in the test sets
test_events = np.load(f"{static_data_dir}/test_SR.npz")
test_bkg = test_events["bkg_events_SR"][:,n_context:]
test_sig = test_events["sig_events_SR"][:,n_context:]
path_to_data = f"{seeded_data_dir}/data_{args.signal}.npz"
if args.full_sup:
full_sup_events = np.load(f"{static_data_dir}/fullsup_SR.npz")
set_1 = full_sup_events["bkg_events_SR"][:,n_context:]
set_2 = full_sup_events["sig_events_SR"][:,n_context:]
run_eval(set_1, set_2, code="full_sup", save_dir=eval_dir, classifier_params=params, device=device, run_test=True, test_B=test_bkg, test_S=test_sig)
print()
if args.ideal:
ideal_bkg_events = np.load(f"{static_data_dir}/ideal_bkg_events.npz")
data_events = np.load(path_to_data)
set_1 = ideal_bkg_events["ideal_bkg_events_sr"][:,n_context:]
set_2 = data_events["data_events_sr"][:,n_context:]
run_eval(set_1, set_2, code=f"ideal_s{args.signal}", save_dir=eval_dir, classifier_params=params, device=device, run_test=False, test_B=test_bkg, test_S=test_sig)
print()
if args.reweight:
reweight_events = np.load(f"{samples_dir}/reweight_SR_s{args.signal}.npz")
data_events = np.load(path_to_data)
set_1 = reweight_events["mc_samples"][:,n_context:]
w_1 = reweight_events["w_sr"]
set_2 = data_events["data_events_sr"][:,n_context:]
run_eval(set_1, set_2, w_1 = w_1, code=f"reweight_s{args.signal}", save_dir=eval_dir, classifier_params=params, device=device, run_test=False, test_B=test_bkg, test_S=test_sig, crop_weights=True)
print()
if args.generate:
generate_events = np.load(f"{samples_dir}/generate_SR_s{args.signal}.npz")
context_weights = np.load(f"{samples_dir}/context_weights_SR_s{args.signal}.npz")
data_events = np.load(path_to_data)
set_1 = generate_events["samples"]
w_1 = context_weights["w_sr"]
set_2 = data_events["data_events_sr"][:,n_context:]
run_eval(set_1, set_2, w_1 = w_1, code=f"generate_s{args.signal}", save_dir=eval_dir, classifier_params=params, device=device, run_test=False, test_B=test_bkg, test_S=test_sig, crop_weights=True)
print()
if args.morph:
morph_events = np.load(f"{samples_dir}/morph_SR_s{args.signal}.npz")
context_weights = np.load(f"{samples_dir}/context_weights_SR_s{args.signal}.npz")
data_events = np.load(path_to_data)
set_1 = morph_events["samples"]
w_1 = context_weights["w_sr"]
set_2 = data_events["data_events_sr"][:,n_context:]
run_eval(set_1, set_2, w_1 = w_1, code=f"morph_s{args.signal}", save_dir=eval_dir, classifier_params=params, device=device, run_test=False, test_B=test_bkg, test_S=test_sig, crop_weights=True)
print()
print("All done!")
if __name__ == "__main__":
main()