-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathrun_SALAD.py
221 lines (158 loc) · 7.38 KB
/
run_SALAD.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
# Imports
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib import gridspec
from helpers.composite_helpers import *
from helpers.datasets import *
from sklearn.model_selection import train_test_split
from sklearn.metrics import roc_curve, roc_auc_score
from sklearn.utils import shuffle
import tensorflow as tf
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Lambda, Dense, Flatten
from keras import backend as K
from scipy import interpolate
import math
import yaml
with open("workflow.yaml", "r") as file:
workflow = yaml.safe_load(file)
import os
os.environ["CUDA_VISIBLE_DEVICES"]="2"
seed = 2
num_signal_to_inject = 3000
project_id = workflow["project_id"]
bands_dict = workflow["bands_dict"]
np.random.seed(seed)
tf.random.set_seed(seed)
gpu_options = tf.compat.v1.GPUOptions(per_process_gpu_memory_fraction=0.5)
sess = tf.compat.v1.Session(config=tf.compat.v1.ConfigProto(gpu_options=gpu_options))
# directory setup
n_features = 5 # Note that the SALAD network will also use the mass feature
feta_dir = "/global/home/users/rrmastandrea/FETA/"
dataset_config_string = f"LHCO_{num_signal_to_inject}sig_{project_id}"
path_to_minmax = f"{feta_dir}/LHCO_STS_{project_id}/data/col_minmax.npy"
col_minmax = np.load(path_to_minmax)
exp_dir = os.path.join(feta_dir, dataset_config_string)
#data_dir = os.path.join(exp_dir, "data")
data_dir = f"/global/ml4hep/spss/rrmastandrea/synthsamp_LHCOinput_{project_id}/nsig_{num_signal_to_inject}/data/"
"""
"
"
"
BEGIN HELPER FUNCTIONS
"
"
"
"""
def get_lhco_loc(filepath):
print(filepath)
df = pd.read_hdf(filepath)
# make_slim(df, directory, lhco_filename)
# Reorder the features such that the jets are ordered according to their invariant masses
jet_order_mask = df['mj1'] < df['mj2']
inverted_keys = ['pxj2', 'pyj2', 'pzj2', 'mj2', 'tau1j2', 'tau2j2', 'tau3j2', 'pxj1', 'pyj1', 'pzj1', 'mj1',
'tau1j1', 'tau2j1', 'tau3j1']
proper_order = df.loc[jet_order_mask]
improper_order = df.loc[~jet_order_mask]
improper_order.columns = inverted_keys
df = pd.concat((proper_order, improper_order))
for jet in ['j1', 'j2']:
df[f'pt{jet}'] = np.sqrt(df[f'px{jet}'] ** 2 + df[f'py{jet}'] ** 2)
df[f'eta{jet}'] = np.arcsinh(df[f'pz{jet}'] / df[f'pt{jet}'])
df[f'phi{jet}'] = np.arctan2(df[f'py{jet}'], df[f'px{jet}'])
df[f'p{jet}'] = np.sqrt(df[f'pz{jet}'] ** 2 + df[f'pt{jet}'] ** 2)
df[f'e{jet}'] = np.sqrt(df[f'm{jet}'] ** 2 + df[f'p{jet}'] ** 2)
data = df[['mj1', 'mj2']].copy()
data['mj2-mj1'] = data['mj2'] - data['mj1']
data[r'$\tau_{21}^{j_1}$'] = df['tau2j1'] / df['tau1j1']
data[r'$\tau_{32}^{j_1}$'] = df['tau3j1'] / df['tau2j1']
data[r'$\tau_{21}^{j_2}$'] = df['tau2j2'] / df['tau1j2']
data[r'$\tau_{32}^{j_2}$'] = df['tau3j2'] / df['tau2j2']
# data = pd.DataFrame()
data[r'$p_t^{j_1}$'] = df['ptj1']
data[r'$p_t^{j_2}$'] = df['ptj2']
phi_1 = df['phij1']
phi_2 = df['phij2']
delPhi = np.arctan2(np.sin(phi_1 - phi_2), np.cos(phi_1 - phi_2))
data[r'$dR_{jj}$'] = ((df['etaj1'] - df['etaj2']) ** 2 + delPhi ** 2) ** (0.5)
data['delPhi'] = abs(delPhi)
data['delEta'] = abs(df['etaj1'] - df['etaj2'])
data['mjj'] = calculate_mass(
np.sum([df[[f'ej{i}', f'pxj{i}', f'pyj{i}', f'pzj{i}']].to_numpy() for i in range(1, 3)], 0))
return data.dropna()
"""
"
"
"
END HELPER FUNCTIONS
"
"
"
"""
new_herwig_samps_path = "/clusterfs/ml4hep/rrmastandrea/LHCO/events_anomalydetection_DelphesHerwig_qcd_extra_inneronly_features.h5"
df_new_herwig = get_lhco_loc(new_herwig_samps_path)
data_herwig_sr = select_lhco_features(df_new_herwig, feature_type = 0).to_numpy()
"""
"""
"""
LOAD IN THE DATASETS AND PROCESS
"""
"""
"""
npull_dataset_train_sim = ToyDataset(data_dir, "train_sim.npy")
npull_dataset_val_sim = ToyDataset(data_dir, "val_sim.npy")
npull_dataset_train_dat = ToyDataset(data_dir, "train_dat.npy")
npull_dataset_val_dat = ToyDataset(data_dir, "val_dat.npy")
print("Num SIM events in SB:", len(npull_dataset_train_sim)+len(npull_dataset_val_sim))
print("Num DAT events in SB:", len(npull_dataset_train_dat)+len(npull_dataset_val_dat))
print()
# Preprocess the data
print("Preproccessing data...")
print()
dataset_train_sim = npull_dataset_train_sim.pull_from_mass_range([bands_dict["sb1"], bands_dict["sb2"]])
dataset_val_sim = npull_dataset_val_sim.pull_from_mass_range([bands_dict["sb1"], bands_dict["sb2"]])
dataset_train_dat = npull_dataset_train_dat.pull_from_mass_range([bands_dict["sb1"], bands_dict["sb2"]])
dataset_val_dat = npull_dataset_val_dat.pull_from_mass_range([bands_dict["sb1"], bands_dict["sb2"]])
dataset_train_sim = minmaxscale(dataset_train_sim.data, col_minmax, lower = 0, upper = 1, forward = True)
dataset_val_sim = minmaxscale(dataset_val_sim.data, col_minmax, lower = 0, upper = 1, forward = True)
dataset_train_dat = minmaxscale(dataset_train_dat.data, col_minmax, lower = 0, upper = 1, forward = True)
dataset_val_dat = minmaxscale(dataset_val_dat.data, col_minmax, lower = 0, upper = 1, forward = True)
dataset_sr_sim = minmaxscale(data_herwig_sr, col_minmax, lower = 0, upper = 1, forward = True)
# RUN SALAD
def get_weights(data, model):
yhat = model.predict(data, batch_size=128)
return np.squeeze(yhat/(1 - yhat))
stored_weights = {}
X_SALAD_sb_train = np.concatenate([dataset_train_sim, dataset_train_dat])
Y_SALAD_sb_train = np.concatenate([np.zeros(len(dataset_train_sim)), np.ones(len(dataset_train_dat))])
X_SALAD_val = np.concatenate([dataset_val_sim, dataset_val_dat])
Y_SALAD_val = np.concatenate([np.zeros(len(dataset_val_sim)),np.ones(len(dataset_val_dat))])
print('Training SALAD model...')
tf.keras.backend.clear_session()
model_SALAD_sb = Sequential()
model_SALAD_sb.add(Dense(100, input_dim=n_features+1, activation='relu'))
model_SALAD_sb.add(Dense(100, activation='relu'))
model_SALAD_sb.add(Dense(100, activation='relu'))
model_SALAD_sb.add(Dense(1, activation='sigmoid'))
model_SALAD_sb.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
EPOCHS = 50
PATIENCE = 10
BATCH_SIZE = 256
earlyStopping = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=PATIENCE, verbose=0, mode='min')
mcp_save = tf.keras.callbacks.ModelCheckpoint(f'SALAD_models_{project_id}/{num_signal_to_inject}.mdl_wts.hdf5', save_best_only=True, monitor='val_loss', mode='min')
reduce_lr_loss = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=PATIENCE, verbose=1, epsilon=1e-7, mode='min')
hist_SALAD_sb = model_SALAD_sb.fit(
X_SALAD_sb_train, Y_SALAD_sb_train, epochs=EPOCHS,
batch_size=int(BATCH_SIZE), verbose=0, callbacks=[earlyStopping, mcp_save, reduce_lr_loss], validation_data=(X_SALAD_val, Y_SALAD_val)
)
print("Done training!")
# Load in the best weights
model_SALAD_sb.load_weights(f'SALAD_models_{project_id}/{num_signal_to_inject}.mdl_wts.hdf5')
print("Evaluating at best val loss epoch:", np.argmin(hist_SALAD_sb.history['val_loss']))
plot_weights = get_weights(dataset_sr_sim, model_SALAD_sb)
# save out
#scaled_data_dir = f"/global/home/users/rrmastandrea/scaled_data_{project_id}_seed_{seed}/"
scaled_data_dir = f"/global/ml4hep/spss/rrmastandrea/synth_SM_AD/scaled_data_{project_id}_seed_{seed}/"
np.save(f"{scaled_data_dir}/nsig_injected_{num_signal_to_inject}/salad", dataset_sr_sim)
np.save(f"{scaled_data_dir}/nsig_injected_{num_signal_to_inject}/salad_weights", plot_weights)