forked from ZhenxuanZhang-Jensen/HHRecoNN
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_RECONN.py
224 lines (203 loc) · 10.7 KB
/
train_RECONN.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
import tensorflow as tf
import pandas as pd
import numpy as np
import os, json, argparse, tqdm, sklearn, plotting
import matplotlib.pyplot as plt
from plotting.plotter import plotter
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder
from sklearn.utils import class_weight
from sklearn.metrics import roc_curve, roc_auc_score, auc, confusion_matrix
from random import seed
from random import random
from random import randint
from os import environ
os.environ['KERAS_BACKEND'] = 'tensorflow'
import keras
from keras import backend as K
from keras.models import Sequential
from keras.layers.core import Dense
from keras.layers.core import Activation
from keras.optimizers import Nadam
from keras.callbacks import EarlyStopping
def compile_model(num_variables,learn_rate=0.001):
model = Sequential()
model.add(Dense(num_variables, input_dim=num_variables))
model.add(Activation('relu'))
model.add(Dense(32))
model.add(Activation('relu'))
model.add(Dense(32))
model.add(Activation('relu'))
model.add(Dense(16))
model.add(Activation('relu'))
model.add(Dense(8))
model.add(Activation('relu'))
model.add(Dense(4))
model.add(Activation('relu'))
model.add(Dense(1, activation="sigmoid"))
optimizer=Nadam(lr=learn_rate)
model.compile(loss='binary_crossentropy',optimizer=optimizer,metrics=['acc'])
return model
def load_trained_model(model_path):
print('<load_trained_model> weights_path: ', model_path)
model = load_model(model_path, compile=False)
return model
def main():
usage = 'usage: %prog [options]'
parser = argparse.ArgumentParser(usage)
parser.add_argument('-i', '--input_dataframe', dest='input_dataframe', help='Path to input dataframe', default='', type=str)
parser.add_argument('-o', '--outputs', dest='outputdir', help='Name of output directory', default='test', type=str)
parser.add_argument('-f', '--fit_model', dest='fit_model', help='Flag: 0 to evaluate a pre-existing model, 1 to fit a new model', default=1, type=int)
args = parser.parse_args()
input_dataframe = args.input_dataframe
outputdir = args.outputdir
fit_model = args.fit_model
# Make instance of plotter tool
Plotter = plotter()
if os.path.isdir(outputdir) != 1:
os.mkdir(outputdir)
plots_dir = os.path.join(outputdir,'plots/')
if os.path.isdir(plots_dir) != 1:
os.mkdir(plots_dir)
# Create dataset from .csv
event_data = pd.read_csv(input_dataframe)
print("Input Dataframe: ", event_data.shape)
input_columns_training = event_data.columns[:-3]
#column_headers = ['HW1_jet1_pt','HW1_jet1_eta','HW1_jet1_phi','HW1_jet1_E','HW1_jet2_pt','HW1_jet2_eta','HW1_jet2_phi','HW1_jet2_E','HW1_jet3_pt','HW1_jet3_eta','HW1_jet3_phi','HW1_jet3_E','HW1_jet4_pt','HW1_jet4_eta','HW1_jet4_phi','HW1_jet4_E','target']
#column_headers = ['HW1_jet1_pt','HW1_jet1_eta','dRj1_photon1','dRj1_photon2','HW1_jet2_pt','HW1_jet2_eta','dRj2_photon1','dRj2_photon2','HW1_jet3_pt','HW1_jet3_eta','dRj3_photon1','dRj3_photon2','HW1_jet4_pt','HW1_jet4_eta','dRj4_photon1','dRj4_photon2','target','event_ID']
model_output = os.path.join(outputdir,'model/')
if fit_model == 1:
traindataset, valdataset = train_test_split(event_data, test_size=0.1)
print('Using columns: ', input_columns_training.values)
print('Using labels: ', event_data['target'].values)
print('Using event_ID: ', event_data['event_ID'].values)
print('Using njets: ', event_data['njets'].values)
train_input_ = traindataset[input_columns_training].values
train_target_ = traindataset['target'].values
test_input_ = valdataset[input_columns_training].values
test_target_ = valdataset['target'].values
# Event weights if wanted
train_weights = np.ones(len(traindataset['target']))
print('train_weights: ' , train_weights)
test_weights = np.ones(len(valdataset['target']))
labels = np.array(event_data['target'])
print(labels)
class_weights = np.array(class_weight.compute_class_weight(class_weight="balanced", classes=np.unique(labels), y=labels ))
class_weights = dict(enumerate(class_weights))
print('class_weights: ', class_weights)
# Fit label encoder to Y_train
newencoder = LabelEncoder()
newencoder.fit(train_target_)
# Transform to encoded array
encoded_Y = newencoder.transform(train_target_)
encoded_Y_test = newencoder.transform(test_target_)
histories = []
labels = []
print(train_input_)
# Fitting the model
early_stopping_monitor = EarlyStopping(patience=100, monitor='val_loss', min_delta=0.01, verbose=1)
model = compile_model(len(input_columns_training), learn_rate=0.001)
history_ = model.fit(train_input_,train_target_,validation_split=0.1,class_weight=class_weights,epochs=200,batch_size=512,verbose=1,shuffle=True,callbacks=[early_stopping_monitor])
# Store model in file
model_output_name = os.path.join(model_output)
model.save(model_output_name)
weights_output_name = os.path.join(model_output,'model_weights.h5')
model.save_weights(weights_output_name)
model_json = model.to_json()
model_json_name = os.path.join(model_output,'model_serialised.json')
with open(model_json_name,'w') as json_file:
json_file.write(model_json)
train_pred_ = model.predict(np.array(train_input_))
test_pred_ = model.predict(np.array(test_input_))
prediction_data = []
for entry in range(0,len(test_pred_)):
prediction_data.append( [test_pred_[entry],test_target_] )
df = pd.DataFrame(prediction_data)
df.columns = ['prediction','label']
df.to_csv(os.path.join(outputdir,"predictions_dataframe.csv"), index=False)
# Make instance of plotter tool (need for DNN response plots)
Plotter = plotter()
# Initialise output directory.
Plotter.plots_directory = os.path.join(plots_dir,'perf')
Plotter.output_directory = outputdir
# History plots
Plotter.history_plot(history_, label='loss')
Plotter.save_plots(dir=Plotter.plots_directory, filename='history_loss.png')
# ROC curves
Plotter.ROC(train_target_,train_pred_,test_target_,test_pred_)
Plotter.save_plots(dir=Plotter.plots_directory, filename='ROC.png')
# Make overfitting plots of output nodes
Plotter.binary_overfitting(model, train_target_, test_target_, train_pred_, test_pred_, train_weights, test_weights)
Plotter.save_plots(dir=Plotter.plots_directory, filename='response.png')
elif fit_model == 0:
model_name = os.path.join(model_output)
model = keras.models.load_model(model_name)
evaluation_inputs_ = event_data[input_columns_training].values
evaluation_targets_ = event_data['target'].values
evaluation_IDs_ = event_data['event_ID'].values
event_njets = event_data['njets'].values
predictions_ = model.predict(np.array(evaluation_inputs_))
previous_evID=0
max_DNN_score = -9
truth_label_of_max_score = -9
correct_predictions = 0
incorrect_predictions = 0
correct_4jet_predictions = 0
incorrect_4jet_predictions = 0
correct_5jet_predictions = 0
incorrect_5jet_predictions = 0
correct_6jet_predictions = 0
incorrect_6jet_predictions = 0
N_4jet_examples = 0
N_5jet_examples = 0
N_6jet_examples = 0
for evID_index in range(0,len(evaluation_IDs_)):
# If first event
if evID_index == 0:
previous_evID = evaluation_IDs_[evID_index]
if event_njets[evID_index] == 4 and evaluation_targets_[evID_index] == 0:
print('WARNING: 4 jet event with target label == 0 ')
print('Event: ', evaluation_IDs_[evID_index-1])
print('NJets: ', event_njets[evID_index-1])
print('Target label: ', evaluation_targets_[evID_index-1])
# Elif same event ID as previous event
elif evaluation_IDs_[evID_index] == previous_evID:
# If prediction has largest response so far
if predictions_[evID_index] > max_DNN_score:
max_DNN_score = predictions_[evID_index]
truth_label_of_max_score = evaluation_targets_[evID_index]
# Else we check the result for the previous event and start a new event
elif evaluation_IDs_[evID_index] != previous_evID:
if event_njets[evID_index-1] == 4:
N_4jet_examples+=1
if truth_label_of_max_score == 1 :
correct_4jet_predictions += 1
else:
incorrect_4jet_predictions += 1
if event_njets[evID_index-1] == 5:
N_5jet_examples+=1
if truth_label_of_max_score == 1 :
correct_5jet_predictions += 1
else:
incorrect_5jet_predictions += 1
if event_njets[evID_index-1] == 6:
N_6jet_examples+=1
if truth_label_of_max_score == 1 :
correct_6jet_predictions += 1
else:
incorrect_6jet_predictions += 1
# If the DNN score a signal permutation (target == 1) as the highest -> correct
if truth_label_of_max_score == 1 :
correct_predictions += 1
# Otherwise incorrect
else:
incorrect_predictions += 1
previous_evID = evaluation_IDs_[evID_index]
max_DNN_score = predictions_[evID_index]
print('Dataset contained %s 4-jet, %s 5-jet, %s 6-jet events' %(N_4jet_examples,N_5jet_examples,N_6jet_examples))
print('# 4-jet correct: %s, # 4-jet incorrect: %s (%s percent)' % (correct_4jet_predictions,incorrect_4jet_predictions,(correct_4jet_predictions/(correct_4jet_predictions+incorrect_4jet_predictions))))
print('# 5-jet correct: %s, # 5-jet incorrect: %s (%s percent)' % (correct_5jet_predictions,incorrect_5jet_predictions,(correct_5jet_predictions/(correct_5jet_predictions+incorrect_5jet_predictions))))
print('# 6-jet correct: %s, # 6-jet incorrect: %s (%s percent)' % (correct_6jet_predictions,incorrect_6jet_predictions,(correct_6jet_predictions/(correct_6jet_predictions+incorrect_6jet_predictions))))
print('# total correct: %s, # total incorrect %s (%s percent)' % (correct_predictions,incorrect_predictions,(correct_predictions/(correct_predictions+incorrect_predictions))))
exit(0)
main()