-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathsplitDataset.py
169 lines (143 loc) · 9.01 KB
/
splitDataset.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
### This script takes as input the dataframes produced in the previous step (buildDataset), computes train weights to compensate for different
### statistics for each signal mass value and for signal/background, scales each feature according to median and interquartile range
### and splits the resulting dataframe in train and test samples.
### Histograms of scaled and unscaled variables will be saved if running with --drawPlots = 1
import seaborn
from Functions import ReadArgParser, checkCreateDir, ReadConfig, DrawVariablesHisto, ShufflingData, ComputeTrainWeights, ScalingFeatures, ComputeScaleFactors#, SaveFeatureScaling,
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
import sys
import shlex
from colorama import init, Fore
init(autoreset = True)
#print(sys.executable, ' '.join(map(shlex.quote, sys.argv)))
#from tensorflow.keras.utils import to_categorical
pd.options.mode.chained_assignment = None ### to suppress the SettingWithCopyWarning
### Reading from command line
tag, analysis, channel, preselectionCuts, background, signal, trainingFraction, drawPlots = ReadArgParser()
### Reading from configuration file
dfPath, signalsList, backgroundsList = ReadConfig(tag, analysis, signal) ### remove input features
### Loading input file
inputDir = dfPath + analysis + '/' + channel + '/' + preselectionCuts + '/' + signal + '/' + background
#inputDir = dfPath + analysis + '/' + channel + '/' + preselectionCuts + '/' + 'ggFandVBF/' + signal + '/' + background
#fileCommonName = tag + '_' + jetCollection + '_' + analysis + '_' + channel + '_' + preselectionCuts + '_' + signal + '_' + background
fileCommonName = tag + '_' + analysis + '_' + channel + '_' + preselectionCuts + '_' + signal + '_' + background
print(Fore.GREEN + 'Loading' + inputDir + '/MixData_' + fileCommonName + '.pkl')
data = pd.read_pickle(inputDir + '/MixData_' + fileCommonName + '.pkl')
### If not already existing, creating output directory
outputDir = inputDir# + '/PDNN_trainSet9'# + '/ggFsameStatAsVBF'
checkCreateDir(outputDir)
fileCommonName += '_' + str(trainingFraction) + 't'
### Creating log file
logFileName = '/logFile_splitDataset_' + fileCommonName + '.txt'
logFile = open(outputDir + logFileName, 'w')
logFile.write('Command executed: ' + sys.executable + ' '.join(map(shlex.quote, sys.argv)) + '\nTag: ' + tag + '\nAnalysis: ' + analysis + '\nChannel: ' + channel + '\nPreselection cuts: ' + preselectionCuts + '\nSignal: ' + signal + '\nBackground: ' + background)
### Creating the list of backgrounds and signal processes to select
if background == 'all':
inputOrigins = backgroundsList.copy()
backgroundLegend = backgroundsList.copy()
logFile.write(' (' + str(backgroundsList) + ')')
else:
inputOrigins = list(background.split('_'))
inputOrigins.append(signal)
### Selecting events according to their origin
data_set = data[data['origin'].isin(inputOrigins)]
### Dividing signal from background
dataSetSignal = data_set[data_set['origin'] == signal]
dataSetBackground = data_set[data_set['origin'] != signal]
massesSignalList = sorted(list(set(list(dataSetSignal['mass']))))
print(Fore.BLUE + 'Masses in the signal sample: ' + str(massesSignalList) + ' GeV (' + str(len(massesSignalList)) + ')')
logFile.write('\nMasses in the signal sample: ' + str(massesSignalList) + ' GeV (' + str(len(massesSignalList)) + ')')
'''
statDict = {}
statDict['Radion'] = {500: 2587, 600: 7075, 700: 10780, 800: 13074, 1000: 13241, 1200: 11704, 1400: 11451, 1500: 11811, 1600: 12185, 1800: 13458, 2000: 14873, 2400: 17913, 2600: 18750, 3000: 19480, 3500: 17941, 4000: 15948, 4500: 14542, 5000: 13232, 6000: 12604}
statDict['RSG'] = {500: 1622, 600: 5964, 700: 5964, 800: 7282, 1000: 7329, 1200: 7208, 1400: 7431, 1500: 7549, 1600: 7620, 1800: 8410, 2000: 9452, 2400: 11098, 2600: 11505, 3000: 11699, 3500: 11033, 4000: 9492, 4500: 8321, 5000: 7532, 6000: 7532}
statDict['HVTWZ'] = {500: 0, 600: 2735, 700: 4526, 800: 6452, 1000: 7293, 1200: 6853, 1400: 0, 1500: 6886, 1600: 0, 1800: 8544, 2000: 9401, 2400: 7708, 2600: 11763, 3000: 12711, 3500: 12100, 4000: 11888, 4500: 0, 5000: 0, 6000: 0}
statBkgDict = {}
statBkgDict['Radion'] = {'Zjets': 41611, 'Diboson': 2603, 'stop': 8, 'ttbar': 272, 'Wjets': 16}
statBkgDict['RSG'] = {'Zjets': 41292, 'Diboson': 2602, 'stop': 8, 'ttbar': 272, 'Wjets': 16}
statBkgDict['HVTWZ'] = {'Zjets': 41611, 'Diboson': 2603, 'stop': 8, 'ttbar': 272, 'Wjets': 16}
newDataSetSignal = []
for mass in massesSignalList:
dataSetSignalMass = dataSetSignal[dataSetSignal['mass'] == mass]
if mass not in statDict[signal]:
continue
statMass = statDict[signal][mass]
dataSetSignalMass = dataSetSignalMass[:statMass]
newDataSetSignal.append(dataSetSignalMass)
dataSetSignal = pd.concat(newDataSetSignal, ignore_index = True)
dataSetSignal = ShufflingData(dataSetSignal)
newDataSetBkg = []
for bkg in backgroundsList:
dataSetBkgOrigin = dataSetBackground[dataSetBackground['origin'] == bkg]
statBkg = statBkgDict[signal][bkg]
dataSetBkgOrigin = dataSetBkgOrigin[:statBkg]
newDataSetBkg.append(dataSetBkgOrigin)
dataSetBackground = pd.concat(newDataSetBkg, ignore_index = True)
dataSetBackground = ShufflingData(dataSetBackground)
print(dataSetSignal.shape)
print(dataSetBackground.shape)
'''
### Creating new column in the dataframes with train weight
dataSetSignal, dataSetBackground, logString = ComputeTrainWeights(dataSetSignal, dataSetBackground, massesSignalList, outputDir, fileCommonName, analysis, channel, signal, backgroundLegend, preselectionCuts, drawPlots)
if logString != '':
print(Fore.GREEN + logString)
logFile.write(logString)
### Printing and saving information on events numbers
stringForBkgEvents = 'Number of background events: ' + str(dataSetBackground.shape[0]) + ' raw, ' + str(dataSetBackground['weight'].sum()) + ' with MC weights, ' + str(dataSetBackground['train_weight'].sum()) + ' with train weights'
print(Fore.BLUE + stringForBkgEvents)
logFile.write('\n' + stringForBkgEvents)
if(len(backgroundsList) > 1):
for bkg in backgroundsList:
dataSetSingleBackground = dataSetBackground[dataSetBackground['origin'] == bkg]
stringToSaveBkg = ' ---> Number of ' + bkg + ' events: ' + str(dataSetSingleBackground.shape[0]) + ' raw, ' + str(dataSetSingleBackground['weight'].sum()) + ' with MC weights, ' + str(dataSetSingleBackground['train_weight'].sum()) + ' with train weights'
print(Fore.BLUE + stringToSaveBkg)
logFile.write('\n' + stringToSaveBkg)
stringForSignalEvents = 'Number of signal events: ' + str(dataSetSignal.shape[0]) + ' raw, ' + str(dataSetSignal['weight'].sum()) + ' with MC weights, ' + str(dataSetSignal['train_weight'].sum()) + ' with train weights'
print(Fore.BLUE + stringForSignalEvents)
logFile.write('\n' + stringForSignalEvents)
for signalMass in massesSignalList:
dataSetSignalMass = dataSetSignal[dataSetSignal['mass'] == signalMass]
stringToSaveSignal = ' ---> Number of signal events with mass ' + str(signalMass) + ' GeV: ' + str(dataSetSignalMass.shape[0]) + ' raw, ' + str(dataSetSignalMass['weight'].sum()) + ' with MC weights, ' + str(dataSetSignalMass['train_weight'].sum()) + ' with train weights'
print(Fore.BLUE + stringToSaveSignal)
logFile.write('\n' + stringToSaveSignal)
### Concatening signal and background dataframes
dataFrame = pd.concat([dataSetSignal, dataSetBackground], ignore_index = True)
### Shuffling the dataframe
dataFrame = ShufflingData(dataFrame)
### Creating a new column in the dataFrame that will store the unscaled mass
dataFrame = dataFrame.assign(unscaledMass = dataFrame['mass'])
### Splitting data into train and test set
data_train, data_test = train_test_split(dataFrame, train_size = trainingFraction)
'''
### Slicing data train for statistic test
nEvents = int(data_train.shape[0] / 4)
data_train = data_train[:nEvents]
'''
### Scaling InputFeatures of train and test set
logString = ComputeScaleFactors(data_train, outputDir)
'''
data_train, data_test, logString = ScalingFeatures(data_train, data_test, InputFeatures, outputDir)
'''
logFile.write(logString)
'''
if drawPlots:
scaledHistogramsPath = outputDir + '/trainScaledHistograms'
print(format('Output directory: ' + Fore.GREEN + scaledHistogramsPath), checkCreateDir(scaledHistogramsPath))
logFile.write('\nSaving histograms of the scaled train dataset in ' + scaledHistogramsPath)
DrawVariablesHisto(data_train, InputFeatures, scaledHistogramsPath, fileCommonName, jetCollection, analysis, channel, signal, backgroundLegend, preselectionCuts, True)
'''
### Saving unscaled dataframes
dataTrainName = outputDir + '/data_train_' + fileCommonName + '.pkl'
data_train.to_pickle(dataTrainName)
print(Fore.GREEN + 'Saved ' + dataTrainName)
logFile.write('\nSaved unscaled train dataframe in ' + dataTrainName)
dataTestName = outputDir + '/data_test_' + fileCommonName + '.pkl'
data_test.to_pickle(dataTestName)
print(Fore.GREEN + 'Saved ' + dataTestName)
logFile.write('\nSaved unscaled test dataframe in ' + dataTestName)
### Closing log file
logFile.close()
print(Fore.GREEN + 'Saved ' + outputDir + logFileName)