-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathNGBoost_MuonEnergyReco_train_true.py
122 lines (103 loc) · 4.92 KB
/
NGBoost_MuonEnergyReco_train_true.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
##### Script for Muon Energy Reconstruction in the water tank
import ngboost as ngb
import numpy as np
import pandas as pd
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from sklearn.metrics import mean_squared_error
import pickle
# import seaborn as sns
#-------- File with events for reconstruction:
#--- evts for training:
infile = "vars_Ereco_train.csv"
#----------------
# Set TF random seed to improve reproducibility
seed = 170
np.random.seed(seed)
E_threshold = 2.
E_low=0
E_high=2000
div=100
bins = int((E_high-E_low)/div)
print('bins: ', bins)
print( "--- opening file with input variables!")
#--- events for training ---
filein = open(str(infile))
print("evts for training in: ",filein)
df00=pd.read_csv(filein)
df0=df00[['totalPMTs','totalLAPPDs','TrueTrackLengthInWater','trueKE','diffDirAbs','TrueTrackLengthInMrd','recoDWallR','recoDWallZ','dirX','dirY','dirZ','vtxX','vtxY','vtxZ','DNNRecoLength']]
#dfsel=df0.loc[df0['neutrinoE'] < E_threshold]
dfsel=df0.dropna()
print("df0.head(): ", df0.head())
#print to check:
print("check training sample: \n",dfsel.head())
# print(dfsel.iloc[5:10,0:5])
#check fr NaN values:
print("The dimensions of training sample ",dfsel.shape)
assert(dfsel.isnull().any().any()==False)
#--- normalisation-training sample:
#dfsel_n = pd.DataFrame([ dfsel['DNNRecoLength']/600., dfsel['TrueTrackLengthInMrd']/200., dfsel['diffDirAbs'], dfsel['recoDWallR']/152.4, dfsel['recoDWallZ']/198., dfsel['totalLAPPDs']/1000., dfsel['totalPMTs']/1000., dfsel['vtxX']/150., dfsel['vtxY']/200., dfsel['vtxZ']/150. ]).T
dfsel_n = pd.DataFrame([ dfsel['DNNRecoLength']/600., dfsel['TrueTrackLengthInMrd'], dfsel['diffDirAbs'], dfsel['recoDWallR'], dfsel['recoDWallZ'], dfsel['totalLAPPDs']/200., dfsel['totalPMTs']/200., dfsel['vtxX']/150., dfsel['vtxY']/200., dfsel['vtxZ']/150. ]).T
print("check normalisation: ", dfsel_n.head())
#--- prepare training & test sample for BDT:
arr_hi_E0 = np.array(dfsel_n[['DNNRecoLength','TrueTrackLengthInMrd','diffDirAbs','recoDWallR','recoDWallZ','totalLAPPDs','totalPMTs','vtxX','vtxY','vtxZ']])
arr3_hi_E0 = np.array(dfsel[['trueKE']])
#---- random split of events ----
rnd_indices = np.random.rand(len(arr_hi_E0)) < 1. #< 0.50
print(rnd_indices[0:5])
#--- select events for training/test:
arr_hi_E0B = arr_hi_E0[rnd_indices]
print(arr_hi_E0B[0:5])
arr2_hi_E_n = arr_hi_E0B #.reshape(arr_hi_E0B.shape + (-1,))
arr3_hi_E = arr3_hi_E0[rnd_indices]
##--- select events for prediction: -- in future we need to replace this with data sample!
#evts_to_predict = arr_hi_E0[~rnd_indices]
#evts_to_predict_n = evts_to_predict #.reshape(evts_to_predict.shape + (-1,))
#test_data_trueKE_hi_E = arr3_hi_E0[~rnd_indices]
#printing..
print('events for training: ',len(arr3_hi_E)) #,' events for predicting: ',len(test_data_trueKE_hi_E))
print('initial train shape: ',arr3_hi_E.shape) #," predict: ",test_data_trueKE_hi_E.shape)
########### NGBoost############
n_estimators=600
params = {
# 'Dist':Exponential,
# 'Base':DecisionTreeRegressor(max_depth=20),
# 'n_estimators':n_estimators,
# 'learning_rate':0.025,
# 'verbose':True
}
print("arr2_hi_E_n.shape: ",arr2_hi_E_n.shape)
#--- select 70% of sample for training and 30% for testing:
offset = int(arr2_hi_E_n.shape[0] * 0.7)
arr2_hi_E_train, arr3_hi_E_train = arr2_hi_E_n[:offset], arr3_hi_E[:offset].reshape(-1) # train sample
arr2_hi_E_test, arr3_hi_E_test = arr2_hi_E_n[offset:], arr3_hi_E[offset:].reshape(-1) # test sample
print("train shape: ", arr2_hi_E_train.shape," label: ",arr3_hi_E_train.shape)
print("test shape: ", arr2_hi_E_test.shape," label: ",arr3_hi_E_test.shape)
print("training NGBoost...")
net_hi_E = ngb.NGBRegressor(**params)
model = net_hi_E.fit(arr2_hi_E_train, arr3_hi_E_train)
# save the model to disk
filename = 'models/finalized_NGBoost_model_forMuonEnergy_trueTrackLengthInMrd.sav'
pickle.dump(model, open(filename, 'wb'))
mse = mean_squared_error(arr3_hi_E_test, net_hi_E.predict(arr2_hi_E_test))
print("MSE: %.4f" % mse)
print("events at training & test samples: ", len(arr_hi_E0))
print("events at train sample: ", len(arr2_hi_E_train))
print("events at test sample: ", len(arr2_hi_E_test))
# test_score = np.zeros((params['n_estimators'],), dtype=np.float64)
# for i, y_pred in enumerate(net_hi_E.staged_predict(arr2_hi_E_test)):
# test_score[i] = net_hi_E.loss_(arr3_hi_E_test, y_pred)
# fig,ax=plt.subplots(ncols=1, sharey=True)
# ax.plot(np.arange(params['n_estimators']) + 1, net_hi_E.train_score_, 'b-',
# label='Training Set Deviance')
# ax.plot(np.arange(params['n_estimators']) + 1, test_score, 'r-',
# label='Test Set Deviance')
# ax.set_ylim(0.,500.)
# ax.set_xlim(0.,n_estimators)
# ax.legend(loc='upper right')
# ax.set_ylabel('Least Absolute Deviations [MeV]')
# ax.set_xlabel('Number of Estimators')
# ax.yaxis.set_label_coords(-0.1, 0.6)
# ax.xaxis.set_label_coords(0.85, -0.08)
# plt.savefig("deviation_train_test.png")