-
Notifications
You must be signed in to change notification settings - Fork 79
/
XGBoostClassifier.py
207 lines (176 loc) · 7.91 KB
/
XGBoostClassifier.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
# -*- coding: utf-8 -*-
"""
Created on oct 20 23:15:24 2015
@author: marios
Script that makes Xgboost scikit-like.
The initial version of the script came from Guido Tapia (or such is his kaggle name!). I have modified it quite a bit though.
the github from where this was retrieved was : https://github.com/gatapia/py_ml_utils
He has done excellent job in making many commonly used algorithms scikit-like
"""
from sklearn.base import BaseEstimator, ClassifierMixin
import sys
from sklearn.cross_validation import StratifiedKFold
import xgboost as xgb
import numpy as np
from scipy.sparse import csr_matrix
class XGBoostClassifier(BaseEstimator, ClassifierMixin):
def __init__(self, silent=True,
use_buffer=True, num_round=10,num_parallel_tree=1, ntree_limit=0,
nthread=None, booster='gbtree',
eta=0.3, gamma=0.01,
max_depth=6, min_child_weight=1, subsample=1,
colsample_bytree=1,
l=0, alpha=0, lambda_bias=0, objective='reg:linear',
eval_metric='logloss', seed=0, num_class=None,
max_delta_step=0,classes_=None ,
colsample_bylevel=1.0 , sketch_eps=0.1 , sketch_ratio=2.0 ,
opt_dense_col=1, size_leaf_vector=0.0, min_split_loss=0.0,
cache_opt=1, default_direction =0 , k_folds=0 ,early_stopping_rounds=200
):
assert booster in ['gbtree', 'gblinear']
assert objective in ['reg:linear', 'reg:logistic',
'binary:logistic', 'binary:logitraw', 'multi:softmax',
'multi:softprob', 'rank:pairwise','count:poisson']
assert eval_metric in [ 'rmse', 'mlogloss', 'logloss', 'error',
'merror', 'auc', 'ndcg', 'map', 'ndcg@n', 'map@n', 'kappa']
if eval_metric=='kappa':
booster='gblinear'
self.silent = silent
self.use_buffer = use_buffer
self.num_round = num_round
self.ntree_limit = ntree_limit
self.nthread = nthread
self.booster = booster
# Parameter for Tree Booster
self.eta=eta
self.gamma=gamma
self.max_depth=max_depth
self.min_child_weight=min_child_weight
self.subsample=subsample
self.colsample_bytree=colsample_bytree
self.colsample_bylevel=colsample_bylevel
self.max_delta_step=max_delta_step
self.num_parallel_tree=num_parallel_tree
self.min_split_loss=min_split_loss
self.size_leaf_vector=size_leaf_vector
self.default_direction=default_direction
self.opt_dense_col=opt_dense_col
self.sketch_eps=sketch_eps
self.sketch_ratio=sketch_ratio
self.k_folds=k_folds
self.k_models=[]
self.early_stopping_rounds=early_stopping_rounds
# Parameter for Linear Booster
self.l=l
self.alpha=alpha
self.lambda_bias=lambda_bias
# Misc
self.objective=objective
self.eval_metric=eval_metric
self.seed=seed
self.num_class = num_class
self.n_classes_ =num_class
self.classes_=classes_
def set_params(self,random_state=1):
self.seed=random_state
def build_matrix(self, X, opt_y=None, weighting=None):
if opt_y==None:
if weighting==None:
return xgb.DMatrix(csr_matrix(X), missing =-999.0)
else :
#scale weight
sumtotal=float(X.shape[0])
sumweights=np.sum(weighting)
for s in range(0,len(weighting)):
weighting[s]*=sumtotal/sumweights
return xgb.DMatrix(csr_matrix(X), missing =-999.0, weight=weighting)
else:
if weighting==None:
return xgb.DMatrix(csr_matrix(X), label=np.array(opt_y), missing =-999.0)
else :
sumtotal=float(X.shape[0])
sumweights=np.sum(weighting)
for s in range(0,len(weighting)):
weighting[s]*=sumtotal/sumweights
return xgb.DMatrix(csr_matrix(X), label=np.array(opt_y), missing =-999.0, weight=weighting)
def fit(self, X, y,sample_weight=None):
self.k_models=[]
X1 = self.build_matrix(X, y,weighting= sample_weight)#sample_weight)
param = {}
param['booster']=self.booster
param['objective'] = self.objective
param['bst:eta'] = self.eta
param['seed']= self.seed
param['bst:max_depth'] = self.max_depth
if self.eval_metric!='kappa':
param['eval_metric'] = self.eval_metric
param['bst:min_child_weight']= self.min_child_weight
param['silent'] = 1
param['nthread'] = self.nthread
param['bst:subsample'] = self.subsample
param['gamma'] = self.gamma
param['colsample_bytree']= self.colsample_bytree
param['num_parallel_tree']= self.num_parallel_tree
param['colsample_bylevel']= self.colsample_bylevel
#param['min_split_loss']=self.min_split_loss
param['default_direction']=self.default_direction
param['opt_dense_col']=self.opt_dense_col
param['sketch_eps']=self.sketch_eps
param['sketch_ratio']=self.sketch_ratio
param['size_leaf_vector']=self.size_leaf_vector
if self.num_class is not None:
param['num_class']= self.num_class
if self.k_folds <2:
self.bst = xgb.train(param.items(), X1, self.num_round)
else :
number_of_folds=self.k_folds
kfolder2=StratifiedKFold(y, n_folds=number_of_folds,shuffle=True, random_state=self.seed)
## we split 64-16 5 times to make certain all the data has been use in modelling at least once
for train_indexnew, test_indexnew in kfolder2:
if sample_weight==None:
dtrain = xgb.DMatrix(X[train_indexnew], label=y[train_indexnew])
dtvalid = xgb.DMatrix(X[test_indexnew], label=y[test_indexnew])
else :
dtrain = xgb.DMatrix(X[train_indexnew], label=y[train_indexnew], weight=sample_weight[train_indexnew])
dtvalid = xgb.DMatrix(X[test_indexnew], label=y[test_indexnew], weight=sample_weight[test_indexnew])
watchlist = [(dtrain, 'train'), (dtvalid, 'valid')]
gbdt = xgb.train(param.items(), dtrain, self.num_round, watchlist, verbose_eval=False, early_stopping_rounds=self.early_stopping_rounds)#, verbose_eval=250) #, early_stopping_rounds=250, verbose_eval=250)
#predsnew = gbdt.predict(dtest, ntree_limit=gbdt.best_iteration)
self.k_models.append(gbdt)
return self
def predict(self, X):
if self.k_models!=None and len(self.k_models)<2:
X1 = self.build_matrix(X)
return self.bst.predict(X1)
else :
dtest = xgb.DMatrix(X)
preds= [0.0 for k in X.shape[0]]
for gbdt in self.k_models:
predsnew = gbdt.predict(dtest, ntree_limit=(gbdt.best_iteration+1)*self.num_parallel_tree)
for g in range (0, predsnew.shape[0]):
preds[g]+=predsnew[g]
for g in range (0, len(preds)):
preds[g]/=float(len(self.k_models))
def predict_proba(self, X):
try:
rows=(X.shape[0])
except:
rows=len(X)
X1 = self.build_matrix(X)
if self.k_models!=None and len(self.k_models)<2:
predictions = self.bst.predict(X1)
else :
dtest = xgb.DMatrix(X)
predictions= None
for gbdt in self.k_models:
predsnew = gbdt.predict(dtest, ntree_limit=(gbdt.best_iteration+1)*self.num_parallel_tree)
if predictions==None:
predictions=predsnew
else:
for g in range (0, predsnew.shape[0]):
predictions[g]+=predsnew[g]
for g in range (0, len(predictions)):
predictions[g]/=float(len(self.k_models))
predictions=np.array(predictions)
if self.objective == 'multi:softprob': return predictions.reshape( rows, self.num_class)
return np.vstack([1 - predictions, predictions]).T