forked from miworking/KDD2015_XUETANG
-
Notifications
You must be signed in to change notification settings - Fork 0
/
extra_feature.py
246 lines (180 loc) · 9.78 KB
/
extra_feature.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
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
import pandas as pd
import pickle
import numpy as np
import os
from course_tree import *
from full_tree import *
from full_tree_node import *
df = pd.DataFrame(pd.read_csv('./object.csv',header = 0))
for course_id in pd.unique(df.course_id):
print "---------------------------" + str(course_id) + "---------------------------"
# load full course tree
course_tree_path = './train/courseDump/' + course_id + ".txt"
f = open(course_tree_path,'rb')
full_tree = pickle.load(f)
f.close()
# load enrollment data
enrollment = pd.DataFrame(pd.read_csv('./train/enrollment_train.csv',header = 0))
enrollmentIds = enrollment[enrollment.course_id == course_id]
del enrollmentIds['username']
# ------------------------- load classification result -------------------------
result = pd.DataFrame(pd.read_csv('./train/truth_train.csv',header = 0))
# merge enrollment and result to 'matrix', get enrollment_id and their result
matrix = pd.merge(enrollmentIds,result,how='inner',on='enrollment_id')
del matrix['course_id']
# ------------------------- get access log of this course -------------------------
access_path = './train/course/' + course_id + ".csv"
access = pd.DataFrame(pd.read_csv(access_path,header = 0))
access_grouped= access.groupby('enrollment_id')
access_count = access_grouped.count()
del access_count['time']
del access_count['event']
del access_count['source']
access_count = access_count.reset_index()
# ------------------------- get total access counts -------------------------
matrix = pd.merge(matrix,access_count,on="enrollment_id",how='inner')
matrix['access_counts'] = matrix.object
del matrix['object']
# ------------------------- get the first_access_in(days: relative to the module's start time) -------------------------
module_path = './train/fullTree/' + course_id + "_moduleIds.csv"
modules = pd.DataFrame(pd.read_csv(module_path,header = 0))
get_simple = lambda x : x[: x.find('.')]
modules['start'] = modules['start'].apply(get_simple)
first_access = access.sort("time").groupby("enrollment_id", as_index=False).first()
first_access_with_time = pd.merge(first_access,modules,left_on='object',right_on='module_id',how='left')
del first_access_with_time['source']
del first_access_with_time['event']
del first_access_with_time['chapterId']
del first_access_with_time['module_id']
del first_access_with_time['category']
del first_access_with_time['sequentialId']
get_timeT = lambda x : datetime.datetime.strptime(str(x),"%Y-%m-%dT%H:%M:%S")
get_time = lambda x : datetime.datetime.strptime(str(x),"%Y-%m-%d %H:%M:%S")
first_access_with_time['time'] = first_access_with_time['time'].apply(get_timeT)
first_access_with_time['start'] = first_access_with_time['start'].apply(get_time)
first_access_with_time['from'] = first_access_with_time['time'] - first_access_with_time['start']
get_float_sec = lambda x : x / np.timedelta64(1, 'D')
first_access_with_time['from'] = first_access_with_time['from'].apply(get_float_sec)
del first_access_with_time['start']
del first_access_with_time['time']
del first_access_with_time['object']
matrix = pd.merge(matrix,first_access_with_time,on='enrollment_id',how='left')
# ------------------------- get access timespan(in days: relative to first access start time) -------------------------
first_access = access.sort("time").groupby("enrollment_id", as_index=False).first()
del first_access['object']
del first_access['event']
del first_access['source']
first_access.rename(columns={'time':'first_time'},inplace=True)
last_access = access.sort("time").groupby("enrollment_id", as_index=False).last()
del last_access['object']
del last_access['event']
del last_access['source']
last_access.rename(columns={'time':'last_time'},inplace=True)
timespan = pd.merge(first_access,last_access,on="enrollment_id",how="left")
get_timeT = lambda x : datetime.datetime.strptime(str(x),"%Y-%m-%dT%H:%M:%S")
timespan['first_time'] = timespan['first_time'].apply(get_timeT)
timespan['last_time'] = timespan['last_time'].apply(get_timeT)
timespan['timespan'] = (timespan['last_time'] - timespan['first_time']) / np.timedelta64(1, 'D')
del timespan['first_time']
del timespan['last_time']
matrix = pd.merge(matrix,timespan,on='enrollment_id',how='left')
# ------------------------- get video_access_counts -------------------------
videos = modules[modules.category == 'video']
videolist = videos.module_id.tolist()
video_access = access[access['object'].isin(videolist)]
video_grouped = video_access.groupby('enrollment_id')
video_access_count = video_grouped.count()
del video_access_count['time']
del video_access_count['source']
del video_access_count['object']
video_access_count = video_access_count.reset_index()
matrix = pd.merge(matrix,video_access_count,on='enrollment_id',how='left')
matrix.fillna(0,inplace=True)
matrix.rename(columns={'event':'video_access_counts'},inplace=True)
# ------------------------- get video_start_in (days) -------------------------
video_grouped_by2 = video_access.groupby(['enrollment_id','object'])
video_first_access_time = video_grouped_by2.min()
video_first_access_time = video_first_access_time.reset_index()
video_start_in = pd.merge(video_first_access_time,videos,left_on='object',right_on='module_id',how='left')
del video_start_in['category']
del video_start_in['sequentialId']
del video_start_in['chapterId']
del video_start_in['module_id']
del video_start_in['source']
del video_start_in['event']
get_timeT = lambda x : datetime.datetime.strptime(str(x),"%Y-%m-%dT%H:%M:%S")
get_time = lambda x : datetime.datetime.strptime(str(x),"%Y-%m-%d %H:%M:%S")
video_start_in['time'] = video_start_in['time'].apply(get_timeT)
video_start_in['start'] = video_start_in['start'].apply(get_time)
video_start_in['video_start_in'] = video_start_in['time'] - video_start_in['start']
get_float_sec = lambda x : x / np.timedelta64(1, 'D')
video_start_in['video_start_in'] = video_start_in['video_start_in'].apply(get_float_sec)
del video_start_in['time']
del video_start_in['start']
video_start_in_days = video_start_in.groupby('enrollment_id').mean()
video_start_in_days.reset_index(inplace=True)
matrix = pd.merge(matrix,video_start_in_days,on='enrollment_id',how='left')
matrix['video_start_in'].fillna(-1,inplace=True)
# ------------------------- get problem access counts -------------------------
problems = modules[modules.category == 'problem']
problemlist = problems.module_id.tolist()
problem_access = access[access['object'].isin(problemlist)]
problem_grouped = problem_access.groupby('enrollment_id')
problem_access_count = problem_grouped.count()
del problem_access_count['time']
del problem_access_count['source']
del problem_access_count['object']
problem_access_count = problem_access_count.reset_index()
matrix = pd.merge(matrix,problem_access_count,on='enrollment_id',how='left')
matrix.fillna(0,inplace=True)
matrix.rename(columns={'event':'problem_access_counts'},inplace=True)
# ------------------------- get problem_start_in (days) -------------------------
problem_grouped_by2 = problem_access.groupby(['enrollment_id','object'])
problem_first_access_time = problem_grouped_by2.min()
problem_first_access_time = problem_first_access_time.reset_index()
problem_start_in = pd.merge(problem_first_access_time,problems,left_on='object',right_on='module_id',how='left')
del problem_start_in['category']
del problem_start_in['sequentialId']
del problem_start_in['chapterId']
del problem_start_in['module_id']
del problem_start_in['source']
del problem_start_in['event']
get_timeT = lambda x : datetime.datetime.strptime(str(x),"%Y-%m-%dT%H:%M:%S")
get_time = lambda x : datetime.datetime.strptime(str(x),"%Y-%m-%d %H:%M:%S")
problem_start_in['time'] = problem_start_in['time'].apply(get_timeT)
problem_start_in['start'] = problem_start_in['start'].apply(get_time)
problem_start_in['problem_start_in'] = problem_start_in['time'] - problem_start_in['start']
get_float_sec = lambda x : x / np.timedelta64(1, 'D')
problem_start_in['problem_start_in'] = problem_start_in['problem_start_in'].apply(get_float_sec)
del problem_start_in['time']
del problem_start_in['start']
problem_start_in_days = problem_start_in.groupby('enrollment_id').mean()
problem_start_in_days.reset_index(inplace=True)
matrix = pd.merge(matrix,problem_start_in_days,on='enrollment_id',how='left')
matrix['problem_start_in'].fillna(-1,inplace=True)
# ------------- adjust the column positions: put result to the last column ------------------------
matrix['res'] = matrix['result']
del matrix['result']
matrix['result'] = matrix['res']
del matrix['res']
# write matrix to csv
path = './train/features/' + course_id + '.csv'
matrix.to_csv(path,index = False,index_label= False,header = 1)
# cross validation on random forest
from sklearn.cross_validation import cross_val_score
from sklearn.datasets import make_blobs
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import ExtraTreesClassifier
from sklearn.tree import DecisionTreeClassifier
y = matrix.result
del matrix['result']
del matrix['enrollment_id']
X = matrix
mode_result_path = './randomForest/' + str(course_id) + '.result'
mode_data_path = './randomForest/' + str(course_id) + '.data'
f1 = open(mode_result_path,'w')
pickle.dump(X,f1,0)
f1.close()
f2 = open(mode_data_path,'w')
pickle.dump(y,f2,0)
f2.close()