/
train_model_2019_run.py
177 lines (138 loc) · 8.15 KB
/
train_model_2019_run.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
import os
import pickle
import gzip
import copy
import torch
from torch import nn
import torch.nn.functional as F
import random, time
import numpy as np
import matplotlib.pyplot as plt
from scipy import sparse
from scipy.stats import rankdata
import networkx as nx
import pandas as pd
from collections import defaultdict,Counter
from datetime import datetime, date
from itertools import combinations
from sklearn.metrics import roc_auc_score, accuracy_score, roc_curve, auc
from general_utils import *
from preprocess_utils import *
from features_utils import *
from train_model_utils import *
rn_time=random.random()*30
time.sleep(rn_time)
if __name__ == '__main__':
split_type=0 # 1 is for conditional case
out_norm=False # we fix this to False, using the raw scores from the neural network output
num_class=2 # binary classfication, fixed
day_origin = date(1990,1,1) # the baseline time
vertex_degree_cutoff=1 # fixed, the vertex has at least one edge connecting to it
min_edges=1 # fixed, minimal number of edges that is considered, not used in the work, can be removed
years_delta=3 # year gap is 3 years
year_start=2019-years_delta # train 2016 for 2019
graph_parameter=[year_start, years_delta, vertex_degree_cutoff, min_edges] # parameters for the knowledge graph
# create folders and subfolders
# it will create a main folder: 2016_train that contains subfolders: t0_c2_log, t0_c2_net, t0_c2_loss, t0_c2_curve, t0_c2_result
save_folders, log_folder=make_folders(year_start, split_type, num_class, "train")
log_run=os.path.join(log_folder,f"train_model_{year_start+years_delta}_run_1") # just a log file to check the running status
with open(log_run+"_logs.txt", "a") as myfile:
myfile.write(f"\n\nstart: {datetime.now()}\n")
# load the full dynamic graph
start_time = time.time()
data_folder="data_concept_graph" # folder that stores the full knowledge graph
graph_file=os.path.join(data_folder,"full_dynamic_graph.parquet")
full_dynamic_graph = pd.read_parquet(graph_file)
with open(log_run+"_logs.txt", "a") as myfile:
myfile.write(f"\n{datetime.now()}: Done, read full_dynamic_graph: {len(full_dynamic_graph)}; elapsed_time: {time.time() - start_time}")
# load data for preparing different type of features
feature_folder="data_for_features" # folder that stores data used for preparing features
start_time=time.time()
adj_mat_sparse=[]
node_neighbor_list=[]
num_neighbor_list=[]
for yy in [year_start,year_start-1,year_start-2]:
data_file=os.path.join(feature_folder, f"adjacency_matrix_{yy}.gz")
adj_mat=get_adjacency_matrix(full_dynamic_graph, yy, data_file) # load the adjacency_matrix file, if not exists, it will automatically generate one
adj_mat_sparse.append(adj_mat) # adjacency_matrix for year y, y-1, y-2
curr_node_neighbor=get_node_neighbor(adj_mat)
node_neighbor_list.append(curr_node_neighbor) # get the neighbors of vertices for years y, y-1, y-2
curr_num_neighbor = np.array(adj_mat.sum(axis=0)).flatten() # array
num_neighbor_list.append(curr_num_neighbor) # get the number of neighbors of vertices for years y, y-1, y-2
with open(log_run+"_logs.txt", "a") as myfile:
myfile.write(f"\n{datetime.now()}: Done, adjacency_matrix_sparse; elapsed_time: {time.time() - start_time}")
start_time=time.time()
vertex_features=get_all_node_feature(adj_mat_sparse, year_start, feature_folder) # prepare all the node features for a vertex in years y, y-1, y-2
# load data for preparing different type of citation features
start_time=time.time()
vc_feature_list=[]
for yy in [year_start,year_start-1,year_start-2]:
data_file=os.path.join(feature_folder, f"concept_node_citation_data_{yy}.parquet") # load the citation information for concepts in year yy
vc_df=pd.read_parquet(data_file)
vc_feature=vc_df.values
vc_feature_list.append(vc_feature)
vertex_cfeatures=get_all_node_cfeature(vc_feature_list)
with open(log_run+"_logs.txt", "a") as myfile:
myfile.write(f"\n{datetime.now()}: Done, vertex_cfeatures; elapsed_time: {time.time() - start_time}")
pair_cf_parameter=[vc_feature_list, node_neighbor_list, num_neighbor_list, vertex_features, vertex_cfeatures] # later used for pair features and cfeatures
# load the whole unconnected pairs for training and testing
train_data_folder = 'data_pair_solution' # folder that stores the unconnected pairs and their citation informations in the future
train_pair_file1=os.path.join(train_data_folder,f"unconnected_{year_start}_pair_solution_connected_{year_start+years_delta}_clean.parquet")
train_pair_file2=os.path.join(train_data_folder,f"unconnected_{year_start}_pair_solution_unconnected_{year_start+years_delta}.parquet")
time_start = time.time()
train_pair_data_yes = pd.read_parquet(train_pair_file1)
with open(log_run+"_logs.txt", "a") as myfile:
myfile.write(f"\nDone, read unconnected_{year_start}_pair_solution_connected_{year_start+years_delta}: {len(train_pair_data_yes)}; elapsed_time: {time.time() - time_start}")
time_start = time.time()
train_pair_data_no = pd.read_parquet(train_pair_file2)
with open(log_run+"_logs.txt", "a") as myfile:
myfile.write(f"\nDone, read unconnected_{year_start}_pair_solution_unconnected_{year_start+years_delta}: {len(train_pair_data_no)}; elapsed_time: {time.time() - time_start}")
time_start = time.time()
full_train_data=np.concatenate((train_pair_data_yes.values, train_pair_data_no.values), axis=0)
with open(log_run+"_logs.txt", "a") as myfile:
myfile.write(f"\nDone, combine all: {len(full_train_data)}; elapsed_time: {time.time() - time_start}")
full_dynamic_graph=pd.DataFrame()
train_pair_data_yes=pd.DataFrame()
train_pair_data_no=pd.DataFrame()
# load the evaluation data feature and solutions
eval_folder="data_eval" # folder that stores the evaluatuion datasets, unconnected pairs, features, solutions
start_time = time.time()
eval_file=os.path.join(eval_folder,"eval_data_pair_solution.parquet")
eval_data_features_df = pd.read_parquet(eval_file)
eval_data_solution=eval_data_features_df.values
eval_data_features_df=pd.DataFrame()
with open(log_run+"_logs.txt", "a") as myfile:
myfile.write(f"finish loading eval_data_features; {time.time()-start_time}")
start_time = time.time()
eval_file=os.path.join(eval_folder,"eval_data_pair_feature.parquet")
eval_data_features_df = pd.read_parquet(eval_file)
eval_data_features=eval_data_features_df.values
eval_data_features_df=pd.DataFrame()
with open(log_run+"_logs.txt", "a") as myfile:
myfile.write(f"\nfinish loading eval_data_solution; {time.time()-start_time}")
# train neural networks for different IR number from 1 to 200
IR_start=1
IR_end=40 # IR_end=200
IR_count=IR_start
while IR_count <= IR_end:
num_impact=random.randint(IR_start, IR_end) # used for parallel computing
IR_num=[num_impact]
IR_Str=format_IR(IR_num, split_type)
logs_file_name=os.path.join(log_folder,f"train_model_{year_start+years_delta}_"+IR_Str)
if not os.path.exists(logs_file_name+"_logs.txt"):
current_time=datetime.now()
open(logs_file_name+"_logs.txt", 'a').close()
batch_size=1000
lr_enc=3*10**-5
rnd_seed=42
hyper_parameter=[batch_size, lr_enc, rnd_seed]
graph_parameter=[year_start,years_delta,vertex_degree_cutoff, min_edges]
user_parameter=[num_class, IR_num, split_type, out_norm]
impact_classfication(full_train_data, eval_data_features, eval_data_solution[:,2], pair_cf_parameter, hyper_parameter, graph_parameter, user_parameter, save_folders, logs_file_name)
IR_count+=1
rn_time=random.random()*30
time.sleep(rn_time)
else:
pass
with open(log_run+"_logs.txt", "a") as myfile:
myfile.write(f"\nfinish: {datetime.now()}\n\n")