/
Pretest.py
121 lines (114 loc) · 6.04 KB
/
Pretest.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
import argparse
import os
import time
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from accelerate import Accelerator
from contextlib import suppress as nullcontext
from DataLoader import TestDataset_MLM
from Utility import init_seeds, xavier_init, evaluatefunction
from MIIR import *
def pretestfunction(accelerator, model, dataloader, epoch, rank, save_path):
results = []
indices = []
ddp_model, ddp_dataloader = accelerator.prepare(model, dataloader)
ddp_model.eval()
with torch.no_grad():
for batch in ddp_dataloader:
index, session_ids, session_side_information, session_masks, session_negatives = batch
output = ddp_model(session_ids[0], session_side_information, session_masks[0])
output = output[0] # item id
index = index.cpu().numpy() # [batch_size]
output = output[session_masks[1] == 1].cpu().numpy() # [batch_size, item_num], note that include padding and missing
session_positives = session_ids[1][session_masks[1] == 1].cpu().numpy() # [batch_size]
session_negatives = session_negatives.cpu().numpy()
result = evaluatefunction(output, session_positives, session_negatives)
results.extend(result)
indices.extend(index.tolist())
with open(save_path+'pretest-'+str(epoch)+'-'+str(rank)+'.txt', 'w') as f:
i = 0
for result in results:
result['index'] = indices[i]
f.write(str(result)+'\n')
i += 1
accelerator.wait_for_everyone()
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('-dataset_files', type=str, help='the files for a dataset')
parser.add_argument('-save_path', type=str, help='the path for saving the model')
parser.add_argument('-n_gpu', type=int, default=1, help='the number of gpus, should be consistent with the num_processes in accelerate_config.yaml')
parser.add_argument('-max_interactions', type=int, default=20, help='the maximum number of the interactions in one input set')
parser.add_argument('-batch_size', type=int, default=256, help='the size of one batch')
parser.add_argument('-num_workers', type=int, default=10, help='how many subprocesses to use for data loading')
parser.add_argument('-resume', type=bool, default=False, help='resume or not')
args = parser.parse_args()
with open(args.dataset_files, 'r') as f:
content = f.readlines()
dataset_files = eval(content[0].strip())
with open(args.save_path+'pretrain_result.txt', 'r') as f:
content = f.readlines()
epochs = len(content)
init_seeds()
accelerator = Accelerator(split_batches=True)
if accelerator.is_main_process:
if not os.path.exists(args.save_path):
os.makedirs(args.save_path)
if args.resume:
fw = open(args.save_path+'pretest_result.txt', 'a')
else:
fw = open(args.save_path+'pretest_result.txt', 'w')
accelerator.wait_for_everyone()
accelerator.print('Evaluate the pretrained model')
model = MIIR(dataset_files['dataset_token'])
dataset = TestDataset_MLM(dataset_files['interactions_filepath'], dataset_files['categories_filepath'], dataset_files['brands_filepath'], dataset_files['titles_filepath'], dataset_files['descriptions_filepath'], dataset_files['missings_filepath'], dataset_files['test_rows_filepath'], dataset_files['test_negatives_filepath'], args.max_interactions, dataset_files['masks_filepath'])
dataloader = DataLoader(dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.num_workers, drop_last=False) # because drop_last=False, it may be some duplicate instances in the last batch
next_epoch = 1
if args.resume:
with open(args.save_path+'pretest_result.txt', 'r') as f:
content = f.readlines()
next_epoch = len(content)+1
accelerator.print('Resume from Epoch %d' % (next_epoch,))
epoch = next_epoch
while epoch <= epochs:
model.load_state_dict(torch.load(args.save_path+'pretrain-'+str(epoch)+'.pth'))
start = time.time()
pretestfunction(accelerator, model, dataloader, epoch, accelerator.process_index, args.save_path)
if accelerator.is_main_process: # merge results from all processes
results = {}
for rank in range(args.n_gpu):
with open(args.save_path+'pretest-'+str(epoch)+'-'+str(rank)+'.txt', 'r') as f:
content = f.readlines()
for line in content:
line = eval(line.strip())
index = line['index']
results[index] = line
os.remove(args.save_path+'pretest-'+str(epoch)+'-'+str(rank)+'.txt')
assert len(results) == len(dataset), 'len(results) should be equal to len(dataset)'
average_result = {}
with open(args.save_path+'pretest-'+str(epoch)+'.txt', 'w') as f:
for index in range(len(results)):
if index == 0:
for metric in results[index]:
if metric != 'index' and metric != 'predict':
average_result[metric] = results[index][metric]
else:
for metric in results[index]:
if metric != 'index' and metric != 'predict':
average_result[metric] += results[index][metric]
f.write(str(results[index])+'\n')
metric_sum = 0
for metric in average_result:
average_result[metric] /= len(results)
metric_sum += average_result[metric]
average_result['sum'] = metric_sum
print('Epoch %d Time %d:' % (epoch,time.time()-start), average_result)
average_result['epoch'] = epoch
fw.write(str(average_result)+'\n')
fw.flush()
accelerator.wait_for_everyone()
epoch += 1
if accelerator.is_main_process:
fw.close()