-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathlaunch_experiment.py
128 lines (87 loc) · 4.04 KB
/
launch_experiment.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
import pandas as pd
import numpy as np
from scipy import sparse
import os
import scipy.sparse.linalg as sp_l
import faiss
import torch
from torch import nn, optim
import time
import matplotlib.pyplot as plt
import seaborn as sns
sns.set()
# Manage all hyperparameters :
from argparse import ArgumentParser
from utils import *
from algorithms import *
parser = ArgumentParser()
parser.add_argument("--exp_folder", type=str, default='experiments/twitch/N:300000_K:1000_seed:0')
parser.add_argument("--method", type=str, default='uniform')
parser.add_argument("--n_samples", type=int, default=1000)
parser.add_argument("--trunc_at", type=int, default=256)
parser.add_argument("--eps", type=float, default=.8)
parser.add_argument("--lr", type=float, default=5e-5)
parser.add_argument("--seed", type=int, default=0)
args = vars(parser.parse_args())
print(args)
seed = args['seed']
np.random.seed(seed)
torch.manual_seed(seed)
print('Loading the different splits')
exp_folder_path = args['exp_folder']
data_folder = os.path.join(exp_folder_path, 'data')
contexts_train, rewards_train = load_tensor_spmx_dataset(data_folder, split = 'train')
contexts_val, rewards_val = load_tensor_spmx_dataset(data_folder, split = 'val')
contexts_test, rewards_test = load_tensor_spmx_dataset(data_folder, split = 'test')
print('Loading the embeddings')
prod_emb = np.load(os.path.join(data_folder, 'prod_emb.npy'))
P = prod_emb.shape[0]
print('number of products is %d'%P)
print('Loading the index')
index = faiss.read_index(os.path.join(data_folder, 'index'))
print('Define the learning algorithms')
epochs = 10
bsize = 32
reg = 1e-10
method = args['method']
lr = args['lr']
n_samples = args['n_samples']
eps = args['eps']
trunc_at = args['trunc_at']
policy = policy_model(prod_emb).cuda()
if method == 'exact' :
path = method + '_Ns:%d_Lr:%.6f'%(n_samples, lr)
print('-------------Exact Reinforce---------------')
train_log, val_log_index, val_log, exec_time = exact_reinforce(policy, index, n_samples, contexts_train, rewards_train,
epochs, bsize, lr, reg, contexts_val, rewards_val)
elif method == 'uniform' :
path = method + '_Ns:%d_Lr:%.6f'%(n_samples, lr)
print('-------------Uniform SNIS---------------')
train_log, val_log_index, val_log, exec_time = uniform_snips_approximate_reinforce(policy, index, P, n_samples, contexts_train, rewards_train,
epochs, bsize, lr, reg, contexts_val, rewards_val)
elif method == 'mixture':
path = method + '_Ns:%d_T@:%d_Eps:%.1f_Lr:%.6f'%(n_samples, trunc_at, eps, lr)
print('-------------Mixture SNIS %.1f---------------'%eps)
train_log, val_log_index, val_log, exec_time = mixture_snips_approximate_reinforce(policy, index, P, n_samples, trunc_at, eps,
contexts_train, rewards_train, epochs, bsize, lr, reg,
contexts_val, rewards_val)
else:
print("method not supported")
print("method should be either exact, uniform, mixture")
test_log_index, test_log = test_policy(policy, index, contexts_test, rewards_test, 512)
print('Test Results : INDEX %.4f, TRUE %.4f '%(test_log_index, test_log))
print('Execution Time : %.4f seconds'%exec_time)
result_path = os.path.join(exp_folder_path, 'results/' + path)
create_dir(result_path)
train_logs_dict = {method: train_log}
train_logs = pd.DataFrame(train_logs_dict)
train_logs.to_csv(result_path + '/train_logs.csv', index=False)
val_logs_dict = {method: val_log}
val_logs = pd.DataFrame(val_logs_dict)
val_logs.to_csv(result_path + '/val_logs.csv', index=False)
val_logs_dict = {method: val_log_index}
val_logs = pd.DataFrame(val_logs_dict)
val_logs.to_csv(result_path + '/val_logs_index.csv', index=False)
exec_time_dict = {method: [exec_time, test_log]}
exec_logs = pd.DataFrame(exec_time_dict)
exec_logs.to_csv(result_path + '/exec_test_logs.csv', index=False)