-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
191 lines (152 loc) · 6.01 KB
/
main.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
import numpy as np
import pandas as pd
import torch
import torch.nn as nn
from datasets import (
FriendsterSurvivalDataset,
SyntheticSurvivalDataset,
SubsetStar,
)
import random
from Common import utils
import sys
import logging
import os
import matplotlib
matplotlib.rcParams["text.usetex"] = False
import matplotlib.pyplot as plt
import survivalUtils
import survivalNet
import argparse
def crossValidationSplit(lenData, reset=False):
"""
Setup cross-validation folds
Parameters
----------
lenData: Length of the dataset
Rest of the parameters come from the configuration
"""
# Filename to save the cross-validation folds
cvFilename = os.path.join(
"data", f"cv_{args.dataset}_nFolds{args.cvFolds}_seed{args.seed}.bin"
)
if not reset and os.path.isfile(cvFilename):
# Use the file if exists
log.info(f"Using CV splits file : {cvFilename}")
foldsIndices = torch.load(cvFilename)
else:
# Create the file with cross-validation folds
log.info(f"CV splits file {cvFilename} not found. Creating one.")
permutation = torch.randperm(lenData)
foldSize = lenData // args.cvFolds
foldsIndices = []
for iFold in range(args.cvFolds):
start = iFold * foldSize
end = start + foldSize
foldsIndices.append(list(permutation[start:end]))
torch.save(foldsIndices, cvFilename)
# Create train/test splits based on the fold and cvIteration
trainIdx = []
testIdx = []
for i, fold in enumerate(foldsIndices):
if i == args.cvIt:
testIdx.extend(fold)
else:
trainIdx.extend(fold)
return trainIdx, testIdx
# Use GPU if available
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
log = utils.getLogger()
# Training settings
parser = argparse.ArgumentParser(description='Deep Lifetime Clustering.')
parser.add_argument('--dataset', type=str, default=None, metavar='N',
help='Dataset (friendster or synthetic)')
parser.add_argument('--download', action='store_true', default=False,
help='Download the dataset')
parser.add_argument('--k', type=int, default=2, metavar='N',
help='Number of clusters (default: 2)')
parser.add_argument('--lossName', type=str, default="kuiper_ub", metavar='N',
help='Loss for Lifetime clustering: kuiper_ub or mmd')
parser.add_argument('--eol', action='store_true', default=False,
help='End of life signals learnt')
parser.add_argument('--cvFolds', type=int, default=5, metavar='N',
help='Number of cross-validation folds (default: 5)')
parser.add_argument('--cvIt', type=int, default=0, metavar='N',
help='Run fold i (default: 0)')
parser.add_argument('--Ntrain', type=int, default=10000, metavar='N',
help='Number of training samples from each CV-fold (default: 10000)')
parser.add_argument('--Ntest', type=int, default=-1, metavar='N',
help='Number of test samples from each CV-fold, only used for debugging faster (default: -1)')
parser.add_argument('--batchSize', type=int, default=1028, metavar='N',
help='input batch size for training (default: 1028)')
parser.add_argument('--epochs', type=int, default=100, metavar='N',
help='number of epochs to train (default: 100)')
parser.add_argument('--lr', type=float, default=0.01, metavar='LR',
help='learning rate (default: 0.01)')
parser.add_argument('--seed', type=int, default=42, metavar='S',
help='random seed (default: 42)')
parser.add_argument('--show', action='store_true', default=False,
help='Show results plots')
parser.add_argument('--save', action='store_true', default=False,
help='Save results plots')
args = parser.parse_args()
# Set seed for random, numpy and torch (for reproducibility)
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
np.random.seed(args.seed)
random.seed(args.seed)
log.info(f"Using {device}")
if args.dataset.lower().startswith("friendster"):
data = FriendsterSurvivalDataset("data", window=10, download=args.download)
elif args.dataset.lower().startswith("synthetic"):
data = SyntheticSurvivalDataset("data", window=0, clusterIds=[1, 2], download=args.download)
else:
raise NotImplementedError
params = {
# Model params
"k": args.k,
"layerDims": [128, 128],
"lossName": args.lossName,
"activationClass": "relu",
"batchNorm": False,
"endOfLifeSignalsLearnt": args.eol,
"nPairs": "k",
# Fit params
"lr": args.lr,
"nMinibatches": -1,
"batchSize": args.batchSize,
"weightDecay": 0,
"patience": max(10, args.epochs// 10), # Patience for early stopping
"numEpochs": args.epochs,
"fileName": f"survivalNet_k={args.k}.torch",
"plotFileName": "plot",
}
utils.logInfoDict(log, params, "Configuration: ")
# Cross validation split according to the configuration parameters.
trainIdx, testIdx = crossValidationSplit(len(data), reset=False)
trainIdx = np.random.choice(
trainIdx, len(trainIdx) if args.Ntrain == -1 else args.Ntrain, replace=False
)
testIdx = np.random.choice(
testIdx, len(testIdx) if args.Ntest == -1 else args.Ntest, replace=False
)
trainData = SubsetStar(data, trainIdx, train=True)
testData = SubsetStar(
data, testIdx, train=False, mean=trainData.mean, std=trainData.std
)
# Metrics to evaluate on
metrics = [
survivalUtils.concordanceIndex,
survivalUtils.multivariateLogRankScore,
survivalUtils.brierScore,
]
params["fileName"] = f"survivalNet_k={args.k}.torch"
# Run the model
log.info("Begin Run : SurvivalNet")
runFunction = survivalNet.run
trainResults, validResults, testResults = runFunction(
trainData, testData, metrics, show=args.show, save=args.save, _config=params, _run=None
)
# Print train and test results
utils.logInfoDict(log, trainResults, "TrainResults: ")
utils.logInfoDict(log, testResults, "TestResults: ")