-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
134 lines (122 loc) · 3.8 KB
/
train.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
import gzip
import json
import time
import torch
import numpy as np
import torch.nn as nn
from pathlib import Path
import torch.optim as optim
from argparse import ArgumentParser
from torch.utils.data import DataLoader
from SNNomics.dataset import SiameseDataset
from SNNomics.model import SNN
from SNNomics.utils import check_dir
from SNNomics.trainer import Trainer
def siamese_collate_fn(batch):
anchors, positives, negatives = zip(*batch)
return torch.tensor(np.stack(anchors)), torch.tensor(np.stack(positives)), torch.tensor(np.stack(negatives))
if __name__ == '__main__':
parser = ArgumentParser()
parser.add_argument(
'-epochs',
help='number of epochs to run',
type=int,
default=25,
)
parser.add_argument(
'-folds',
help='Path to json file containing training folds',
type=str,
required=True,
)
parser.add_argument(
'-expression_mat',
help='Path to .npz file containing a samples x genes expression matrix',
type=str,
required=True,
)
parser.add_argument(
'--save',
help='if applied, will save the model weights',
action='store_true',
)
parser.add_argument(
'-batch_size',
help='size of each batch',
type=int,
default=128,
)
parser.add_argument(
'-lr',
help='learning rate',
type=float,
default=0.001,
)
parser.add_argument(
'-weight_decay',
help='weight_decay',
type=float,
default=0.0005,
)
parser.add_argument(
'-outdir',
help='directory to save results to',
type=str,
default='results',
)
parser.add_argument(
'-out_prefix',
help='prefix of results outfiles',
type=str,
default=None,
)
args = parser.parse_args()
start = time.time()
# Set paths
folds_file = Path(args.folds)
expression_file = Path(args.expression_mat)
outdir = Path(args.outdir)
check_dir(outdir)
# Assign training variables
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
batch_size = args.batch_size
epochs = args.epochs
save_model = args.save
lr = args.lr
weight_decay = args.weight_decay
# Load data
expression_data = np.load(expression_file, allow_pickle=True)
expression = expression_data['expression']
samples = expression_data['gsms']
genes = expression_data['genes']
num_genes = len(genes)
# Load folds
with gzip.open(folds_file, 'r') as f:
folds_bytes = f.read()
folds_str = folds_bytes.decode('utf-8')
folds = json.loads(folds_str)
# Train model with k-fold CV
for k in folds:
print(f"Training fold {k}")
# Assign training arguments
model = SNN(num_genes, 'train')
mpdel = model.to(device)
optimizer = optim.Adam(model.parameters(), lr=lr, weight_decay=weight_decay)
triplet_loss = nn.TripletMarginLoss(margin=1.0, p=2, eps=1e-7)
train_data = SiameseDataset(expression_mat=expression, gsms=samples, training_json=folds, fold=k, split='train')
test_data = SiameseDataset(expression_mat=expression, gsms=samples, training_json=folds, fold=k, split='test')
train_loader = DataLoader(train_data, batch_size=batch_size, collate_fn=siamese_collate_fn, num_workers=1, shuffle=True)
test_loader = DataLoader(test_data, batch_size=batch_size, num_workers=1, shuffle=True)
trainer = Trainer(
model,
optimizer,
triplet_loss,
train_loader,
test_loader,
device,
outdir,
k,
)
trainer.train(epochs=epochs)
trainer.test()
trainer.save_weights(f'model_fold-{k}.pt')