Repository navigation
Expand file tree
/
Copy pathtest_train.py
More file actions
134 lines (112 loc) · 4.51 KB
/
Copy pathtest_train.py
File metadata and controls
134 lines (112 loc) · 4.51 KB
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
from src.nanopore_dataset import BaseNanoporeDataset, custom_collate
from torch.utils.data import DataLoader
import jax
from jax import numpy as jnp
from flax.training import checkpoints
import time
import wandb
NON_RECURRENT_ENCODING_DICT = {'A':1, 'C':2, 'G':3, 'T':4, '':0}
NON_RECURRENT_DECODING_DICT = {1:'A', 2:'C', 3:'G', 4:'T', 0:''}
BASES = ['A', 'C', 'G', 'T']
BASES_CTC = 'N' + ''.join(BASES)
data_dir = "/home/townsag/nanopore_project/test_nn_input"
#data_dir = "/home/townsag/nanopore_project/SimpleCall/practice_data/training_input"
checkpoint_dir = "/home/townsag/nanopore_project/SimpleCall/checkpoints"
batch_size = 64
s2s = False
dataset = BaseNanoporeDataset(
data_dir = data_dir,
decoding_dict = NON_RECURRENT_DECODING_DICT,
encoding_dict = NON_RECURRENT_ENCODING_DICT,
split = 0.95,
shuffle = True,
seed = 1,
s2s = s2s,
)
print(f"total num samples in dataset: {len(dataset)}")
dataloader_train = DataLoader(
dataset,
batch_size = batch_size,
sampler = dataset.train_sampler,
collate_fn=custom_collate,
num_workers = 1
)
dataloader_validation = DataLoader(
dataset,
batch_size = batch_size,
sampler = dataset.validation_sampler,
collate_fn=custom_collate,
num_workers = 1
)
print(f"num train batches {len(dataloader_train)}")
print(f"num test batches{len(dataloader_validation)}")
seed = 0
learning_rate = 1e-3
min_learning_rate = 1e-5
num_epochs = 5
linear_warmup_steps = 10
from ml_collections import config_dict
initial_dictionary = {
'linear_warmup_steps': linear_warmup_steps,
'num_batches_per_epoch': len(dataloader_train),
'num_epochs': num_epochs,
'base_learning_rate': learning_rate,
'min_learning_rate': min_learning_rate
}
cfg = config_dict.ConfigDict(initial_dictionary)
wandb.init(
# set the wandb project where this run will be logged
project="SimpleCall-20231213",
# track hyperparameters and run metadata
config={
"base_learning_rate": learning_rate,
"min_learning_rate": min_learning_rate,
"architecture": "CNN-LSTM-CTC",
"dataset": "demo_data",
"batch_size": batch_size,
"epochs": num_epochs,
"optimizer": "adam with linear warmup and cosine decay",
"num_linear_warmup_steps": linear_warmup_steps,
"batches_per_epoch": len(dataloader_train)
}
)
from src.model import create_train_state, train_one_epoch, evaluate_model
from jax import device_put
print("initializing train_state")
start = time.time()
train_state = create_train_state(key=jax.random.PRNGKey(seed), config=cfg)
end = time.time()
print("time to create train state: ", end - start)
print("starting training")
for epoch in range(1, num_epochs + 1):
start = time.time()
train_state, train_metrics = train_one_epoch(state=train_state, train_dataloader=dataloader_train,
alphabet=BASES_CTC, epoch=epoch, batches_per_epoch=cfg.num_batches_per_epoch, checkpoint_dir=checkpoint_dir)
end = time.time()
print(f"========== Total Epoch Train Time: {end - start} ==========")
print(f"Train epoch: {epoch}, loss: {train_metrics['mean_epoch_loss']}, accuracy: {train_metrics['mean_epoch_accuracy'] * 100}")
start = time.time()
print(f"Starting evaluation for epoch: {epoch}")
test_metrics = evaluate_model(state=train_state, test_dataloader=dataloader_validation, alphabet=BASES_CTC, epoch=epoch)
print(f"\Validation epoch: {epoch}, loss: {test_metrics['loss']}, accuracy: {test_metrics['accuracy'] * 100}")
end = time.time()
print(f"========== Total Epoch Test Time: {end - start} ==========")
checkpoints.save_checkpoint(ckpt_dir=checkpoint_dir, target=train_state.params, step=train_state.step, keep=3)
wandb.finish()
"""
first_batch = next(iter(dataloader_train))
features = device_put(first_batch["x"])
labels = device_put(first_batch["y"])
labels_padding_mask = jnp.zeros_like(labels)
labels_padding_mask = labels_padding_mask.at[labels == 0].set(1)
print(f"features size: {features.shape}")
print(f"labels size: {labels.shape}")
from src.model import train_step
print("features type and device: ", type(features), features.devices())
print("labels type and device: ", type(labels), labels.devices())
print("padding type and device: ", type(labels_padding_mask), labels_padding_mask.devices())
start = time.time()
state, batch_logits, mean_batch_loss = train_step(state=train_state, samples=features, ground_truth_labels=labels,labels_padding_mask=labels_padding_mask, batch=1)
end = time.time()
print("time to train one batch: ", end - start)
"""