harry
commited on
Commit
·
aaea685
1
Parent(s):
f774571
feat: update training loop with learning rate scheduler and progress bar enhancements
Browse files
mnist_classifier/train.py
CHANGED
@@ -10,6 +10,7 @@ import os
|
|
10 |
import random
|
11 |
import numpy as np
|
12 |
from tqdm import tqdm
|
|
|
13 |
|
14 |
def set_seed(seed):
|
15 |
torch.manual_seed(seed)
|
@@ -23,7 +24,7 @@ def set_seed(seed):
|
|
23 |
def train():
|
24 |
# Training loop
|
25 |
learning_rate = 0.001
|
26 |
-
batch_size =
|
27 |
epochs = 10
|
28 |
|
29 |
# Set seed for reproducibility
|
@@ -34,7 +35,7 @@ def train():
|
|
34 |
print(f"Using device: {device}")
|
35 |
|
36 |
# Initialize tensorboard
|
37 |
-
log_dir = 'runs/mnist_experiment_' + datetime.now().strftime('%Y%m%d-%H%M%S')
|
38 |
writer = SummaryWriter(log_dir)
|
39 |
|
40 |
# Setup data
|
@@ -44,6 +45,7 @@ def train():
|
|
44 |
# Initialize model, optimizer, and loss function
|
45 |
model = MNISTModel().to(device)
|
46 |
optimizer = optim.Adam(model.parameters(), lr=learning_rate)
|
|
|
47 |
criterion = nn.CrossEntropyLoss()
|
48 |
|
49 |
|
@@ -53,6 +55,7 @@ def train():
|
|
53 |
running_loss = 0.0
|
54 |
correct = 0
|
55 |
total = 0
|
|
|
56 |
|
57 |
with tqdm(total=len(train_loader), desc=f"Epoch {epoch+1}/{num_epochs}", unit="batch") as pbar:
|
58 |
for batch_idx, batch in enumerate(train_loader):
|
@@ -76,11 +79,13 @@ def train():
|
|
76 |
total += labels.size(0)
|
77 |
correct += predicted.eq(labels).sum().item()
|
78 |
|
|
|
79 |
# Update tqdm progress bar
|
80 |
pbar.set_postfix({
|
81 |
'loss': running_loss / (batch_idx + 1),
|
82 |
'accuracy': 100. * correct / total,
|
83 |
-
'step': batch_idx + 1
|
|
|
84 |
})
|
85 |
pbar.update(1)
|
86 |
|
@@ -93,6 +98,10 @@ def train():
|
|
93 |
epoch * len(train_loader) + batch_idx)
|
94 |
running_loss = 0.0
|
95 |
|
|
|
|
|
|
|
|
|
96 |
# Validation phase
|
97 |
model.eval()
|
98 |
test_loss = 0
|
|
|
10 |
import random
|
11 |
import numpy as np
|
12 |
from tqdm import tqdm
|
13 |
+
from torch.optim.lr_scheduler import StepLR
|
14 |
|
15 |
def set_seed(seed):
|
16 |
torch.manual_seed(seed)
|
|
|
24 |
def train():
|
25 |
# Training loop
|
26 |
learning_rate = 0.001
|
27 |
+
batch_size = 64
|
28 |
epochs = 10
|
29 |
|
30 |
# Set seed for reproducibility
|
|
|
35 |
print(f"Using device: {device}")
|
36 |
|
37 |
# Initialize tensorboard
|
38 |
+
log_dir = 'runs/mnist_experiment_' + f"lr{learning_rate}_bs{batch_size}_ep{epochs}_" + datetime.now().strftime('%Y%m%d-%H%M%S')
|
39 |
writer = SummaryWriter(log_dir)
|
40 |
|
41 |
# Setup data
|
|
|
45 |
# Initialize model, optimizer, and loss function
|
46 |
model = MNISTModel().to(device)
|
47 |
optimizer = optim.Adam(model.parameters(), lr=learning_rate)
|
48 |
+
scheduler = StepLR(optimizer, step_size=2, gamma=0.5) # Decay LR by a factor of 0.1 every 2 epochs
|
49 |
criterion = nn.CrossEntropyLoss()
|
50 |
|
51 |
|
|
|
55 |
running_loss = 0.0
|
56 |
correct = 0
|
57 |
total = 0
|
58 |
+
current_lr = optimizer.param_groups[0]['lr'] # Get current learning rate
|
59 |
|
60 |
with tqdm(total=len(train_loader), desc=f"Epoch {epoch+1}/{num_epochs}", unit="batch") as pbar:
|
61 |
for batch_idx, batch in enumerate(train_loader):
|
|
|
79 |
total += labels.size(0)
|
80 |
correct += predicted.eq(labels).sum().item()
|
81 |
|
82 |
+
|
83 |
# Update tqdm progress bar
|
84 |
pbar.set_postfix({
|
85 |
'loss': running_loss / (batch_idx + 1),
|
86 |
'accuracy': 100. * correct / total,
|
87 |
+
'step': batch_idx + 1,
|
88 |
+
'lr': current_lr,
|
89 |
})
|
90 |
pbar.update(1)
|
91 |
|
|
|
98 |
epoch * len(train_loader) + batch_idx)
|
99 |
running_loss = 0.0
|
100 |
|
101 |
+
writer.add_scalar('learning rate', current_lr, epoch)
|
102 |
+
|
103 |
+
scheduler.step() # Update the learning rate
|
104 |
+
|
105 |
# Validation phase
|
106 |
model.eval()
|
107 |
test_loss = 0
|
models/mnist_model_lr0.001_bs32_ep10.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d77abc86d2b18d3a3b4b7a560186ac1695d0c2e8e708028dd2b65211cefde6ae
|
3 |
+
size 4803144
|
models/mnist_model_lr0.001_bs64_ep10.pth
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 4803144
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b7ed0e38b2b8663379c3655a73c1e6f7e4165bf7d9f792491c7cc9fa99e1e97f
|
3 |
size 4803144
|