Below code is mainly from @ptrblck (not my code) and modification and help from a friend.
#model_ft = model_ft.cuda()
nb_samples = 931
nb_classes = 9
from __future__ import print_function, division
import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
import numpy as np
import torchvision
from torchvision import datasets, models, transforms
import matplotlib.pyplot as plt
import time
import os
import copy
import torch.utils.data as data_utils
from torch.utils import data
data_transforms = {
'train': transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.RandomRotation(20),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
'test': transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
}
val_loader = data.DataLoader(
image_datasets['test'],
num_workers=2,
batch_size=1
)
val_loader = iter(val_loader)
data_dir = "images"
image_datasets = {x: datasets.ImageFolder(os.path.join(data_dir, x),
data_transforms[x])
for x in ['train', 'test']}
dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'test']}
print(dataset_sizes)
class_names = image_datasets['train'].classes
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# LOOCV
loocv_preds = []
loocv_targets = []
for idx in range(nb_samples):
print('Using sample {} as test data'.format(idx))
# Get all indices and remove test sample
train_indices = list(range(len(image_datasets)))
del train_indices[idx]
# Create new sampler
sampler = data.SubsetRandomSampler(train_indices)
dataloader = data.DataLoader(
image_datasets['train'],
num_workers=2,
batch_size=1,
sampler=sampler
)
# Train model
for batch_idx, (samples, target) in enumerate(dataloader):
print('Batch {}'.format(batch_idx))
model_ft = train_model(model_ft, criterion, optimizer_ft, exp_lr_scheduler, num_epochs=25) # do I add this line here?
# Test on LOO sample
model_ft.eval()
# test_data, test_target = image_datasets['train'][idx]
test_data, test_target = val_loader.next()
test_data = test_data.cuda()
test_target = test_target.cuda()
#test_data.unsqueeze_(1)
#test_target.unsqueeze_(0)
output = model_ft(test_data)
pred = torch.argmax(output, 1)
loocv_preds.append(pred)
loocv_targets.append(test_target.item())
Error is:
{'train': 791, 'test': 140}
Using sample 0 as test data
Batch 0
Epoch 0/24
----------
train Loss: 2.0170 Acc: 0.3009
test Loss: 2.2370 Acc: 0.3571
Epoch 1/24
----------
train Loss: 2.0436 Acc: 0.3236
test Loss: 1.8071 Acc: 0.4071
Epoch 2/24
----------
train Loss: 1.9560 Acc: 0.3300
test Loss: 1.9911 Acc: 0.3714
Epoch 3/24
----------
train Loss: 1.9052 Acc: 0.3413
test Loss: 2.4123 Acc: 0.3286
Epoch 4/24
----------
train Loss: 1.8917 Acc: 0.3692
test Loss: 2.8163 Acc: 0.3929
Epoch 5/24
----------
train Loss: 1.8890 Acc: 0.3515
test Loss: 2.3379 Acc: 0.3429
Epoch 6/24
----------
train Loss: 1.9090 Acc: 0.3388
test Loss: 2.3404 Acc: 0.3929
Epoch 7/24
----------
train Loss: 1.6069 Acc: 0.4475
test Loss: 2.0714 Acc: 0.4143
Epoch 8/24
----------
train Loss: 1.4701 Acc: 0.5019
test Loss: 2.1390 Acc: 0.4071
Epoch 9/24
----------
train Loss: 1.5322 Acc: 0.4867
test Loss: 2.0545 Acc: 0.4286
Epoch 10/24
----------
train Loss: 1.4743 Acc: 0.5082
test Loss: 2.0200 Acc: 0.4143
Epoch 11/24
----------
train Loss: 1.4498 Acc: 0.5006
test Loss: 2.0408 Acc: 0.4071
Epoch 12/24
----------
train Loss: 1.4314 Acc: 0.5095
test Loss: 2.0807 Acc: 0.4214
Epoch 13/24
----------
train Loss: 1.4097 Acc: 0.5133
test Loss: 2.1612 Acc: 0.4214
Epoch 14/24
----------
train Loss: 1.4116 Acc: 0.5234
test Loss: 2.0989 Acc: 0.4000
Epoch 15/24
----------
train Loss: 1.3736 Acc: 0.5234
test Loss: 2.1113 Acc: 0.4214
Epoch 16/24
----------
train Loss: 1.3562 Acc: 0.5310
test Loss: 2.1046 Acc: 0.4143
Epoch 17/24
----------
train Loss: 1.3730 Acc: 0.5297
test Loss: 2.1644 Acc: 0.4214
Epoch 18/24
----------
train Loss: 1.3351 Acc: 0.5398
test Loss: 2.1070 Acc: 0.4357
Epoch 19/24
----------
train Loss: 1.3228 Acc: 0.5449
test Loss: 2.1646 Acc: 0.4214
Epoch 20/24
----------
train Loss: 1.3837 Acc: 0.5272
test Loss: 2.1686 Acc: 0.4214
Epoch 21/24
----------
train Loss: 1.3377 Acc: 0.5424
test Loss: 2.1626 Acc: 0.4143
Epoch 22/24
----------
train Loss: 1.3879 Acc: 0.5158
test Loss: 2.1593 Acc: 0.4286
Epoch 23/24
----------
train Loss: 1.3329 Acc: 0.5563
test Loss: 2.2069 Acc: 0.4357
Epoch 24/24
----------
train Loss: 1.3181 Acc: 0.5613
test Loss: 2.1064 Acc: 0.4143
Training complete in 5m 5s
Using sample 1 as test data
Batch 0
Epoch 0/24
----------
train Loss: 1.3487 Acc: 0.5310
test Loss: 2.1172 Acc: 0.4286
Epoch 1/24
----------
train Loss: 1.3218 Acc: 0.5474
test Loss: 2.0875 Acc: 0.4214
Epoch 2/24
----------
train Loss: 1.3528 Acc: 0.5436
test Loss: 2.2111 Acc: 0.4286
Epoch 3/24
----------
train Loss: 1.3242 Acc: 0.5499
test Loss: 2.1504 Acc: 0.4286
Epoch 4/24
----------
train Loss: 1.3318 Acc: 0.5499
test Loss: 2.1848 Acc: 0.4286
Epoch 5/24
----------
train Loss: 1.3252 Acc: 0.5449
test Loss: 2.1637 Acc: 0.4357
Epoch 6/24
----------
train Loss: 1.3135 Acc: 0.5424
test Loss: 2.0848 Acc: 0.4357
Epoch 7/24
----------
train Loss: 1.3492 Acc: 0.5373
test Loss: 2.1031 Acc: 0.4071
Epoch 8/24
----------
train Loss: 1.2792 Acc: 0.5575
test Loss: 2.1071 Acc: 0.4286
Epoch 9/24
----------
train Loss: 1.3436 Acc: 0.5386
test Loss: 2.1486 Acc: 0.4500
Epoch 10/24
----------
train Loss: 1.3596 Acc: 0.5373
test Loss: 2.1443 Acc: 0.4214
Epoch 11/24
----------
train Loss: 1.3750 Acc: 0.5499
test Loss: 2.1618 Acc: 0.4214
Epoch 12/24
----------
train Loss: 1.3241 Acc: 0.5424
test Loss: 2.1119 Acc: 0.4214
Epoch 13/24
----------
train Loss: 1.3706 Acc: 0.5335
test Loss: 2.0601 Acc: 0.4143
Epoch 14/24
----------
train Loss: 1.3505 Acc: 0.5525
test Loss: 2.0726 Acc: 0.4214
Epoch 15/24
----------
train Loss: 1.3460 Acc: 0.5550
test Loss: 2.0473 Acc: 0.4429
Epoch 16/24
----------
train Loss: 1.3337 Acc: 0.5563
test Loss: 2.0629 Acc: 0.4143
Epoch 17/24
----------
train Loss: 1.3146 Acc: 0.5601
test Loss: 2.1087 Acc: 0.4214
Epoch 18/24
----------
train Loss: 1.3320 Acc: 0.5512
test Loss: 2.1290 Acc: 0.4286
Epoch 19/24
----------
train Loss: 1.3393 Acc: 0.5601
test Loss: 2.0548 Acc: 0.4143
Epoch 20/24
----------
train Loss: 1.3232 Acc: 0.5474
test Loss: 2.0624 Acc: 0.4429
Epoch 21/24
----------
train Loss: 1.3537 Acc: 0.5386
test Loss: 2.0688 Acc: 0.4286
Epoch 22/24
----------
train Loss: 1.2933 Acc: 0.5664
test Loss: 2.1751 Acc: 0.4143
Epoch 23/24
----------
train Loss: 1.3747 Acc: 0.5424
test Loss: 2.1009 Acc: 0.4143
Epoch 24/24
----------
train Loss: 1.3455 Acc: 0.5183
test Loss: 2.1395 Acc: 0.4143
Training complete in 5m 10s
Using sample 2 as test data
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
<ipython-input-10-15f05fc7d5ca> in <module>()
68 # Get all indices and remove test sample
69 train_indices = list(range(len(image_datasets)))
---> 70 del train_indices[idx]
71
72 # Create new sampler
IndexError: list assignment index out of range
How can we fix this error?
Code is using segments from the below fine-tuning tutorial:
https://pytorch.org/tutorials/beginner/transfer_learning_tutorial.html