Deep Learning Models -- A collection of various deep learning architectures, models, and tips for TensorFlow and PyTorch in Jupyter Notebooks.
- Author: Sebastian Raschka
- GitHub Repository: https://github.com/rasbt/deeplearning-models
In [1]:
%load_ext watermark
%watermark -a 'Sebastian Raschka' -v -p torch
Sebastian Raschka CPython 3.7.3 IPython 7.9.0 torch 1.7.0
AlexNet CIFAR-10 Classifier¶
Network Architecture¶
References
- [1] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "Imagenet classification with deep convolutional neural networks." In Advances in Neural Information Processing Systems, pp. 1097-1105. 2012.
Imports¶
In [2]:
import os
import time
import random
import numpy as np
import pandas as pd
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import DataLoader
from torch.utils.data.dataset import Subset
from torchvision import datasets
from torchvision import transforms
import matplotlib.pyplot as plt
from PIL import Image
if torch.cuda.is_available():
torch.backends.cudnn.deterministic = True
Model Settings¶
Setting a random seed¶
I recommend using a function like the following one prior to using dataset loaders and initializing a model if you want to ensure the data is shuffled in the same manner if you rerun this notebook and the model gets the same initial random weights:
In [3]:
def set_all_seeds(seed):
os.environ["PL_GLOBAL_SEED"] = str(seed)
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
Setting cuDNN and PyTorch algorithmic behavior to deterministic¶
Similar to the set_all_seeds function above, I recommend setting the behavior of PyTorch and cuDNN to deterministic (this is particulary relevant when using GPUs). We can also define a function for that:
In [4]:
def set_deterministic():
if torch.cuda.is_available():
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
torch.set_deterministic(True)
In [5]:
##########################
### SETTINGS
##########################
# Hyperparameters
RANDOM_SEED = 1
LEARNING_RATE = 0.0001
BATCH_SIZE = 256
NUM_EPOCHS = 40
# Architecture
NUM_CLASSES = 10
# Other
DEVICE = "cuda:0"
set_all_seeds(RANDOM_SEED)
# Deterministic behavior not yet supported by AdaptiveAvgPool2d
#set_deterministic()
Import utility functions¶
In [6]:
import sys
sys.path.insert(0, "..") # to include ../helper_evaluate.py etc.
from helper_evaluate import compute_accuracy
from helper_data import get_dataloaders_cifar10
from helper_train import train_classifier_simple_v1
Dataset¶
In [7]:
### Set random seed ###
set_all_seeds(RANDOM_SEED)
##########################
### Dataset
##########################
train_transforms = transforms.Compose([transforms.Resize((70, 70)),
transforms.RandomCrop((64, 64)),
transforms.ToTensor()])
test_transforms = transforms.Compose([transforms.Resize((70, 70)),
transforms.CenterCrop((64, 64)),
transforms.ToTensor()])
train_loader, valid_loader, test_loader = get_dataloaders_cifar10(
batch_size=BATCH_SIZE,
num_workers=2,
train_transforms=train_transforms,
test_transforms=test_transforms,
validation_fraction=0.1)
Files already downloaded and verified
In [8]:
# Checking the dataset
print('Training Set:\n')
for images, labels in train_loader:
print('Image batch dimensions:', images.size())
print('Image label dimensions:', labels.size())
print(labels[:10])
break
# Checking the dataset
print('\nValidation Set:')
for images, labels in valid_loader:
print('Image batch dimensions:', images.size())
print('Image label dimensions:', labels.size())
print(labels[:10])
break
# Checking the dataset
print('\nTesting Set:')
for images, labels in train_loader:
print('Image batch dimensions:', images.size())
print('Image label dimensions:', labels.size())
print(labels[:10])
break
Training Set: Image batch dimensions: torch.Size([256, 3, 64, 64]) Image label dimensions: torch.Size([256]) tensor([6, 9, 9, 4, 1, 1, 2, 7, 8, 3]) Validation Set: Image batch dimensions: torch.Size([256, 3, 64, 64]) Image label dimensions: torch.Size([256]) tensor([7, 1, 4, 1, 0, 2, 2, 5, 9, 6]) Testing Set: Image batch dimensions: torch.Size([256, 3, 64, 64]) Image label dimensions: torch.Size([256]) tensor([6, 9, 9, 4, 1, 1, 2, 7, 8, 3])
Model¶
In [9]:
##########################
### MODEL
##########################
class AlexNet(nn.Module):
def __init__(self, num_classes):
super(AlexNet, self).__init__()
self.features = nn.Sequential(
nn.Conv2d(3, 64, kernel_size=11, stride=4, padding=2),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=3, stride=2),
nn.Conv2d(64, 192, kernel_size=5, padding=2),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=3, stride=2),
nn.Conv2d(192, 384, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(384, 256, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(256, 256, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=3, stride=2),
)
self.avgpool = nn.AdaptiveAvgPool2d((6, 6))
self.classifier = nn.Sequential(
nn.Dropout(0.5),
nn.Linear(256 * 6 * 6, 4096),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Linear(4096, 4096),
nn.ReLU(inplace=True),
nn.Linear(4096, num_classes)
)
def forward(self, x):
x = self.features(x)
x = self.avgpool(x)
x = x.view(x.size(0), 256 * 6 * 6)
logits = self.classifier(x)
probas = F.softmax(logits, dim=1)
return logits
In [10]:
torch.manual_seed(RANDOM_SEED)
model = AlexNet(NUM_CLASSES)
model.to(DEVICE)
optimizer = torch.optim.Adam(model.parameters(), lr=LEARNING_RATE)
Training¶
In [11]:
log_dict = train_classifier_simple_v1(num_epochs=NUM_EPOCHS, model=model,
optimizer=optimizer, device=DEVICE,
train_loader=train_loader, valid_loader=valid_loader,
logging_interval=50)
Epoch: 001/040 | Batch 0000/0176 | Loss: 2.3030 Epoch: 001/040 | Batch 0050/0176 | Loss: 1.9805 Epoch: 001/040 | Batch 0100/0176 | Loss: 1.8457 Epoch: 001/040 | Batch 0150/0176 | Loss: 1.7629 Epoch: 001/040 | Train Acc.: 33.087% | Loss: 1.697 Epoch: 001/040 | Validation Acc.: 34.160% | Loss: 1.671 Time elapsed: 0.44 min Epoch: 002/040 | Batch 0000/0176 | Loss: 1.7060 Epoch: 002/040 | Batch 0050/0176 | Loss: 1.5886 Epoch: 002/040 | Batch 0100/0176 | Loss: 1.6655 Epoch: 002/040 | Batch 0150/0176 | Loss: 1.4965 Epoch: 002/040 | Train Acc.: 41.529% | Loss: 1.521 Epoch: 002/040 | Validation Acc.: 41.880% | Loss: 1.511 Time elapsed: 0.88 min Epoch: 003/040 | Batch 0000/0176 | Loss: 1.4985 Epoch: 003/040 | Batch 0050/0176 | Loss: 1.4119 Epoch: 003/040 | Batch 0100/0176 | Loss: 1.3107 Epoch: 003/040 | Batch 0150/0176 | Loss: 1.2836 Epoch: 003/040 | Train Acc.: 47.098% | Loss: 1.406 Epoch: 003/040 | Validation Acc.: 47.520% | Loss: 1.400 Time elapsed: 1.32 min Epoch: 004/040 | Batch 0000/0176 | Loss: 1.4325 Epoch: 004/040 | Batch 0050/0176 | Loss: 1.2214 Epoch: 004/040 | Batch 0100/0176 | Loss: 1.1575 Epoch: 004/040 | Batch 0150/0176 | Loss: 1.2118 Epoch: 004/040 | Train Acc.: 55.071% | Loss: 1.224 Epoch: 004/040 | Validation Acc.: 54.980% | Loss: 1.229 Time elapsed: 1.77 min Epoch: 005/040 | Batch 0000/0176 | Loss: 1.1903 Epoch: 005/040 | Batch 0050/0176 | Loss: 1.1410 Epoch: 005/040 | Batch 0100/0176 | Loss: 1.0611 Epoch: 005/040 | Batch 0150/0176 | Loss: 1.0817 Epoch: 005/040 | Train Acc.: 57.938% | Loss: 1.144 Epoch: 005/040 | Validation Acc.: 57.640% | Loss: 1.147 Time elapsed: 2.21 min Epoch: 006/040 | Batch 0000/0176 | Loss: 1.1344 Epoch: 006/040 | Batch 0050/0176 | Loss: 1.0903 Epoch: 006/040 | Batch 0100/0176 | Loss: 0.9820 Epoch: 006/040 | Batch 0150/0176 | Loss: 1.0716 Epoch: 006/040 | Train Acc.: 60.222% | Loss: 1.104 Epoch: 006/040 | Validation Acc.: 58.700% | Loss: 1.120 Time elapsed: 2.65 min Epoch: 007/040 | Batch 0000/0176 | Loss: 1.0856 Epoch: 007/040 | Batch 0050/0176 | Loss: 0.9864 Epoch: 007/040 | Batch 0100/0176 | Loss: 0.9559 Epoch: 007/040 | Batch 0150/0176 | Loss: 0.9611 Epoch: 007/040 | Train Acc.: 63.953% | Loss: 1.008 Epoch: 007/040 | Validation Acc.: 61.700% | Loss: 1.055 Time elapsed: 3.09 min Epoch: 008/040 | Batch 0000/0176 | Loss: 0.9608 Epoch: 008/040 | Batch 0050/0176 | Loss: 0.9866 Epoch: 008/040 | Batch 0100/0176 | Loss: 0.9418 Epoch: 008/040 | Batch 0150/0176 | Loss: 0.9364 Epoch: 008/040 | Train Acc.: 65.887% | Loss: 0.946 Epoch: 008/040 | Validation Acc.: 63.220% | Loss: 1.006 Time elapsed: 3.53 min Epoch: 009/040 | Batch 0000/0176 | Loss: 0.9465 Epoch: 009/040 | Batch 0050/0176 | Loss: 0.9415 Epoch: 009/040 | Batch 0100/0176 | Loss: 0.9320 Epoch: 009/040 | Batch 0150/0176 | Loss: 0.8411 Epoch: 009/040 | Train Acc.: 67.782% | Loss: 0.908 Epoch: 009/040 | Validation Acc.: 64.740% | Loss: 0.995 Time elapsed: 3.97 min Epoch: 010/040 | Batch 0000/0176 | Loss: 0.8873 Epoch: 010/040 | Batch 0050/0176 | Loss: 0.9821 Epoch: 010/040 | Batch 0100/0176 | Loss: 0.8360 Epoch: 010/040 | Batch 0150/0176 | Loss: 0.7949 Epoch: 010/040 | Train Acc.: 71.416% | Loss: 0.806 Epoch: 010/040 | Validation Acc.: 67.360% | Loss: 0.912 Time elapsed: 4.41 min Epoch: 011/040 | Batch 0000/0176 | Loss: 0.7469 Epoch: 011/040 | Batch 0050/0176 | Loss: 0.8446 Epoch: 011/040 | Batch 0100/0176 | Loss: 0.7797 Epoch: 011/040 | Batch 0150/0176 | Loss: 0.7813 Epoch: 011/040 | Train Acc.: 72.213% | Loss: 0.779 Epoch: 011/040 | Validation Acc.: 67.640% | Loss: 0.913 Time elapsed: 4.85 min Epoch: 012/040 | Batch 0000/0176 | Loss: 0.7076 Epoch: 012/040 | Batch 0050/0176 | Loss: 0.8669 Epoch: 012/040 | Batch 0100/0176 | Loss: 0.7664 Epoch: 012/040 | Batch 0150/0176 | Loss: 0.7205 Epoch: 012/040 | Train Acc.: 73.053% | Loss: 0.762 Epoch: 012/040 | Validation Acc.: 67.440% | Loss: 0.918 Time elapsed: 5.29 min Epoch: 013/040 | Batch 0000/0176 | Loss: 0.6941 Epoch: 013/040 | Batch 0050/0176 | Loss: 0.7947 Epoch: 013/040 | Batch 0100/0176 | Loss: 0.7407 Epoch: 013/040 | Batch 0150/0176 | Loss: 0.6998 Epoch: 013/040 | Train Acc.: 75.658% | Loss: 0.686 Epoch: 013/040 | Validation Acc.: 69.320% | Loss: 0.883 Time elapsed: 5.74 min Epoch: 014/040 | Batch 0000/0176 | Loss: 0.6040 Epoch: 014/040 | Batch 0050/0176 | Loss: 0.7107 Epoch: 014/040 | Batch 0100/0176 | Loss: 0.6712 Epoch: 014/040 | Batch 0150/0176 | Loss: 0.6610 Epoch: 014/040 | Train Acc.: 74.813% | Loss: 0.725 Epoch: 014/040 | Validation Acc.: 68.140% | Loss: 0.937 Time elapsed: 6.18 min Epoch: 015/040 | Batch 0000/0176 | Loss: 0.7104 Epoch: 015/040 | Batch 0050/0176 | Loss: 0.6965 Epoch: 015/040 | Batch 0100/0176 | Loss: 0.6800 Epoch: 015/040 | Batch 0150/0176 | Loss: 0.6326 Epoch: 015/040 | Train Acc.: 76.784% | Loss: 0.660 Epoch: 015/040 | Validation Acc.: 69.560% | Loss: 0.926 Time elapsed: 6.62 min Epoch: 016/040 | Batch 0000/0176 | Loss: 0.5889 Epoch: 016/040 | Batch 0050/0176 | Loss: 0.6394 Epoch: 016/040 | Batch 0100/0176 | Loss: 0.5344 Epoch: 016/040 | Batch 0150/0176 | Loss: 0.5194 Epoch: 016/040 | Train Acc.: 78.216% | Loss: 0.620 Epoch: 016/040 | Validation Acc.: 69.060% | Loss: 0.925 Time elapsed: 7.06 min Epoch: 017/040 | Batch 0000/0176 | Loss: 0.4660 Epoch: 017/040 | Batch 0050/0176 | Loss: 0.6028 Epoch: 017/040 | Batch 0100/0176 | Loss: 0.5386 Epoch: 017/040 | Batch 0150/0176 | Loss: 0.5476 Epoch: 017/040 | Train Acc.: 79.882% | Loss: 0.563 Epoch: 017/040 | Validation Acc.: 70.940% | Loss: 0.900 Time elapsed: 7.50 min Epoch: 018/040 | Batch 0000/0176 | Loss: 0.4768 Epoch: 018/040 | Batch 0050/0176 | Loss: 0.6519 Epoch: 018/040 | Batch 0100/0176 | Loss: 0.5177 Epoch: 018/040 | Batch 0150/0176 | Loss: 0.5190 Epoch: 018/040 | Train Acc.: 79.673% | Loss: 0.575 Epoch: 018/040 | Validation Acc.: 69.540% | Loss: 0.961 Time elapsed: 7.94 min Epoch: 019/040 | Batch 0000/0176 | Loss: 0.4714 Epoch: 019/040 | Batch 0050/0176 | Loss: 0.5929 Epoch: 019/040 | Batch 0100/0176 | Loss: 0.5053 Epoch: 019/040 | Batch 0150/0176 | Loss: 0.5581 Epoch: 019/040 | Train Acc.: 81.029% | Loss: 0.535 Epoch: 019/040 | Validation Acc.: 70.120% | Loss: 0.971 Time elapsed: 8.38 min Epoch: 020/040 | Batch 0000/0176 | Loss: 0.3960 Epoch: 020/040 | Batch 0050/0176 | Loss: 0.5640 Epoch: 020/040 | Batch 0100/0176 | Loss: 0.5285 Epoch: 020/040 | Batch 0150/0176 | Loss: 0.4373 Epoch: 020/040 | Train Acc.: 83.053% | Loss: 0.497 Epoch: 020/040 | Validation Acc.: 71.060% | Loss: 0.936 Time elapsed: 8.82 min Epoch: 021/040 | Batch 0000/0176 | Loss: 0.3562 Epoch: 021/040 | Batch 0050/0176 | Loss: 0.4917 Epoch: 021/040 | Batch 0100/0176 | Loss: 0.5012 Epoch: 021/040 | Batch 0150/0176 | Loss: 0.5080 Epoch: 021/040 | Train Acc.: 82.573% | Loss: 0.489 Epoch: 021/040 | Validation Acc.: 70.460% | Loss: 0.943 Time elapsed: 9.26 min Epoch: 022/040 | Batch 0000/0176 | Loss: 0.4068 Epoch: 022/040 | Batch 0050/0176 | Loss: 0.4859 Epoch: 022/040 | Batch 0100/0176 | Loss: 0.4582 Epoch: 022/040 | Batch 0150/0176 | Loss: 0.4259 Epoch: 022/040 | Train Acc.: 83.702% | Loss: 0.468 Epoch: 022/040 | Validation Acc.: 71.920% | Loss: 0.940 Time elapsed: 9.70 min Epoch: 023/040 | Batch 0000/0176 | Loss: 0.4236 Epoch: 023/040 | Batch 0050/0176 | Loss: 0.4781 Epoch: 023/040 | Batch 0100/0176 | Loss: 0.4485 Epoch: 023/040 | Batch 0150/0176 | Loss: 0.3734 Epoch: 023/040 | Train Acc.: 83.813% | Loss: 0.461 Epoch: 023/040 | Validation Acc.: 70.860% | Loss: 0.958 Time elapsed: 10.14 min Epoch: 024/040 | Batch 0000/0176 | Loss: 0.3325 Epoch: 024/040 | Batch 0050/0176 | Loss: 0.3642 Epoch: 024/040 | Batch 0100/0176 | Loss: 0.3749 Epoch: 024/040 | Batch 0150/0176 | Loss: 0.3309 Epoch: 024/040 | Train Acc.: 85.018% | Loss: 0.425 Epoch: 024/040 | Validation Acc.: 71.560% | Loss: 0.965 Time elapsed: 10.58 min Epoch: 025/040 | Batch 0000/0176 | Loss: 0.3156 Epoch: 025/040 | Batch 0050/0176 | Loss: 0.3344 Epoch: 025/040 | Batch 0100/0176 | Loss: 0.3945 Epoch: 025/040 | Batch 0150/0176 | Loss: 0.3581 Epoch: 025/040 | Train Acc.: 84.291% | Loss: 0.450 Epoch: 025/040 | Validation Acc.: 69.320% | Loss: 1.077 Time elapsed: 11.02 min Epoch: 026/040 | Batch 0000/0176 | Loss: 0.3812 Epoch: 026/040 | Batch 0050/0176 | Loss: 0.3231 Epoch: 026/040 | Batch 0100/0176 | Loss: 0.4320 Epoch: 026/040 | Batch 0150/0176 | Loss: 0.3551 Epoch: 026/040 | Train Acc.: 83.344% | Loss: 0.467 Epoch: 026/040 | Validation Acc.: 69.180% | Loss: 1.108 Time elapsed: 11.45 min Epoch: 027/040 | Batch 0000/0176 | Loss: 0.3591 Epoch: 027/040 | Batch 0050/0176 | Loss: 0.3114 Epoch: 027/040 | Batch 0100/0176 | Loss: 0.2481 Epoch: 027/040 | Batch 0150/0176 | Loss: 0.3428 Epoch: 027/040 | Train Acc.: 84.931% | Loss: 0.433 Epoch: 027/040 | Validation Acc.: 69.600% | Loss: 1.126 Time elapsed: 11.90 min Epoch: 028/040 | Batch 0000/0176 | Loss: 0.3464 Epoch: 028/040 | Batch 0050/0176 | Loss: 0.2866 Epoch: 028/040 | Batch 0100/0176 | Loss: 0.2776 Epoch: 028/040 | Batch 0150/0176 | Loss: 0.3536 Epoch: 028/040 | Train Acc.: 86.289% | Loss: 0.399 Epoch: 028/040 | Validation Acc.: 70.460% | Loss: 1.114 Time elapsed: 12.34 min Epoch: 029/040 | Batch 0000/0176 | Loss: 0.3019 Epoch: 029/040 | Batch 0050/0176 | Loss: 0.3655 Epoch: 029/040 | Batch 0100/0176 | Loss: 0.2486 Epoch: 029/040 | Batch 0150/0176 | Loss: 0.4025 Epoch: 029/040 | Train Acc.: 89.231% | Loss: 0.308 Epoch: 029/040 | Validation Acc.: 71.820% | Loss: 1.050 Time elapsed: 12.78 min Epoch: 030/040 | Batch 0000/0176 | Loss: 0.2827 Epoch: 030/040 | Batch 0050/0176 | Loss: 0.3269 Epoch: 030/040 | Batch 0100/0176 | Loss: 0.2699 Epoch: 030/040 | Batch 0150/0176 | Loss: 0.3385 Epoch: 030/040 | Train Acc.: 89.320% | Loss: 0.308 Epoch: 030/040 | Validation Acc.: 71.940% | Loss: 1.063 Time elapsed: 13.22 min Epoch: 031/040 | Batch 0000/0176 | Loss: 0.2108 Epoch: 031/040 | Batch 0050/0176 | Loss: 0.2322 Epoch: 031/040 | Batch 0100/0176 | Loss: 0.2662 Epoch: 031/040 | Batch 0150/0176 | Loss: 0.1860 Epoch: 031/040 | Train Acc.: 88.807% | Loss: 0.316 Epoch: 031/040 | Validation Acc.: 71.720% | Loss: 1.073 Time elapsed: 13.66 min Epoch: 032/040 | Batch 0000/0176 | Loss: 0.2714 Epoch: 032/040 | Batch 0050/0176 | Loss: 0.3018 Epoch: 032/040 | Batch 0100/0176 | Loss: 0.2323 Epoch: 032/040 | Batch 0150/0176 | Loss: 0.2668 Epoch: 032/040 | Train Acc.: 91.173% | Loss: 0.249 Epoch: 032/040 | Validation Acc.: 71.980% | Loss: 1.025 Time elapsed: 14.10 min Epoch: 033/040 | Batch 0000/0176 | Loss: 0.1913 Epoch: 033/040 | Batch 0050/0176 | Loss: 0.3458 Epoch: 033/040 | Batch 0100/0176 | Loss: 0.1746 Epoch: 033/040 | Batch 0150/0176 | Loss: 0.2072 Epoch: 033/040 | Train Acc.: 91.171% | Loss: 0.245 Epoch: 033/040 | Validation Acc.: 72.680% | Loss: 1.060 Time elapsed: 14.54 min Epoch: 034/040 | Batch 0000/0176 | Loss: 0.2103 Epoch: 034/040 | Batch 0050/0176 | Loss: 0.2894 Epoch: 034/040 | Batch 0100/0176 | Loss: 0.2032 Epoch: 034/040 | Batch 0150/0176 | Loss: 0.2118 Epoch: 034/040 | Train Acc.: 91.756% | Loss: 0.239 Epoch: 034/040 | Validation Acc.: 73.200% | Loss: 1.038 Time elapsed: 14.98 min Epoch: 035/040 | Batch 0000/0176 | Loss: 0.2010 Epoch: 035/040 | Batch 0050/0176 | Loss: 0.2106 Epoch: 035/040 | Batch 0100/0176 | Loss: 0.1802 Epoch: 035/040 | Batch 0150/0176 | Loss: 0.2216 Epoch: 035/040 | Train Acc.: 91.060% | Loss: 0.264 Epoch: 035/040 | Validation Acc.: 71.280% | Loss: 1.146 Time elapsed: 15.42 min Epoch: 036/040 | Batch 0000/0176 | Loss: 0.1478 Epoch: 036/040 | Batch 0050/0176 | Loss: 0.1735 Epoch: 036/040 | Batch 0100/0176 | Loss: 0.1186 Epoch: 036/040 | Batch 0150/0176 | Loss: 0.1835 Epoch: 036/040 | Train Acc.: 91.376% | Loss: 0.254 Epoch: 036/040 | Validation Acc.: 72.160% | Loss: 1.181 Time elapsed: 15.86 min Epoch: 037/040 | Batch 0000/0176 | Loss: 0.1154 Epoch: 037/040 | Batch 0050/0176 | Loss: 0.1817 Epoch: 037/040 | Batch 0100/0176 | Loss: 0.1166 Epoch: 037/040 | Batch 0150/0176 | Loss: 0.1973 Epoch: 037/040 | Train Acc.: 91.278% | Loss: 0.259 Epoch: 037/040 | Validation Acc.: 71.480% | Loss: 1.241 Time elapsed: 16.31 min Epoch: 038/040 | Batch 0000/0176 | Loss: 0.1402 Epoch: 038/040 | Batch 0050/0176 | Loss: 0.1672 Epoch: 038/040 | Batch 0100/0176 | Loss: 0.1366 Epoch: 038/040 | Batch 0150/0176 | Loss: 0.1037 Epoch: 038/040 | Train Acc.: 91.447% | Loss: 0.256 Epoch: 038/040 | Validation Acc.: 71.260% | Loss: 1.256 Time elapsed: 16.75 min Epoch: 039/040 | Batch 0000/0176 | Loss: 0.1505 Epoch: 039/040 | Batch 0050/0176 | Loss: 0.1537 Epoch: 039/040 | Batch 0100/0176 | Loss: 0.1592 Epoch: 039/040 | Batch 0150/0176 | Loss: 0.1947 Epoch: 039/040 | Train Acc.: 92.571% | Loss: 0.223 Epoch: 039/040 | Validation Acc.: 72.040% | Loss: 1.266 Time elapsed: 17.19 min Epoch: 040/040 | Batch 0000/0176 | Loss: 0.1707 Epoch: 040/040 | Batch 0050/0176 | Loss: 0.1500 Epoch: 040/040 | Batch 0100/0176 | Loss: 0.1422 Epoch: 040/040 | Batch 0150/0176 | Loss: 0.1099 Epoch: 040/040 | Train Acc.: 95.191% | Loss: 0.139 Epoch: 040/040 | Validation Acc.: 72.740% | Loss: 1.201 Time elapsed: 17.63 min Total Training Time: 17.63 min
Evaluation¶
In [12]:
import matplotlib.pyplot as plt
%matplotlib inline
In [13]:
plt.plot(log_dict['train_loss_per_batch'], label='Minibatch cost')
plt.plot(np.convolve(cost_list,
np.ones(200,)/200, mode='valid'),
label='Running average')
plt.ylabel('Cross Entropy')
plt.xlabel('Iteration')
plt.legend()
plt.show()
--------------------------------------------------------------------------- NameError Traceback (most recent call last) <ipython-input-13-550123761591> in <module> 1 plt.plot(log_dict['train_loss_per_batch'], label='Minibatch cost') ----> 2 plt.plot(np.convolve(cost_list, 3 np.ones(200,)/200, mode='valid'), 4 label='Running average') 5 NameError: name 'cost_list' is not defined
In [ ]:
plt.plot(np.arange(1, NUM_EPOCHS+1), log_dict['train_acc_per_batch'], label='Training')
plt.plot(np.arange(1, NUM_EPOCHS+1), log_dict['valid_acc_per_batch'], label='Validation')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.legend()
plt.show()
In [ ]:
with torch.set_grad_enabled(False):
train_acc = compute_accuracy(model=model,
data_loader=test_loader,
device=DEVICE)
test_acc = compute_accuracy(model=model,
data_loader=test_loader,
device=DEVICE)
valid_acc = compute_accuracy(model=model,
data_loader=valid_loader,
device=DEVICE)
print(f'Train ACC: {valid_acc:.2f}%')
print(f'Validation ACC: {valid_acc:.2f}%')
print(f'Test ACC: {test_acc:.2f}%')
In [ ]:
%watermark -iv