In [168]:
# !pip install aeon
In [169]:
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import torch
import torch.nn as nn
import torch
from torch import nn
import torch.nn.functional as F
from skorch import NeuralNetClassifier
from skorch.callbacks import EpochScoring
from skorch.dataset import ValidSplit
from sklearn.metrics import accuracy_score
from skorch.callbacks import LRScheduler, EarlyStopping, Checkpoint
DEVICE = "mps" if torch.backends.mps.is_available() else "cuda" if torch.cuda.is_available() else "cpu"
2D Convolutional Neural Networks¶
DermaMNIST is a small benchmark dataset from MedMNIST for skin-lesion image classification. It is based on HAM10000 and contains 10,015 dermatoscopic images of pigmented skin lesions, resized to MNIST-like $28 \times 28$ color images. The task is 7-class multi-class classification, with official train, validation, and test splits of 7,007, 1,003, and 2,005 images.
Labels are:
- akiec: Actinic keratoses and intraepithelial carcinoma
- bcc: Basal cell carcinoma
- bkl: Benign keratosis-like lesions
- df: Dermatofibroma
- nv: Melanocytic nevi
- vasc: Vascular lesions
- mel: Melanoma
In [170]:
files = np.load("Data/dermamnist.npz")
train_images = files["train_images"]
train_labels = files["train_labels"]
validation_images = files["val_images"]
validation_labels = files["val_labels"]
test_images = files["test_images"]
test_labels = files["test_labels"]
train_images.shape, train_labels.shape, validation_images.shape, validation_labels.shape, test_images.shape, test_labels.shape
Out[170]:
((7007, 28, 28, 3), (7007, 1), (1003, 28, 28, 3), (1003, 1), (2005, 28, 28, 3), (2005, 1))
In [171]:
np.unique(train_labels)
Out[171]:
array([0, 1, 2, 3, 4, 5, 6], dtype=uint8)
In [172]:
i = 0
plt.imshow(train_images[i])
plt.title(train_labels[i])
Out[172]:
Text(0.5, 1.0, '[0]')
In [173]:
train_images = train_images.astype(np.float32).transpose(0, 3, 1, 2) / 255.0
validation_images = validation_images.astype(np.float32).transpose(0, 3, 1, 2) / 255.0
test_images = test_images.astype(np.float32).transpose(0, 3, 1, 2) / 255.0
train_labels = train_labels.ravel()
validation_labels = validation_labels.ravel()
test_labels = test_labels.ravel()
train_images.shape, train_labels.shape, validation_images.shape, validation_labels.shape, test_images.shape, test_labels.shape
Out[173]:
((7007, 3, 28, 28), (7007,), (1003, 3, 28, 28), (1003,), (2005, 3, 28, 28), (2005,))
In [174]:
from sklearn.dummy import DummyClassifier
dummy_clf = DummyClassifier(strategy="most_frequent")
dummy_clf.fit(train_images, train_labels)
dummy_clf.score(test_images, test_labels)
Out[174]:
0.6688279301745635
In [175]:
class Conv2DClassifier(nn.Module):
def __init__(self, in_channels: int, num_classes: int):
super().__init__()
self.features = nn.Sequential(
nn.Conv2d(
in_channels=in_channels,
out_channels=32,
kernel_size=3,
padding=1
),
nn.BatchNorm2d(32),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(
in_channels=32,
out_channels=64,
kernel_size=3,
padding=1
),
nn.BatchNorm2d(64),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(
in_channels=64,
out_channels=128,
kernel_size=3,
padding=1
),
nn.BatchNorm2d(128),
nn.ReLU()
)
self.pool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
self.classifier = nn.Sequential(
nn.Flatten(),
nn.Linear(128, num_classes)
)
def forward(self, x):
"""
x shape: (batch_size, in_channels, height, width)
"""
x = self.features(x)
x = self.pool(x)
x = self.classifier(x)
return x
In [176]:
validation_images.shape, validation_labels.shape
Out[176]:
((1003, 3, 28, 28), (1003,))
In [177]:
from skorch.helper import predefined_split
from torch.utils.data import TensorDataset
valid_ds = TensorDataset(torch.Tensor(validation_images), torch.Tensor(validation_labels))
In [ ]:
Vanilla CNN¶
In [178]:
net = NeuralNetClassifier(
Conv2DClassifier,
module__in_channels=train_images.shape[1],
module__num_classes=np.unique(train_labels).shape[0],
criterion=nn.CrossEntropyLoss,
optimizer=torch.optim.Adam,
max_epochs=10,
batch_size=32,
lr=1e-4,
callbacks=[
EpochScoring(
scoring='accuracy',
name='train_acc',
on_train=True,
),
Checkpoint(
monitor="valid_acc_best",
load_best=True,
f_history=None,
f_optimizer=None,
)
],
device=DEVICE,
train_split=predefined_split(valid_ds)
)
net
Out[178]:
<class 'skorch.classifier.NeuralNetClassifier'>[uninitialized]( module=<class '__main__.Conv2DClassifier'>, module__in_channels=3, module__num_classes=7, )In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
<class 'skorch.classifier.NeuralNetClassifier'>[uninitialized]( module=<class '__main__.Conv2DClassifier'>, module__in_channels=3, module__num_classes=7, )
In [179]:
net.fit(train_images, train_labels)
epoch train_acc train_loss valid_acc valid_loss cp dur
------- ----------- ------------ ----------- ------------ ---- ------
1 0.5807 1.2946 0.6979 0.9958 + 2.6244
2 0.6956 0.9130 0.7149 0.8589 + 2.4661
3 0.7089 0.8358 0.7298 0.8131 + 2.4988
4 0.7153 0.8031 0.7358 0.7858 + 2.7542
5 0.7211 0.7824 0.7388 0.7711 + 2.7795
6 0.7250 0.7671 0.7438 0.7595 + 2.5161
7 0.7280 0.7544 0.7488 0.7529 + 2.4621
8 0.7293 0.7432 0.7488 0.7434 2.6305
9 0.7313 0.7331 0.7478 0.7349 2.4915
10 0.7330 0.7240 0.7507 0.7301 + 2.4509
Out[179]:
<class 'skorch.classifier.NeuralNetClassifier'>[initialized](
module_=Conv2DClassifier(
(features): Sequential(
(0): Conv2d(3, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
(3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(4): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(6): ReLU()
(7): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(8): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(9): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(10): ReLU()
)
(pool): AdaptiveAvgPool2d(output_size=(1, 1))
(classifier): Sequential(
(0): Flatten(start_dim=1, end_dim=-1)
(1): Linear(in_features=128, out_features=7, bias=True)
)
),
)In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook. On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
<class 'skorch.classifier.NeuralNetClassifier'>[initialized](
module_=Conv2DClassifier(
(features): Sequential(
(0): Conv2d(3, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
(3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(4): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(6): ReLU()
(7): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(8): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(9): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(10): ReLU()
)
(pool): AdaptiveAvgPool2d(output_size=(1, 1))
(classifier): Sequential(
(0): Flatten(start_dim=1, end_dim=-1)
(1): Linear(in_features=128, out_features=7, bias=True)
)
),
)In [180]:
y_pred = net.predict(test_images)
accuracy_score(test_labels, y_pred)
Out[180]:
0.7276807980049875
In [181]:
history = net.history
plt.plot(history[:, 'train_loss'], label='train_loss')
plt.plot(history[:, 'valid_loss'], label='valid_loss')
plt.legend()
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.show()
plt.plot(history[:, 'train_acc'], label='train_acc')
plt.plot(history[:, 'valid_acc'], label='valid_acc')
plt.legend()
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.show()
Resnet¶
In [182]:
from torchvision.models import resnet18, ResNet18_Weights
class ResNet18Classifier(nn.Module):
def __init__(self, num_classes, pretrained=True, freeze_features=True):
super().__init__()
weights = ResNet18_Weights.DEFAULT if pretrained else None
self.model = resnet18(weights=weights)
if freeze_features:
for name, param in self.model.named_parameters():
if not name.startswith("fc."):
param.requires_grad = False
in_features = self.model.fc.in_features
self.model.fc = nn.Linear(in_features, num_classes)
def forward(self, x):
return self.model(x)
In [184]:
net = NeuralNetClassifier(
ResNet18Classifier,
module__num_classes=np.unique(train_labels).shape[0],
criterion=nn.CrossEntropyLoss,
optimizer=torch.optim.Adam,
max_epochs=10,
batch_size=32,
lr=1e-4,
callbacks=[
EpochScoring(
scoring='accuracy',
name='train_acc',
on_train=True,
),
Checkpoint(
monitor="valid_acc_best",
load_best=True,
f_history=None,
f_optimizer=None,
)
],
device=DEVICE,
train_split=predefined_split(valid_ds)
)
net
Out[184]:
<class 'skorch.classifier.NeuralNetClassifier'>[uninitialized]( module=<class '__main__.ResNet18Classifier'>, module__num_classes=7, )In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
<class 'skorch.classifier.NeuralNetClassifier'>[uninitialized]( module=<class '__main__.ResNet18Classifier'>, module__num_classes=7, )
In [185]:
net.fit(train_images, train_labels)
epoch train_acc train_loss valid_acc valid_loss cp dur
------- ----------- ------------ ----------- ------------ ---- ------
1 0.5477 1.4547 0.6550 1.2086 + 3.7200
2 0.6586 1.1212 0.6630 1.1236 + 3.9270
3 0.6652 1.0523 0.6670 1.0686 + 3.5048
4 0.6716 1.0047 0.6730 1.0314 + 3.6844
5 0.6769 0.9698 0.6780 1.0052 + 3.7846
6 0.6806 0.9429 0.6830 0.9861 + 3.5550
7 0.6835 0.9214 0.6830 0.9719 3.5019
8 0.6869 0.9036 0.6830 0.9610 3.5804
9 0.6897 0.8886 0.6869 0.9526 + 3.9056
10 0.6923 0.8757 0.6859 0.9460 3.9266
Out[185]:
<class 'skorch.classifier.NeuralNetClassifier'>[initialized](
module_=ResNet18Classifier(
(model): ResNet(
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer2): Sequential(
(0): BasicBlock(
(conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer3): Sequential(
(0): BasicBlock(
(conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer4): Sequential(
(0): BasicBlock(
(conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(fc): Linear(in_features=512, out_features=7, bias=True)
)
),
)In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook. On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
<class 'skorch.classifier.NeuralNetClassifier'>[initialized](
module_=ResNet18Classifier(
(model): ResNet(
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer2): Sequential(
(0): BasicBlock(
(conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer3): Sequential(
(0): BasicBlock(
(conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer4): Sequential(
(0): BasicBlock(
(conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(fc): Linear(in_features=512, out_features=7, bias=True)
)
),
)In [186]:
y_pred = net.predict(test_images)
accuracy_score(test_labels, y_pred)
Out[186]:
0.686284289276808
In [187]:
history = net.history
plt.plot(history[:, 'train_loss'], label='train_loss')
plt.plot(history[:, 'valid_loss'], label='valid_loss')
plt.legend()
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.show()
plt.plot(history[:, 'train_acc'], label='train_acc')
plt.plot(history[:, 'valid_acc'], label='valid_acc')
plt.legend()
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.show()
1D Convolutional Neural Networks¶
In [4]:
from aeon.datasets import load_classification
Dataset¶
Heartbeat
This dataset is derived from the PhysioNet/CinC Challenge 2016. Heart sound recordings were sourced from several contributors around the world, collected at either a clinical or nonclinical environment, from both healthy subjects and pathological patients. The heart sound recordings were collected from different locations on the body. The typical four locations are aortic area, pulmonic area, tricuspid area and mitral area, but could be one of nine different locations. The sounds were divided into two classes: normal and abnormal. The normal recordings were from healthy subjects and the abnormal ones were from patients with a confirmed cardiac diagnosis. The patients suffer from a variety of illnesses, but typically they are heart valve defects and coronary artery disease patients. Heart valve defects include mitral valve prolapse, mitral regurgitation, aortic stenosis and valvular surgery. All the recordings from the patients were generally labeled as abnormal. Both healthy subjects and pathological patients include both children and adults. Each recording was truncated to 5 seconds. A Spectrogram of each instance was then created with a window size of 0.061 seconds and an overlap of 70%. Each instance in this multivariate dataset is arranged such that each dimension is a frequency band from the spectrogram. The two classes normal and abnormal consist of 113 and 296 respectivley.
In [18]:
# Import the dataset
X_train, y_train = load_classification("Heartbeat", split="train")
X_test, y_test = load_classification("Heartbeat", split="test")
X_train = X_train.astype(np.float32)
X_test = X_test.astype(np.float32)
X_train.shape, y_train.shape, X_test.shape, y_test.shape
Out[18]:
((204, 61, 405), (204,), (205, 61, 405), (205,))
In [19]:
# Encode the labels
from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()
y_train = le.fit_transform(y_train)
y_test = le.transform(y_test)
LABELS = le.classes_
le.classes_
Out[19]:
array(['abnormal', 'normal'], dtype='<U8')
In [34]:
i = 100
plt.imshow(X_train[i], aspect='auto', cmap='viridis')
plt.suptitle('Class: {}'.format(LABELS[y_train[i]]), fontsize=16)
plt.show()
In [36]:
from sklearn.dummy import DummyClassifier
dummy_clf = DummyClassifier(strategy="most_frequent")
dummy_clf.fit(X_train, y_train)
dummy_clf.score(X_test, y_test)
Out[36]:
0.7219512195121951
In [37]:
class Conv1DClassifier(nn.Module):
def __init__(self, in_channels: int, num_classes: int):
super().__init__()
self.features = nn.Sequential(
nn.Conv1d(
in_channels=in_channels,
out_channels=32,
kernel_size=7,
padding=3
),
nn.BatchNorm1d(32),
nn.ReLU(),
nn.MaxPool1d(kernel_size=2),
nn.Conv1d(
in_channels=32,
out_channels=64,
kernel_size=5,
padding=2
),
nn.BatchNorm1d(64),
nn.ReLU(),
nn.MaxPool1d(kernel_size=2),
nn.Conv1d(
in_channels=64,
out_channels=128,
kernel_size=3,
padding=1
),
nn.BatchNorm1d(128),
nn.ReLU()
)
self.pool = nn.AdaptiveAvgPool1d(output_size=1)
self.classifier = nn.Sequential(
nn.Flatten(),
nn.Linear(128, num_classes)
)
def forward(self, x):
"""
x shape: (batch_size, in_channels, sequence_length)
"""
x = self.features(x)
x = self.pool(x)
x = self.classifier(x)
return x
In [88]:
net = NeuralNetClassifier(
Conv1DClassifier,
module__in_channels=X_train.shape[1],
module__num_classes=len(le.classes_),
criterion=nn.CrossEntropyLoss,
optimizer=torch.optim.Adam,
max_epochs=300,
batch_size=128,
lr=1e-4,
callbacks=[
EpochScoring(
scoring='accuracy',
name='train_acc',
on_train=True,
),
Checkpoint(
monitor="valid_acc_best",
load_best=True,
f_history=None,
f_optimizer=None,
)
],
device=DEVICE,
train_split=ValidSplit(cv=0.2, stratified=True)
)
net
Out[88]:
<class 'skorch.classifier.NeuralNetClassifier'>[uninitialized]( module=<class '__main__.Conv1DClassifier'>, module__in_channels=61, module__num_classes=2, )In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
<class 'skorch.classifier.NeuralNetClassifier'>[uninitialized]( module=<class '__main__.Conv1DClassifier'>, module__in_channels=61, module__num_classes=2, )
In [89]:
net.fit(X_train, y_train)
epoch train_acc train_loss valid_acc valid_loss cp dur
------- ----------- ------------ ----------- ------------ ---- ------
1 0.3497 0.7176 0.2683 0.6971 + 0.0698
2 0.5215 0.6985 0.2683 0.7001 0.0317
3 0.5460 0.6841 0.2683 0.7029 0.0324
4 0.6074 0.6725 0.2683 0.7050 0.0383
5 0.6319 0.6629 0.3415 0.7066 + 0.0412
6 0.6503 0.6546 0.4390 0.7079 + 0.0287
7 0.6871 0.6474 0.4878 0.7090 + 0.0440
8 0.7117 0.6410 0.5366 0.7099 + 0.0391
9 0.7178 0.6351 0.5854 0.7107 + 0.0407
10 0.7178 0.6294 0.6341 0.7115 + 0.0315
11 0.7178 0.6239 0.6341 0.7122 0.0343
12 0.7178 0.6184 0.6341 0.7123 0.0371
13 0.7301 0.6130 0.6585 0.7117 + 0.0412
14 0.7362 0.6076 0.6829 0.7103 + 0.0390
15 0.7423 0.6024 0.6829 0.7079 0.0299
16 0.7362 0.5973 0.7073 0.7045 + 0.0320
17 0.7362 0.5923 0.7073 0.7001 0.0343
18 0.7362 0.5876 0.7073 0.6950 0.0375
19 0.7423 0.5830 0.7073 0.6895 0.0439
20 0.7423 0.5786 0.7073 0.6837 0.0412
21 0.7423 0.5744 0.7073 0.6779 0.0298
22 0.7423 0.5703 0.7073 0.6719 0.0315
23 0.7423 0.5663 0.6829 0.6657 0.0357
24 0.7423 0.5625 0.6829 0.6594 0.0468
25 0.7423 0.5587 0.6829 0.6530 0.0548
26 0.7423 0.5551 0.6829 0.6467 0.0449
27 0.7485 0.5516 0.6829 0.6405 0.0354
28 0.7546 0.5482 0.6829 0.6342 0.0339
29 0.7546 0.5449 0.6829 0.6279 0.0356
30 0.7546 0.5416 0.6829 0.6215 0.0412
31 0.7485 0.5384 0.7073 0.6149 0.0742
32 0.7485 0.5352 0.7073 0.6083 0.0444
33 0.7546 0.5320 0.6829 0.6019 0.0308
34 0.7546 0.5289 0.6829 0.5959 0.0319
35 0.7546 0.5257 0.7073 0.5903 0.0355
36 0.7546 0.5227 0.7073 0.5852 0.0361
37 0.7607 0.5196 0.7073 0.5805 0.0392
38 0.7669 0.5166 0.7073 0.5766 0.0377
39 0.7669 0.5136 0.7073 0.5732 0.0289
40 0.7669 0.5107 0.7073 0.5702 0.0309
41 0.7669 0.5078 0.7073 0.5674 0.0319
42 0.7669 0.5049 0.7073 0.5649 0.0428
43 0.7669 0.5020 0.7073 0.5625 0.0469
44 0.7669 0.4991 0.7073 0.5598 0.0504
45 0.7669 0.4962 0.7073 0.5567 0.0335
46 0.7730 0.4934 0.7073 0.5530 0.0326
47 0.7791 0.4905 0.7073 0.5491 0.0397
48 0.7791 0.4876 0.7073 0.5452 0.0418
49 0.7730 0.4847 0.7073 0.5412 0.0298
50 0.7791 0.4818 0.7073 0.5372 0.0309
51 0.7791 0.4788 0.7073 0.5332 0.0320
52 0.7791 0.4758 0.7073 0.5295 0.0357
53 0.7791 0.4728 0.7073 0.5261 0.0355
54 0.7853 0.4698 0.7073 0.5230 0.0425
55 0.7914 0.4668 0.7073 0.5197 0.0469
56 0.7975 0.4637 0.7073 0.5160 0.0337
57 0.7975 0.4606 0.7073 0.5117 0.0360
58 0.7975 0.4574 0.7073 0.5069 0.0438
59 0.7975 0.4542 0.7317 0.5020 + 0.0398
60 0.8037 0.4510 0.7561 0.4971 + 0.0323
61 0.8037 0.4477 0.7317 0.4924 0.0328
62 0.8098 0.4445 0.7317 0.4879 0.0368
63 0.8160 0.4412 0.7561 0.4833 0.0461
64 0.8160 0.4380 0.7561 0.4786 0.0376
65 0.8160 0.4347 0.7805 0.4740 + 0.0286
66 0.8160 0.4314 0.7805 0.4698 0.0312
67 0.8160 0.4281 0.7805 0.4661 0.0325
68 0.8221 0.4248 0.7805 0.4631 0.0365
69 0.8221 0.4216 0.7805 0.4606 0.0395
70 0.8405 0.4184 0.7805 0.4582 0.0432
71 0.8466 0.4151 0.7805 0.4559 0.0420
72 0.8466 0.4119 0.7805 0.4538 0.0308
73 0.8528 0.4087 0.7805 0.4520 0.0311
74 0.8528 0.4055 0.7805 0.4504 0.0321
75 0.8466 0.4024 0.7805 0.4491 0.0332
76 0.8466 0.3992 0.7805 0.4479 0.0368
77 0.8466 0.3962 0.8049 0.4466 + 0.0402
78 0.8466 0.3931 0.8049 0.4453 0.0408
79 0.8528 0.3901 0.8049 0.4441 0.0296
80 0.8528 0.3871 0.8049 0.4430 0.0317
81 0.8528 0.3841 0.8049 0.4421 0.0320
82 0.8712 0.3812 0.8049 0.4411 0.0342
83 0.8773 0.3784 0.8049 0.4402 0.0371
84 0.8896 0.3755 0.8293 0.4392 + 0.0445
85 0.8896 0.3727 0.8293 0.4383 0.0747
86 0.8896 0.3700 0.8293 0.4374 0.0485
87 0.8896 0.3672 0.8293 0.4366 0.0380
88 0.9018 0.3645 0.8293 0.4358 0.0294
89 0.9018 0.3618 0.8293 0.4349 0.0326
90 0.9018 0.3591 0.8293 0.4341 0.0342
91 0.9018 0.3564 0.8293 0.4334 0.0399
92 0.9080 0.3537 0.8537 0.4326 + 0.0507
93 0.9080 0.3510 0.8537 0.4318 0.0394
94 0.9080 0.3483 0.8537 0.4313 0.0422
95 0.9141 0.3457 0.8537 0.4309 0.0431
96 0.9141 0.3431 0.8537 0.4302 0.0377
97 0.9141 0.3405 0.8537 0.4291 0.0417
98 0.9141 0.3380 0.8537 0.4283 0.0397
99 0.9141 0.3354 0.8537 0.4275 0.0294
100 0.9141 0.3329 0.8537 0.4269 0.0319
101 0.9141 0.3303 0.8537 0.4262 0.0325
102 0.9264 0.3278 0.8537 0.4256 0.0359
103 0.9264 0.3252 0.8537 0.4251 0.0376
104 0.9264 0.3227 0.8537 0.4245 0.0432
105 0.9325 0.3201 0.8537 0.4239 0.0385
106 0.9325 0.3175 0.8537 0.4235 0.0302
107 0.9325 0.3150 0.8537 0.4231 0.0342
108 0.9387 0.3125 0.8537 0.4227 0.0345
109 0.9387 0.3100 0.8537 0.4225 0.0399
110 0.9387 0.3075 0.8537 0.4219 0.0464
111 0.9387 0.3051 0.8537 0.4216 0.0425
112 0.9448 0.3026 0.8537 0.4214 0.0314
113 0.9448 0.3002 0.8537 0.4208 0.0323
114 0.9448 0.2977 0.8537 0.4202 0.0347
115 0.9509 0.2953 0.8537 0.4199 0.0371
116 0.9509 0.2929 0.8537 0.4194 0.0413
117 0.9509 0.2904 0.8537 0.4186 0.0404
118 0.9509 0.2880 0.8537 0.4186 0.0303
119 0.9509 0.2856 0.8537 0.4181 0.0318
120 0.9509 0.2832 0.8537 0.4176 0.0320
121 0.9509 0.2808 0.8537 0.4176 0.0368
122 0.9509 0.2785 0.8537 0.4169 0.0361
123 0.9509 0.2761 0.8293 0.4167 0.0433
124 0.9509 0.2738 0.8293 0.4165 0.0398
125 0.9448 0.2714 0.8293 0.4160 0.0290
126 0.9509 0.2690 0.8293 0.4163 0.0314
127 0.9571 0.2667 0.8049 0.4158 0.0418
128 0.9571 0.2643 0.8049 0.4156 0.0401
129 0.9571 0.2619 0.8049 0.4157 0.0458
130 0.9571 0.2595 0.8049 0.4149 0.0453
131 0.9571 0.2572 0.8049 0.4155 0.0457
132 0.9571 0.2549 0.8049 0.4146 0.0460
133 0.9571 0.2525 0.8049 0.4148 0.0455
134 0.9571 0.2502 0.8049 0.4149 0.0380
135 0.9571 0.2479 0.8049 0.4143 0.0296
136 0.9571 0.2456 0.8049 0.4149 0.0320
137 0.9571 0.2434 0.8293 0.4139 0.0337
138 0.9571 0.2411 0.8049 0.4151 0.0360
139 0.9571 0.2388 0.8293 0.4138 0.0409
140 0.9632 0.2365 0.8049 0.4155 0.0409
141 0.9693 0.2342 0.8293 0.4135 0.0299
142 0.9632 0.2320 0.8049 0.4162 0.0310
143 0.9693 0.2297 0.8293 0.4127 0.0329
144 0.9632 0.2275 0.7805 0.4173 0.0343
145 0.9693 0.2252 0.8293 0.4105 0.0368
146 0.9632 0.2231 0.7805 0.4221 0.0442
147 0.9693 0.2208 0.8293 0.4064 0.0434
148 0.9632 0.2193 0.8049 0.4363 0.0357
149 0.9693 0.2174 0.8537 0.4035 0.0592
150 0.9632 0.2168 0.7805 0.4490 0.0472
151 0.9693 0.2142 0.8293 0.4058 0.0300
152 0.9632 0.2114 0.7805 0.4220 0.0334
153 0.9693 0.2080 0.7805 0.4271 0.0356
154 0.9693 0.2060 0.8293 0.4073 0.0376
155 0.9632 0.2049 0.8049 0.4361 0.0420
156 0.9693 0.2023 0.8293 0.4130 0.0458
157 0.9693 0.1999 0.8049 0.4188 0.0453
158 0.9693 0.1976 0.8049 0.4291 0.0385
159 0.9693 0.1957 0.8293 0.4113 0.0295
160 0.9693 0.1942 0.8049 0.4310 0.0427
161 0.9693 0.1918 0.8049 0.4144 0.0350
162 0.9693 0.1898 0.8049 0.4208 0.0386
163 0.9693 0.1876 0.8049 0.4227 0.0443
164 0.9693 0.1856 0.8049 0.4147 0.0402
165 0.9693 0.1839 0.8293 0.4270 0.0296
166 0.9693 0.1818 0.8049 0.4144 0.0310
167 0.9693 0.1800 0.8049 0.4256 0.0323
168 0.9693 0.1778 0.8049 0.4173 0.0337
169 0.9693 0.1759 0.8049 0.4222 0.0368
170 0.9755 0.1739 0.8049 0.4212 0.0418
171 0.9755 0.1719 0.8049 0.4202 0.0421
172 0.9755 0.1700 0.8049 0.4249 0.0330
173 0.9755 0.1681 0.8049 0.4192 0.0407
174 0.9755 0.1663 0.8293 0.4287 0.0440
175 0.9755 0.1642 0.8049 0.4180 0.0414
176 0.9755 0.1626 0.8293 0.4346 0.0369
177 0.9755 0.1606 0.8293 0.4155 0.0420
178 0.9755 0.1595 0.8293 0.4481 0.0420
179 0.9755 0.1579 0.8049 0.4152 0.0292
180 0.9755 0.1579 0.8049 0.4718 0.0315
181 0.9755 0.1567 0.8049 0.4193 0.0344
182 0.9755 0.1561 0.8293 0.4592 0.0372
183 0.9755 0.1517 0.7805 0.4221 0.0438
184 0.9755 0.1483 0.7805 0.4217 0.0407
185 0.9755 0.1468 0.8293 0.4499 0.0306
186 0.9755 0.1458 0.8049 0.4204 0.0330
187 0.9755 0.1446 0.8293 0.4400 0.0342
188 0.9755 0.1417 0.8049 0.4298 0.0401
189 0.9755 0.1397 0.7805 0.4244 0.0486
190 0.9755 0.1386 0.8293 0.4430 0.0426
191 0.9755 0.1371 0.8049 0.4251 0.0315
192 0.9755 0.1357 0.8293 0.4363 0.0315
193 0.9755 0.1336 0.7805 0.4322 0.0350
194 0.9755 0.1320 0.7805 0.4299 0.0420
195 0.9755 0.1308 0.8293 0.4401 0.0413
196 0.9755 0.1293 0.7805 0.4306 0.0376
197 0.9755 0.1280 0.8293 0.4406 0.0287
198 0.9816 0.1263 0.7805 0.4347 0.0308
199 0.9816 0.1249 0.8049 0.4377 0.0319
200 0.9816 0.1235 0.8049 0.4406 0.0371
201 0.9816 0.1221 0.7805 0.4369 0.0358
202 0.9816 0.1208 0.8293 0.4447 0.0439
203 0.9816 0.1195 0.7805 0.4379 0.0394
204 0.9816 0.1183 0.8049 0.4457 0.0300
205 0.9816 0.1169 0.7805 0.4402 0.0329
206 0.9816 0.1156 0.8049 0.4458 0.0337
207 0.9816 0.1143 0.8049 0.4431 0.0365
208 0.9816 0.1131 0.8049 0.4460 0.0408
209 0.9816 0.1118 0.8049 0.4461 0.0407
210 0.9816 0.1106 0.8049 0.4468 0.0301
211 0.9816 0.1094 0.8049 0.4484 0.0309
212 0.9816 0.1082 0.8049 0.4473 0.0320
213 0.9816 0.1070 0.8049 0.4505 0.0336
214 0.9816 0.1058 0.7805 0.4480 0.0363
215 0.9816 0.1048 0.8049 0.4532 0.0413
216 0.9816 0.1036 0.7805 0.4491 0.0458
217 0.9816 0.1027 0.8049 0.4582 0.0457
218 0.9816 0.1016 0.7805 0.4505 0.0698
219 0.9877 0.1011 0.8049 0.4707 0.0454
220 0.9816 0.1005 0.8049 0.4571 0.0450
221 0.9939 0.1012 0.8293 0.5002 0.0446
222 0.9816 0.1017 0.7805 0.4727 0.0365
223 0.9939 0.1030 0.8293 0.5142 0.0288
224 0.9816 0.1009 0.8049 0.4615 0.0306
225 0.9939 0.0967 0.8049 0.4610 0.0317
226 0.9877 0.0936 0.8049 0.4823 0.0360
227 0.9816 0.0938 0.8049 0.4617 0.0376
228 0.9939 0.0940 0.8049 0.4786 0.0440
229 0.9816 0.0914 0.8049 0.4655 0.0454
230 0.9877 0.0898 0.7805 0.4613 0.0302
231 0.9939 0.0897 0.8049 0.4818 0.0312
232 0.9816 0.0890 0.7805 0.4637 0.0332
233 0.9939 0.0876 0.8049 0.4671 0.0363
234 0.9877 0.0864 0.8049 0.4738 0.0402
235 0.9877 0.0858 0.7805 0.4669 0.0397
236 0.9939 0.0853 0.8049 0.4748 0.0290
237 0.9877 0.0841 0.8049 0.4704 0.0304
238 0.9877 0.0832 0.7805 0.4699 0.0321
239 0.9939 0.0825 0.8049 0.4764 0.0332
240 0.9877 0.0818 0.7805 0.4709 0.0370
241 0.9939 0.0810 0.8049 0.4748 0.0432
242 0.9877 0.0801 0.8049 0.4749 0.0458
243 0.9939 0.0793 0.7805 0.4741 0.0451
244 0.9939 0.0786 0.8049 0.4783 0.0460
245 0.9877 0.0779 0.7805 0.4757 0.0455
246 0.9939 0.0772 0.8049 0.4779 0.0443
247 0.9939 0.0764 0.8049 0.4783 0.0390
248 0.9939 0.0757 0.8049 0.4780 0.0292
249 0.9939 0.0750 0.8049 0.4808 0.0311
250 0.9939 0.0743 0.8049 0.4792 0.0331
251 0.9939 0.0737 0.8049 0.4818 0.0353
252 0.9939 0.0729 0.8049 0.4815 0.0399
253 0.9939 0.0723 0.8049 0.4827 0.0455
254 1.0000 0.0716 0.8049 0.4842 0.0450
255 1.0000 0.0710 0.8049 0.4846 0.0444
256 1.0000 0.0703 0.8049 0.4867 0.0384
257 1.0000 0.0697 0.8049 0.4869 0.0292
258 1.0000 0.0691 0.8049 0.4884 0.0320
259 1.0000 0.0684 0.8049 0.4888 0.0415
260 1.0000 0.0678 0.8049 0.4897 0.0404
261 1.0000 0.0672 0.8049 0.4904 0.0404
262 1.0000 0.0666 0.8049 0.4910 0.0299
263 1.0000 0.0660 0.8049 0.4924 0.0314
264 1.0000 0.0654 0.8049 0.4935 0.0315
265 1.0000 0.0648 0.8049 0.4949 0.0336
266 1.0000 0.0643 0.8049 0.4957 0.0369
267 1.0000 0.0637 0.8049 0.4967 0.0417
268 1.0000 0.0631 0.8049 0.4976 0.0457
269 1.0000 0.0626 0.8049 0.4986 0.0458
270 1.0000 0.0620 0.8049 0.4996 0.0455
271 1.0000 0.0615 0.8049 0.5010 0.0387
272 1.0000 0.0609 0.8049 0.5025 0.0297
273 1.0000 0.0604 0.8049 0.5036 0.0320
274 1.0000 0.0599 0.8049 0.5045 0.0333
275 1.0000 0.0593 0.8049 0.5049 0.0357
276 1.0000 0.0588 0.8049 0.5054 0.0404
277 1.0000 0.0583 0.8049 0.5059 0.0411
278 1.0000 0.0578 0.8049 0.5068 0.0296
279 1.0000 0.0573 0.8049 0.5080 0.0312
280 1.0000 0.0568 0.8049 0.5092 0.0324
281 1.0000 0.0564 0.8049 0.5103 0.0348
282 1.0000 0.0559 0.8049 0.5114 0.0375
283 1.0000 0.0554 0.8049 0.5123 0.0443
284 1.0000 0.0549 0.8049 0.5135 0.0398
285 1.0000 0.0545 0.8049 0.5147 0.0298
286 1.0000 0.0540 0.8049 0.5157 0.0309
287 1.0000 0.0536 0.8049 0.5175 0.0316
288 1.0000 0.0531 0.8049 0.5183 0.0337
289 1.0000 0.0527 0.7805 0.5203 0.0368
290 1.0000 0.0523 0.8049 0.5203 0.0422
291 1.0000 0.0518 0.7805 0.5222 0.0453
292 1.0000 0.0515 0.8049 0.5221 0.0462
293 1.0000 0.0511 0.7805 0.5251 0.0463
294 1.0000 0.0509 0.8049 0.5252 0.0473
295 1.0000 0.0505 0.7805 0.5296 0.0451
296 1.0000 0.0505 0.8049 0.5303 0.0383
297 1.0000 0.0502 0.8049 0.5350 0.0683
298 1.0000 0.0503 0.8049 0.5374 0.0396
299 1.0000 0.0501 0.8049 0.5387 0.0285
300 1.0000 0.0501 0.8049 0.5404 0.0304
Out[89]:
<class 'skorch.classifier.NeuralNetClassifier'>[initialized](
module_=Conv1DClassifier(
(features): Sequential(
(0): Conv1d(61, 32, kernel_size=(7,), stride=(1,), padding=(3,))
(1): BatchNorm1d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
(3): MaxPool1d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(4): Conv1d(32, 64, kernel_size=(5,), stride=(1,), padding=(2,))
(5): BatchNorm1d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(6): ReLU()
(7): MaxPool1d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(8): Conv1d(64, 128, kernel_size=(3,), stride=(1,), padding=(1,))
(9): BatchNorm1d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(10): ReLU()
)
(pool): AdaptiveAvgPool1d(output_size=1)
(classifier): Sequential(
(0): Flatten(start_dim=1, end_dim=-1)
(1): Linear(in_features=128, out_features=2, bias=True)
)
),
)In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook. On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
<class 'skorch.classifier.NeuralNetClassifier'>[initialized](
module_=Conv1DClassifier(
(features): Sequential(
(0): Conv1d(61, 32, kernel_size=(7,), stride=(1,), padding=(3,))
(1): BatchNorm1d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
(3): MaxPool1d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(4): Conv1d(32, 64, kernel_size=(5,), stride=(1,), padding=(2,))
(5): BatchNorm1d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(6): ReLU()
(7): MaxPool1d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(8): Conv1d(64, 128, kernel_size=(3,), stride=(1,), padding=(1,))
(9): BatchNorm1d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(10): ReLU()
)
(pool): AdaptiveAvgPool1d(output_size=1)
(classifier): Sequential(
(0): Flatten(start_dim=1, end_dim=-1)
(1): Linear(in_features=128, out_features=2, bias=True)
)
),
)In [90]:
y_pred = net.predict(X_test)
accuracy_score(y_test, y_pred)
Out[90]:
0.7609756097560976
In [91]:
history = net.history
plt.plot(history[:, 'train_loss'], label='train_loss')
plt.plot(history[:, 'valid_loss'], label='valid_loss')
plt.legend()
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.show()
plt.plot(history[:, 'train_acc'], label='train_acc')
plt.plot(history[:, 'valid_acc'], label='valid_acc')
plt.legend()
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.show()
Rocket Variants¶
In [71]:
from aeon.classification.convolution_based import MiniRocketClassifier, MultiRocketHydraClassifier
In [72]:
clf = MiniRocketClassifier(n_jobs=-1)
In [73]:
clf.fit(X_train, y_train)
Out[73]:
MiniRocketClassifier(n_jobs=-1)In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
MiniRocketClassifier(n_jobs=-1)
In [74]:
clf.score(X_test, y_test)
Out[74]:
0.7560975609756098
In [75]:
clf = MultiRocketHydraClassifier(n_jobs=-1)
In [76]:
clf.fit(X_train, y_train)
Out[76]:
MultiRocketHydraClassifier(n_jobs=-1)In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
MultiRocketHydraClassifier(n_jobs=-1)
In [77]:
clf.score(X_test, y_test)
Out[77]:
0.7414634146341463
Exercises¶
- 1D-CNN: build a 1d cnn for the ECG200 dataset (see previous notebooks for how to import it)
- 2D-CNN: build a 2d cnn for the Pneumonia MNIST dataset (https://zenodo.org/records/10519652)
In [ ]:
In [ ]: