In [1]:
import numpy as np
import torch
from sklearn.metrics import confusion_matrix
from mnist_ten.data import test_loader
from mnist_ten.models import classifier, weights_path
from tqdm.auto import tqdm
import matplotlib.pyplot as plt
In [2]:
classifier.load_state_dict(torch.load(weights_path))
classifier.eval()
classifier
Out[2]:
Sequential(
(0): Sequential(
(0): Conv2d(1, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): LeakyReLU(negative_slope=0.01)
(2): SqueezeLayer(
(short): AvgPool2d(kernel_size=2, stride=2, padding=0)
(long): Sequential(
(0): Conv2d(64, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): LeakyReLU(negative_slope=0.01)
(2): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): LeakyReLU(negative_slope=0.01)
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
)
(3): SqueezeLayer(
(short): AvgPool2d(kernel_size=2, stride=2, padding=0)
(long): Sequential(
(0): Conv2d(64, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): LeakyReLU(negative_slope=0.01)
(2): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): LeakyReLU(negative_slope=0.01)
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
)
(4): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1))
(5): LeakyReLU(negative_slope=0.01)
(6): Flatten()
)
(1): Sequential(
(0): Linear(in_features=3136, out_features=64, bias=True)
(1): LeakyReLU(negative_slope=0.01)
(2): Linear(in_features=64, out_features=10, bias=True)
)
)
In [3]:
real_labels = []
predicted_labels = []
for images, labels in tqdm(test_loader):
real_labels += list(labels)
predictions = classifier(images).detach().numpy()
predicted_labels += np.argmax(predictions, axis=1).tolist()
real_labels = np.array(real_labels)
predicted_labels = np.array(predicted_labels)
HBox(children=(FloatProgress(value=0.0, max=625.0), HTML(value='')))
In [4]:
print(f'Accuracy: {np.mean(predicted_labels == real_labels)}')
Accuracy: 0.5263
In [5]:
confusion = confusion_matrix(real_labels, predicted_labels)
plt.xlabel('Predicted label')
plt.ylabel('Real label')
plt.imshow(confusion)
plt.colorbar()
plt.show()
