dataset, [train_size, test_size])
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=0)
test_loader = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=0)
## LOAD previous cnn
# cnn.load(PATH_TO_LOAD)
## TRAIN and SAVE the CNN
train_result = train_cnn(epochs, cnn, criterion, train_loader, optimizer,
batch_rate)
cnn.save(PATH_TO_SAVE)
## TEST the CNN
test_cnn(cnn, test_loader)
## CHECK the CNN
check_cnn(cnn, test_loader)
## Show the respective graphs
show_statistics(cnn, test_loader, train_result)
## Generate the scv with test data
generate(cnn)
## Make random prediction with the cnn
Inception(512, 112, (144, 288), (32, 64), 64),
Inception(528, 256, (160, 320), (32, 128), 128),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
b5 = nn.Sequential(Inception(832, 256, (160, 320), (32, 128), 128),
Inception(832, 384, (192, 384), (48, 128), 128),
utils.GlobalAvgPool2d())
print('查看网络结构')
net = nn.Sequential(b1, b2, b3, b4, b5, utils.FlattenLayer(),
nn.Linear(1024, 10))
X = torch.rand(1, 1, 96, 96)
for blk in net.children():
X = blk(X)
print('output shape: ', X.shape)
print('获取和读取数据，这里缩减尺寸为 96')
batch_size = 256
train_iter, test_iter = utils.load_data_fashion_mnist(batch_size, resize=96)
print('训练模型，只 1 轮')
lr, num_epochs = 0.002, 1
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
utils.train_cnn(net, train_iter, test_iter, batch_size, optimizer, device,
num_epochs)
'''
训练模型，只 1 轮
training on cpu
epoch 1, loss 0.0058, train acc 0.414, test acc 0.754, time 2282.1 sec
'''
def train():
focus = 'histogram' if request.form['focus'] == 'Histograma' else 'pixel'
n_neighbors = request.form['n_neighbors']
max_iter_svm = request.form['max_iter_svm']
hidden_layer_sizes = request.form['hidden_layer_sizes']
max_iter_bpnn = request.form['max_iter_bpnn']
epochs = request.form['epochs']
steps_per_epoch = request.form['steps_per_epoch']
validation_steps = request.form['validations_steps']
training_steps = request.form['training_steps']
class_positive = request.form['class_positive'].upper()
class_negative = request.form['class_negative'].upper()
response_cnn = utils.train_cnn(steps_per_epoch, epochs, validation_steps,
class_positive, class_negative)
if response_cnn['success'] == True:
response_svm_knn_bpnn = utils.train_svm_knn_bpnn(
n_neighbors, focus, hidden_layer_sizes, max_iter_bpnn,
max_iter_svm)
if response_svm_knn_bpnn['success'] == True:
if app.config['RUN_IMAGE_RETRAINING'] == True:
response_image_retraining = utils.train_image_retraining(
training_steps)
if response_image_retraining['success'] == True:
return render_template(
'form_predict.html',
val_loss_cnn=round(
Decimal(response_cnn['val_loss']) * 100),
val_acc_cnn=round(
Decimal(response_cnn['val_acc']) * 100),
loss_cnn=round(Decimal(response_cnn['loss']) * 100),
acc_cnn=round(Decimal(response_cnn['acc']) * 100),
positive_class=response_cnn['positive_class'],
negative_class=response_cnn['negative_class'],
length_images=response_svm_knn_bpnn['length_images'],
size_package=response_svm_knn_bpnn['size_package'],
accuracy_knn=response_svm_knn_bpnn['accuracy_knn'],
accuracy_bpnn=response_svm_knn_bpnn['accuracy_bpnn'],
accuracy_svm=response_svm_knn_bpnn['accuracy_svm'])
else:
return jsonify({
'success': False,
'msg': 'error-train-image-retraining'
})
else:
return render_template(
'form_predict.html',
val_loss_cnn=round(
Decimal(response_cnn['val_loss']) * 100),
val_acc_cnn=round(Decimal(response_cnn['val_acc']) * 100),
loss_cnn=round(Decimal(response_cnn['loss']) * 100),
acc_cnn=round(Decimal(response_cnn['acc']) * 100),
positive_class=response_cnn['positive_class'],
negative_class=response_cnn['negative_class'],
length_images=response_svm_knn_bpnn['length_images'],
size_package=response_svm_knn_bpnn['size_package'],
accuracy_knn=response_svm_knn_bpnn['accuracy_knn'],
accuracy_bpnn=response_svm_knn_bpnn['accuracy_bpnn'],
accuracy_svm=response_svm_knn_bpnn['accuracy_svm'])
else:
return jsonify({
'success': False,
'msg': 'error-train-svm-knn-bpnn'
})
else:
return jsonify({'success': False, 'msg': 'error-train-cnn'})