def train_model():
dataloader = load_dataset(train=True)
model = cnn.Net()
model.to(cnn.get_device())
model.train()
model, dictionary = cnn.train_model(model=model,
dataloader=dataloader,
num_epochs=10)
print("Time spent: {:.2f}s".format(dictionary['exec_time']))
torch.save(model.state_dict(), "./modelo/mnist.pt")
save_stats(dictionary)
def train_cnn(rootpath, dataset, dataloader, device):
"""
Trains the convolutional Neural Network
"""
num_classes = len(dataset.classes)
model, input_size = cnn.initialize_model("resnet",
num_classes,
resume_from=None)
criterion = cnn.get_loss()
optimizer = cnn.make_optimizer(model)
scheduler = cnn.make_scheduler(optimizer)
num_epochs = 20
save_dir = os.path.join(rootpath, "weights")
trained_model, val_acc_history = cnn.train_model(model,
device,
dataloader,
criterion,
optimizer,
scheduler=scheduler,
save_dir=save_dir,
num_epochs=num_epochs)
return trained_model, val_acc_history
def main():
args = init()
logger = logging.getLogger(__name__)
###########
# Load the Data
###########
data_path, gt_path = args["dataset"], args["groundTruth"]
if not Path(data_path).absolute().is_file() or not Path(gt_path).absolute().is_file():
logger.critical("Dataset or GT do not exist!")
data, ground_truth = load_mat(data_path, gt_path)
logger.info("Loaded data...")
############
# Extract Patches
############
window_size = args["windowSize"]
stride = args["stride"]
maxPatches = args["maxPatches"] if "maxPatches" in args else None
samples, labels = extract_patches(data, ground_truth, window_size, stride, maxPatches=maxPatches)
logger.info("Extracted {} patches".format(len(samples)))
###########
# ACE If necessary
###########
ACE = args["ace"]
cpu_count = args["cpu_count"] if "cpu_count" in args else os.cpu_count()
if ACE:
samples, labels = ace_transform_samples(samples, labels, data, ground_truth, cpu=cpu_count)
###########
# Train the Network
###########
trainParams = args["training"] if "training" in args else {}
buildParams = args["building"] if "building" in args else{}
model = train_model(samples, labels, window_size, buildParams, trainParams)
def train(model,case_type,number=1):
average_accuracy=0
test_accuracy=list()
if model=='svm':
import svm
x_train,y_train,x_test,y_test=svm.loadText(case_type)
for _ in range(int(number)):
test_accuracy.append(svm.train(x_train,y_train,x_test,y_test))
average_accuracy=average_accuracy+test_accuracy[_]
average_accuracy=average_accuracy/int(number)
print("aveage accuract:" +str(average_accuracy))
elif model=='cnn':
import cnn
train_data, test_data, train_label, test_label, vocab = cnn.get_data(case_type,mode='sequence')
for _ in range(int(number)):
test_accuracy.append(cnn.train_model(case_type,train_data, test_data, train_label, test_label, vocab))
average_accuracy=average_accuracy+test_accuracy[_]
elif model=='lstm':
import lstm
train_data, test_data, train_label, test_label, vocab = lstm.get_data(case_type,mode='sequence')
for _ in range(int(number)):
test_accuracy.append(lstm.train_model(case_type,train_data, test_data, train_label, test_label, vocab))
average_accuracy=average_accuracy+test_accuracy[_]
elif model=='keras_text_cnn':
import keras_text_cnn as text_cnn
train_data, test_data, train_label, test_label, vocab = text_cnn.get_data(case_type,mode='sequence')
for _ in range(int(number)):
test_accuracy.append(text_cnn.train_model(case_type,train_data, test_data, train_label, test_label, vocab))
average_accuracy=average_accuracy+test_accuracy[_]
average_accuracy=average_accuracy/int(number)
print("aveage accuract:" +str(average_accuracy))
with open(file='D:/judgement_prediction/judgement_prediction/'+case_type+'/information.txt', mode="a",encoding='utf-8') as target_file:
target_file.write(case_type)
for i in range(int(number)):
target_file.write(str(test_accuracy[i])+' ')
target_file.write(',average:'+str(average_accuracy)+'\n')
#import video
import vis
import datetime
import shutil
import os
# The sketch dataset is licensed under a Creative Commons Attribution 4.0 International License.
# https://creativecommons.org/licenses/by/4.0/
# Copyright (C) 2012 Mathias Eitz, James Hays, and Marc Alexa. 2012. How Do Humans Sketch Objects? ACM Trans. Graph. (Proc. SIGGRAPH) 31, 4 (2012), 44:1--44:10.
# http://cybertron.cg.tu-berlin.de/eitz/projects/classifysketch/
# The data has been modified from original by splitting the images in train sets and test sets, and subsets
if __name__ == '__main__':
print("Hello and welcome to the CNN training visualizer.")
start_time = datetime.datetime.now()
output_dir = f"output/{start_time.strftime('%Y-%m-%d %H.%M.%S')} {train_dir.split('/')[-1]} {test_dir.split('/')[-1]}"
print(f"Started {start_time}\n")
if not os.path.exists(output_dir):
os.mkdir(output_dir)
shutil.copyfile("const.py", output_dir + "/parameters.py")
m = cnn.ConvNet()
cnn.train_model(m, output_dir)
# classes, train_loader, losses, accuracies = cnn.train_model(m, output_dir)
# vis.show_images(train_loader, classes)
# vis.plot_results(losses, accuracies)
# Load a model instead: m = torch.load(MODEL_STORE_PATH + 'conv_net_model.ckpt')
# cnn.test_model(m)
# video.make(output_dir)
# output dimensions
num_classes = 10
################################################################################
# callbacks for Save weights, Tensorboard
# creating a new directory for each run using timestamp
folder = os.path.join(os.getcwd(), datetime.now().strftime("%d-%m-%Y_%H-%M-%S"), str(ACTIV_FN))
history_file = folder + "\cnn_" + str(ACTIV_FN) + ".h5"
save_callback = ModelCheckpoint(filepath=history_file, verbose=1)
tb_callback = TensorBoard(log_dir=folder)
# Build, train, and test model
model = cnn.build_model(input_shape, activation_fn, LEARNING_RATE, DROP_PROB, NUM_NEURONS_IN_DENSE_1, num_classes)
train_accuracy, train_loss, valid_accuracy, valid_loss = cnn.train_model(model, train_images, train_labels, BATCH_SIZE,
NUM_EPOCHS, valid_images, valid_labels,
save_callback, tb_callback)
test_accuracy, test_loss, predictions = cnn.test_model(model, test_images, test_labels)
# save test set results to csv
predictions = np.round(predictions)
predictions = predictions.astype(int)
df = pd.DataFrame(predictions)
df.to_csv("mnist.csv", header=None, index=None)
################################################################################
# Visualization and Output
num_epochs_plot = range(1, len(train_accuracy) + 1)
# Loss curves
plt.figure(1)
# HYPERPARAMETERS AND DESIGN CHOICES
num_neurons = 128
batch_size = 64
ACTIV_FN = "relu"
activation_fn = cnn.get_activ_fn(ACTIV_FN)
num_epochs = 5
max_count = 50
for count in range(0, max_count):
learn_rate = 10**uniform(-2, -4)
drop_prob = 10**uniform(-2, 0)
# callbacks for Save weights, Tensorboard
# creating a new directory for each run using timestamp
folder = os.path.join(os.getcwd(),
datetime.now().strftime("%d-%m-%Y_%H-%M-%S"),
str(ACTIV_FN))
tb_callback = TensorBoard(log_dir=folder)
# Build, train, and test model
model = cnn.build_model(input_shape, activation_fn, learn_rate, drop_prob,
num_neurons, num_classes)
train_accuracy, train_loss, valid_accuracy, valid_loss = cnn.train_model(
model, train_images, train_labels, batch_size, num_epochs,
valid_images, valid_labels, tb_callback)
print(
'Step: {:d}/{:d}, learn: {:.6f}, dropout: {:.4f},'
'Train_loss: {:.4f}, Train_acc: {:.4f}, Val_loss: {:.4f}, Val_acc: {:.4f}'
.format(count, max_count, learn_rate, drop_prob, train_loss[-1],
train_accuracy[-1], valid_loss[-1], valid_accuracy[-1]))
# callbacks for Save weights, Tensorboard
# creating a new directory for each run using timestamp
folder = os.path.join(os.getcwd(),
datetime.now().strftime("%d-%m-%Y_%H-%M-%S"))
history_file = folder + "/cnn_nodatagen" + ".h5"
save_callback = ModelCheckpoint(filepath=history_file, verbose=1)
tb_callback = TensorBoard(log_dir=folder)
# Build, train, and test model
model = cnn.build_model(input_shape=input_shape,
learn_rate=LEARNING_RATE,
drop_prob=DROP_PROB,
num_neurons=NUM_NEURONS_IN_DENSE_1)
train_accuracy, train_loss, valid_accuracy, valid_loss = cnn.train_model(
model, x_train, y_train, BATCH_SIZE, NUM_EPOCHS, x_valid, y_valid,
save_callback, tb_callback)
test_accuracy, test_loss, predictions = cnn.test_model(model, x_test, y_test)
# save test set results to csv
predictions = np.round(predictions)
predictions = predictions.astype(int)
df = pd.DataFrame(predictions)
df.to_csv("predictions_nodatagen.csv", header=None, index=None)
# Visualization and Output
num_epochs_plot = range(1, len(train_accuracy) + 1) # x axis range
directory = os.getcwd()
# Loss curves
plt.figure()
plt.plot(num_epochs_plot, train_loss, "b", label="Training Loss")
plt.plot(num_epochs_plot, valid_loss, "r", label="Validation Loss")
x_train, y_train = utils.apply_undersampling(x_train,y_train,undersample_val=undersample_val)
# initialize output bias
neg, pos = np.bincount(y_train)
output_bias = np.log(pos/neg)
output_bias = keras.initializers.Constant(output_bias)
print("Positive Class Counter:",pos)
print("Negative Class Counter:",neg)
model = cnn.build_model(input_shape = input_shape, layers = n_layers, filters = n_filters, opt=opt, output_bias=output_bias)
weightsFile = 'numSplit'+str(numSplit)+'_params'+str(parameterNum)
history = cnn.train_model(model,x_train,y_train,x_val,y_val,
weightsDir,weightsFile,
patience_count=patience_count,
epochs=max_epochs,
batch_size=batch_size,
class_weights = class_weights)
model.load_weights(weightsDir+weightsFile+'.h5')
predictions = model.predict(x_val)
cm = utils.calc_cm(y_val,predictions)
precision, recall = utils.calc_binary_metrics(cm)
auc = roc_auc_score(y_val,predictions)
#save the metrics for the best epoch, or the last one
if(len(history.history['acc']) == max_epochs):
iterations += max_epochs
training_acc += history.history['acc'][max_epochs-1]
training_loss += history.history['loss'][max_epochs-1]
def main(train_spectrum_path=r"dataset/train_spectrum.npy",
test_spectrum_path=r"dataset/test_spectrum.npy",
train_labels_path=r"dataset/train_labels.npy",
test_labels_path=r"dataset/test_labels.npy",
batch_size=1,
learning_rate=0.01,
num_epochs=20,
kernel_size=(1, 2),
padding=(0, 0),
dropout=True,
drop_prob=0.2,
batch_normalization=True,
weight_decay=True,
weight_decay_amount=0.01,
data_width=2100,
model_save_path=r"model.pth",
fruits=("apple", "banana", "mix"),
create_dataset_now=False,
root_dir="YOMIRAN",
num_channels_layer1=3,
num_channels_layer2=6,
sample_time="after 5",
sample_location="anal",
sample_type="pos",
tolerance=5,
number_of_samples_to_alter=100,
size_of_dataset=60000,
train_data_percentage=0.8,
train_now=False,
show_statistics=True,
predict_now=False,
file_to_predict=r"apple neg.txt",
confidence_threshold=0.7,
validate_hierarchy=True,
validate_filename_format=True,
validate_empty_file=True,
create_dataset_progress_bar_intvar=None,
fc1_amount_output_nodes=1000,
fc2_amount_output_nodes=500,
fc3_amount_output_node=100,
stretch_data=False,
knn=False,
cross_validation_iterations=1,
n_components=2,
k="auto"):
# create data set
if create_dataset_now:
valid_files, _ = get_valid_and_invalid_files(
root_dir=root_dir,
validate_empty_file=validate_empty_file,
validate_filename_format=validate_filename_format,
validate_hierarchy=validate_hierarchy)
create_dataset(data_files=valid_files,
fruits=fruits,
size_of_dataset=size_of_dataset,
train_data_percentage=train_data_percentage,
tolerance=tolerance,
number_of_samples_to_alter=number_of_samples_to_alter,
train_spectrum_path=Path(train_spectrum_path),
train_labels_path=Path(train_labels_path),
test_spectrum_path=Path(test_spectrum_path),
test_labels_path=Path(test_labels_path),
data_width=data_width,
sample_time=sample_time,
sample_location=sample_location,
create_dataset_progress_bar_intvar=
create_dataset_progress_bar_intvar,
stretch_data=stretch_data,
sample_type=sample_type,
n_components=n_components)
# transformation of dataset
transform = compose(transforms.ToTensor(), minmax_scale)
# get the labels enum
if fruits:
fruit_label_enum = create_fruit_labels(fruits=fruits)
# transform = transforms.ToTensor()
if train_now:
# Get the dataset
train_data_loader = DataLoader("train",
train_spectrum_path=train_spectrum_path,
train_labels_path=train_labels_path,
batch_size=batch_size,
transform=transform)
test_data_loader = DataLoader("test",
test_spectrum_path=test_spectrum_path,
test_labels_path=test_labels_path,
batch_size=batch_size,
transform=transform)
if knn:
train_data_loader_size_calculator = copy.deepcopy(
train_data_loader)
amount_train_data = 0
for spectrum, labels in train_data_loader_size_calculator.load_data(
):
amount_train_data += spectrum.shape[0]
if k == "auto":
k = math.ceil(math.sqrt(amount_train_data))
cross_validation_accuracies = []
cross_validation_true_labels = []
cross_validation_predictions = []
for i in range(cross_validation_iterations):
print("Cross validation iteration: {}/{}".format(
i + 1, cross_validation_iterations))
# Get the dataset
train_data_loader = DataLoader(
"train",
train_spectrum_path=train_spectrum_path,
train_labels_path=train_labels_path,
batch_size=1,
transform=transform)
test_data_loader = DataLoader(
"test",
test_spectrum_path=test_spectrum_path,
test_labels_path=test_labels_path,
batch_size=1,
transform=transform)
model = KNN(k=k,
train_data_loader=train_data_loader,
test_data_loader=test_data_loader)
accuracy, true_labels, predictions = model.train()
cross_validation_accuracies.append(accuracy * 100)
cross_validation_true_labels.extend(true_labels)
cross_validation_predictions.extend(predictions)
print("k={}\tAccuracy: {:.3f}%".format(k, accuracy * 100))
shuffle_data_for_cross_validation(
train_data_loader=train_data_loader,
test_data_loader=test_data_loader,
train_spectrum_path=train_spectrum_path,
train_labels_path=train_labels_path,
test_spectrum_path=test_spectrum_path,
test_labels_path=test_labels_path)
accuracies_mean = stat.mean(cross_validation_accuracies)
accuracies_std = stat.stdev(cross_validation_accuracies)
print("Test accuracies mean: {}".format(accuracies_mean))
print("Test accuracies standard deviation: {}".format(
accuracies_std))
plot_data.plot_box_plot(
test_accuracies=cross_validation_accuracies)
plot_data.plot_confusion_matrix(
true_labels=cross_validation_true_labels,
predictions=cross_validation_predictions,
fruits=fruits,
show_null_values=True,
show_plot=True)
# for k in range(1, amount_train_data):
# model = KNN(k=k, train_data_loader=train_data_loader, test_data_loader=test_data_loader)
# accuracy = model.train()
# print("k={}\tAccuracy: {:.3f}%".format(k, accuracy * 100))
else:
train_data_loader_size_calculator = copy.deepcopy(
train_data_loader)
amount_train_data = 0
fruits_from_dataset = []
for spectrum, labels in train_data_loader_size_calculator.load_data(
):
amount_train_data += spectrum.shape[0]
# deprecated
if fruits is None:
for label in labels:
if label not in fruits_from_dataset:
fruits_from_dataset.append(label)
# deprecated
if fruits is None:
fruit_label_enum = create_fruit_labels(
fruits=fruits_from_dataset)
fruits = fruits_from_dataset
cross_validation_losses = []
cross_validation_accuracies_train = []
cross_validation_accuracies_test = []
cross_validation_true_labels = []
cross_validation_predictions_of_last_epoch = []
statistics_of_all_iterations = []
for i in range(cross_validation_iterations):
print("Cross validation iteration: {}/{}".format(
i + 1, cross_validation_iterations))
# Get the dataset
train_data_loader = DataLoader(
"train",
train_spectrum_path=train_spectrum_path,
train_labels_path=train_labels_path,
batch_size=batch_size,
transform=transform)
test_data_loader = DataLoader(
"test",
test_spectrum_path=test_spectrum_path,
test_labels_path=test_labels_path,
batch_size=batch_size,
transform=transform)
# initialize the neural net
model = CNN(amount_of_labels=len(fruit_label_enum),
batch_normalization=batch_normalization,
dropout=dropout,
drop_prob=drop_prob,
kernel_size=kernel_size,
padding=padding,
data_width=data_width,
data_height=1,
num_channels_layer1=num_channels_layer1,
num_channels_layer2=num_channels_layer2,
fc1_amount_output_nodes=fc1_amount_output_nodes,
fc2_amount_output_nodes=fc2_amount_output_nodes,
fc3_amount_output_node=fc3_amount_output_node,
n_components=n_components)
# train the model
statistics = train_model(
model=model,
fruit_label_enum=fruit_label_enum,
train_data_loader=train_data_loader,
test_data_loader=test_data_loader,
num_epochs=num_epochs,
learning_rate=learning_rate,
batch_size=batch_size,
weight_decay=weight_decay,
weight_decay_amount=weight_decay_amount,
model_save_path=model_save_path,
train_dataset_size=amount_train_data)
statistics_of_all_iterations.append(statistics)
losses, accuracies_train, accuracies_test, true_labels, predictions_of_last_epoch = statistics
cross_validation_losses.extend([losses[-1]])
cross_validation_accuracies_train.extend(
[accuracies_train[-1]])
cross_validation_accuracies_test.extend([accuracies_test[-1]])
cross_validation_true_labels.extend(true_labels),
cross_validation_predictions_of_last_epoch.extend(
list(predictions_of_last_epoch))
shuffle_data_for_cross_validation(
train_data_loader=train_data_loader,
test_data_loader=test_data_loader,
train_spectrum_path=train_spectrum_path,
train_labels_path=train_labels_path,
test_spectrum_path=test_spectrum_path,
test_labels_path=test_labels_path)
accuracies_test_mean = stat.mean(cross_validation_accuracies_test)
accuracies_test_std = stat.stdev(cross_validation_accuracies_test)
print("Test accuracies mean: {}".format(accuracies_test_mean))
print("Test accuracies standard deviation: {}".format(
accuracies_test_std))
# plot_data.plot_box_plot(test_accuracies=cross_validation_accuracies_test, show_plot=True)
# plot the statistics
if show_statistics:
# plot_data.plot_train_statistics(x_values=range(len(losses)), y_values=losses, x_label="Epoch",
# y_label="Loss")
# plot_data.plot_train_statistics(x_values=range(len(accuracies_train)), y_values=accuracies_train,
# x_label="Epoch", y_label="Train accuracy")
# plot_data.plot_train_statistics(x_values=range(len(accuracies_test)), y_values=accuracies_test,
# x_label="Epoch", y_label="Test accuracy")
# max_test_accuracy = max(cross_validation_accuracies_test)
# statistics = statistics_of_all_iterations[cross_validation_accuracies_test.index(max_test_accuracy)]
# losses, accuracies_train, accuracies_test, true_labels, predictions_of_last_epoch = statistics
# plot_data.plot_train_statistics1(losses=losses, train_accuracy=accuracies_train,
# test_accuracy=accuracies_test)
plot_data.plot_box_plot(
test_accuracies=cross_validation_accuracies_test)
plot_data.plot_confusion_matrix(
true_labels=cross_validation_true_labels,
predictions=cross_validation_predictions_of_last_epoch,
fruits=fruits,
show_null_values=True)
plot_data.plot_classification_report(
true_labels=cross_validation_true_labels,
predictions=cross_validation_predictions_of_last_epoch,
show_plot=True)
max_test_accuracy = max(cross_validation_accuracies_test)
statistics = statistics_of_all_iterations[
cross_validation_accuracies_test.index(max_test_accuracy)]
losses, accuracies_train, accuracies_test, true_labels, predictions_of_last_epoch = statistics
plot_data.plot_train_statistics1(
losses=losses,
train_accuracy=accuracies_train,
test_accuracy=accuracies_test,
show_plot=True)
if predict_now:
# fit the pca
valid_files, _ = get_valid_and_invalid_files(
root_dir=root_dir,
validate_empty_file=validate_empty_file,
validate_filename_format=validate_filename_format,
validate_hierarchy=validate_hierarchy)
# Get existing data
existing_data, _ = get_existing_data(fruits=fruits,
data_files=valid_files,
sample_time=sample_time,
sample_location=sample_location,
data_width=data_width,
sample_type=sample_type)
# fit pca
pca = PCA(n_components=n_components)
pca = pca.fit(existing_data)
model = load_model(model_save_path,
amount_of_labels=len(fruit_label_enum),
batch_normalization=batch_normalization,
dropout=dropout,
drop_prob=drop_prob,
kernel_size=kernel_size,
padding=padding,
data_width=data_width,
data_height=1,
num_channels_layer1=num_channels_layer1,
num_channels_layer2=num_channels_layer2,
fc1_amount_output_nodes=fc1_amount_output_nodes,
fc2_amount_output_nodes=fc2_amount_output_nodes,
fc3_amount_output_node=fc3_amount_output_node,
n_components=n_components)
confidence, prediction = predict(
model=model,
data_file=file_to_predict,
pca=pca,
transform=transform,
fruit_label_enum=fruit_label_enum,
data_width=data_width,
confidence_threshold=confidence_threshold)
return confidence, prediction