def main():
# Get command-line arguments
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
'--verbose',
type=int,
default=1,
help='Specify level of verbosity: 0=none, 1=default, 2=extra-verbose')
parser.add_argument('--nepochs',
type=int,
default=2,
help='Specify the number of training epochs')
args = parser.parse_args()
# Check for output directory to store plots
if not os.path.isdir(OUTPUT_DIR):
os.system('mkdir ' + OUTPUT_DIR)
# Set up logging
if args.verbose:
tee = subprocess.Popen(["tee", LOGFILE], stdin=subprocess.PIPE)
os.dup2(tee.stdin.fileno(), sys.stdout.fileno())
os.dup2(tee.stdin.fileno(), sys.stderr.fileno())
# Load train/test dataset
train_dataset, test_dataset, label_dict = dt.load_CIFAR10_data(
verbose=args.verbose)
# Plot one example image from each category
dt.display_example_data(train_dataset,
label_dict,
filename=OUTPUT_DIR +
'training_example_images.png')
# Construct the model
model = CNN(len(label_dict))
# Check to see if GPU is available and move model to GPU if it is
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
if args.verbose:
print('GPU is available?: {}'.format(torch.cuda.is_available()))
print('Using device: {}'.format(device))
model.to(device)
# Define our loss function and optimizer
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters())
# Use batch gradient descent; we don't want to load the whole dataset into memory all at once!
batch_size = 128
train_loader, test_loader = dt.get_dataloaders(train_dataset,
test_dataset,
batch_size=batch_size)
# Execute the training loop
train_losses, test_losses = ml.train(train_loader,
model,
optimizer,
criterion,
device=device,
n_epochs=args.nepochs,
test_loader=test_loader,
verbose=bool(args.verbose),
print_every=1)
# Plot loss per epoch to output directory
ml.plot_losses(train_losses,
test_losses,
filename=OUTPUT_DIR + 'model_losses.png')
# Get model predictions
train_predictions, train_targets = ml.predict(train_loader,
model,
device=device,
multiclass=True)
test_predictions, test_targets = ml.predict(test_loader,
model,
device=device,
multiclass=True)
# Get overall accuracy
ml.get_accuracy(train_predictions,
train_targets,
test_predictions,
test_targets,
verbose=args.verbose)
# Plot examples of images we got wrong
correct, wrong = ml.get_correct_wrong(test_predictions,
test_targets,
verbose=args.verbose)
dt.display_pred_examples(wrong,
test_dataset,
test_targets,
test_predictions,
label_dict,
filename=OUTPUT_DIR + 'wrong_examples.png')
# Make the confusion_matrix
ml.plot_confusion_matrix(test_targets,
test_predictions,
labels=label_dict,
log_color=True,
filename=OUTPUT_DIR + 'confusion_matrix.png')
# Plot model summary info
model.summarize()
# Flush logging
if args.verbose:
print("\nstdout flushed", flush=True)
print("stderr flushed", file=sys.stderr, flush=True)
def make_prediction(data_train, labels, data_test, indexes):
predictions = predict(data_train, labels, data_test, PARAMS, ITERATIONS)
dump_submission(predictions, indexes, OUTPUT)
def make_prediction(data_train, labels, data_test, indexes):
predictions = predict(data_train, labels, data_test, PARAMS, ITERATIONS)
dump_submission(predictions, indexes, OUTPUT)
def main():
# Get command-line arguments
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
'--verbose',
type=int,
default=1,
help='Specify level of verbosity: 0=none, 1=default, 2=extra-verbose')
parser.add_argument('--nepochs',
type=int,
default=2,
help='Specify the number of training epochs')
args = parser.parse_args()
# Check for output directory to store plots
if not os.path.isdir(OUTPUT_DIR):
os.system('mkdir ' + OUTPUT_DIR)
# Load train/test dataset
train_dataset, test_dataset, label_dict = mdata.load_FashionMNIST_data(
verbose=args.verbose)
# Plot one example image from each category
mdata.display_example_data(train_dataset,
label_dict,
filename=OUTPUT_DIR + filename_root +
'training_example_images.png')
# Construct the model
model = LSTM(n_inputs=28, n_hidden=128, n_outputs=10, n_rnnlayers=2)
# Check to see if GPU is available and move model to GPU if it is
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
if args.verbose:
print('GPU is available?: {}'.format(torch.cuda.is_available()))
print('Using device: {}'.format(device))
model.to(device)
# Define our loss function and optimizer
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters())
# Use batch gradient descent; we don't want to load the whole dataset into memory all at once!
batch_size = 256
train_loader, test_loader = mdata.get_dataloaders(train_dataset,
test_dataset,
batch_size=batch_size)
# Execute the training loop
train_losses, test_losses = ml.train(train_loader,
model,
optimizer,
criterion,
device=device,
n_epochs=args.nepochs,
test_loader=test_loader,
reshape_inp=reshape_func,
verbose=bool(args.verbose),
print_every=1)
# Plot loss per epoch to output directory
ml.plot_losses(train_losses,
test_losses,
filename=OUTPUT_DIR + filename_root + 'model_losses.png')
# Get model predictions
train_predictions, train_targets = ml.predict(train_loader,
model,
device=device,
reshape_inp=reshape_func,
multiclass=True)
test_predictions, test_targets = ml.predict(test_loader,
model,
device=device,
reshape_inp=reshape_func,
multiclass=True)
# Get overall accuracy
ml.get_accuracy(train_predictions,
train_targets,
test_predictions,
test_targets,
verbose=args.verbose)
# Plot examples of images we got wrong
correct, wrong = ml.get_correct_wrong(test_predictions,
test_targets,
verbose=args.verbose)
mdata.display_pred_examples(wrong,
test_dataset,
test_targets,
test_predictions,
label_dict,
filename=OUTPUT_DIR + filename_root +
'wrong_examples.png')
# Make the confusion_matrix
ml.plot_confusion_matrix(test_targets,
test_predictions,
labels=label_dict,
log_color=True,
filename=OUTPUT_DIR + filename_root +
'confusion_matrix.png')