def load_data(batch_size):
train_dataset = AdultDataset(training_data, training_labels)
valid_dataset = AdultDataset(validation_data, validation_labels)
train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
val_loader = DataLoader(valid_dataset, batch_size=batch_size, shuffle=False)
return train_loader, val_loader
def load_data(batch_size):
trainDataSet = AdultDataset(trainingData, trainingLabel)
validationDataSet = AdultDataset(validationData, validationLabel)
train_loader = DataLoader(trainDataSet,
batch_size=batch_size,
shuffle=True)
val_loader = DataLoader(validationDataSet,
batch_size=batch_size,
shuffle=True)
return train_loader, val_loader
def load_data(batch_size):
######
# 3.2 YOUR CODE HERE
train_dataset = AdultDataset(feat_train,label_train)
val_dataset = AdultDataset(feat_valid,label_valid)
######
train_loader = DataLoader(train_dataset, batch_size = batch_size, shuffle= True)
val_loader = DataLoader(val_dataset, batch_size = batch_size, shuffle = False)
return train_loader, val_loader
def load_data(batch_size, train_data, val_data, train_labels, val_labels):
###### DO I GIVE THE VALIDATION SET A BATCH SIZE?
# 3.2 YOUR CODE HERE
train_data = AdultDataset(train_data, train_labels)
val_data = AdultDataset(val_data, val_labels)
train_loader = DataLoader(train_data, batch_size=batch_size, shuffle=True)
val_loader = DataLoader(val_data, batch_size=batch_size, shuffle=False)
######
return train_loader, val_loader
def load_data(batch_size=5):
######
train_loader = DataLoader(AdultDataset(X_train, Y_train),
batch_size=batch_size,
shuffle=True)
val_loader = DataLoader(AdultDataset(X_test, Y_test),
batch_size=1,
shuffle=True)
######
return train_loader, val_loader
def load_data(batch_size):
training_set = AdultDataset(X_train, y_train)
train_loader = DataLoader(training_set,
batch_size=batch_size,
shuffle=True)
validation_set = AdultDataset(X_test, y_test)
val_loader = DataLoader(validation_set,
batch_size=batch_size,
shuffle=True)
return train_loader, val_loader
def load_data(batchsize):
######
# 3.2 YOUR CODE HERE
training_data = AdultDataset(training_features, training_income)
validation_data = AdultDataset(validation_features, validation_income)
train_loader = DataLoader(training_data, batch_size=batchsize, shuffle=True)
val_loader = DataLoader(validation_data, batch_size=batchsize)
######
return train_loader, val_loader
def load_data(batch_size):
######
# 4.2 YOUR CODE HERE
traindata = AdultDataset(feat_train, label_train)
validdata = AdultDataset(feat_valid, label_valid)
torch.manual_seed(seed)
train_loader = DataLoader(traindata, batch_size=batch_size, shuffle=True)
val_loader = DataLoader(validdata, batch_size=batch_size, shuffle=False)
######
return train_loader, val_loader
def load_data(batch_size):
######
# 4.1 YOUR CODE HERE
trainDataset = AdultDataset(trainData, trainLabel)
valDataset = AdultDataset(valData, valLabel)
train_loader = DataLoader(dataset=trainDataset,
batch_size=batch_size,
shuffle=True)
val_loader = DataLoader(dataset=valDataset,
batch_size=batch_size,
shuffle=True)
######
return train_loader, val_loader
def load_data(batch_size, lr):
train_dataset = AdultDataset(X_train, y_train)
train_loader = DataLoader(train_dataset,
batch_size=batch_size,
num_workers=1,
shuffle=True)
test_dataset = AdultDataset(X_test, y_test)
test_loader = DataLoader(test_dataset,
batch_size=batch_size,
num_workers=1,
shuffle=False)
loss_fnc = torch.nn.BCELoss()
model = MultiLayerPerceptron(X_train.shape[1])
optimizer = torch.optim.SGD(filter(lambda p: p.requires_grad,
model.parameters()),
lr=lr)
return train_loader, test_loader, model, loss_fnc, optimizer
torch.set_num_threads(8)
logger = get_logger(args.name)
# Set random number seed.
np.random.seed(args.seed)
torch.manual_seed(args.seed)
dtype = np.float32
logger.info(
"UCI Adult data set, target attribute: {}, sensitive attribute: {}".format(
args.target, args.private))
# Load UCI Adult dataset.
time_start = time.time()
adult_train = AdultDataset(root_dir='data',
phase='train',
tar_attr=args.target,
priv_attr=args.private)
adult_test = AdultDataset(root_dir='data',
phase='test',
tar_attr=args.target,
priv_attr=args.private)
train_loader = DataLoader(adult_train,
batch_size=args.batch_size,
shuffle=True)
test_loader = DataLoader(adult_test, batch_size=args.batch_size, shuffle=False)
time_end = time.time()
logger.info(
"Time used to load all the data sets: {} seconds.".format(time_end -
time_start))
input_dim = adult_train.xdim
num_classes = 2
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--batch_size', type=int)
parser.add_argument('--lr', type=float)
parser.add_argument('--epochs', type=int, default=20)
parser.add_argument('--eval_every', type=int, default=10)
args = parser.parse_args()
######
train_loader, val_loader = load_data(args.batch_size)
model, loss_fnc, optimizer = load_model(args.lr)
t_loss, v_loss, t_acc, v_acc = [], [], [], []
for epoch in range(args.epochs):
n = args.eval_every
printI = 1
lastN = []
for datum, label in train_loader:
# set up optimizer with blank gradients
optimizer.zero_grad()
# predict with the current weights
predict = model(datum.float())
# evaluate the loss for the predictions
loss = loss_fnc(input=predict.squeeze(), target=label.float())
# calculate the gradients per the loss function we chose
loss.backward()
# changes the weights one step in the right direction (for the batch)
optimizer.step()
if len(lastN) < n:
lastN += [[datum, label]]
else:
lastN = lastN[1:] + [[datum, label]]
if printI == n:
t_loss += [loss.item()]
predict = model(X_test.float())
v_loss += [
loss_fnc(input=predict.squeeze(),
target=Y_test.float()).item()
]
t_acc_this = 0
for every in range(n):
t_acc_this += evaluate(
model,
DataLoader(
AdultDataset(lastN[every][0], lastN[every][1])))
t_acc += [t_acc_this / n]
v_acc += [
evaluate(model, DataLoader(AdultDataset(X_test, Y_test)))
]
printI = 1
print("Status update: currently on epoch", epoch,
"and Loss is", loss)
print("Number of correct predictions is ",
t_acc[-1] * len(Y_train))
print("Validation accuracy is currently: ", v_acc[-1])
printI += 1
if plotting:
# Plot losses
plt.figure()
x = [i for i in range(len(t_loss))]
plt.plot(x, t_loss, label='Training Loss')
plt.plot(x, v_loss, label='Validation Loss')
plt.title("Losses as Function of Gradient Step")
plt.ylabel("Loss")
plt.xlabel("Gradient Steps")
plt.legend()
plt.show()
# Plot accuracies
x = [i for i in range(len(t_acc))]
plt.figure()
plt.plot(x, t_acc, label='Training Accuracy')
plt.plot(x, v_acc, label='Validation Accuracy')
plt.ylim(0, 1)
plt.title("Accuracy as Function of Gradient Step")
plt.ylabel("Accuracy")
plt.xlabel("Gradient Steps")
plt.legend()
plt.show()
# Plot losses smoothed
plt.figure()
x = [i for i in range(len(t_loss))]
plt.plot(x, smooth(t_loss, 151, 3), label='Training Loss Smoothed')
plt.plot(x, smooth(v_loss, 101, 4), label='Validation Loss')
plt.title("Smoothed Losses as Function of Gradient Step")
plt.ylabel("Loss")
plt.xlabel("Gradient Steps")
plt.legend()
plt.show()
# Plot accuracies smoothed
plt.figure()
plt.plot(x, smooth(t_acc, 151, 5), label='Training Accuracy Smoothed')
plt.plot(x,
smooth(v_acc, 101, 4),
label='Validation Accuracy Smoothed')
plt.ylim(0, 1)
plt.title("Smoothed Accuracy as Function of Gradient Step")
plt.ylabel("Accuracy")
plt.xlabel("Gradient Steps")
plt.legend()
plt.show()
