# Load dataset
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
X_train = pre_process_images(X_train)
X_val = pre_process_images(X_val)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
current_train_history, current_val_history = trainer.train(num_epochs)
train_history[model_name] = current_train_history
val_history[model_name] = current_val_history
plt.figure(figsize=(16, 10))
plt.subplot(1, 2, 1)
plt.ylim([0, .5])
for model_name in train_history.keys():
utils.plot_loss(train_history[model_name]["loss"],
model_name,
# Load dataset
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
X_train = pre_process_images(X_train)
X_val = pre_process_images(X_val)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
# 32 neurons per layer
model_a = SoftmaxModel(
neurons_per_layer,
use_improved_sigmoid,
use_improved_weight_init)
trainer_a = SoftmaxTrainer(
momentum_gamma, use_momentum,
model_a, learning_rate, batch_size, shuffle_data,
X_train, Y_train, X_val, Y_val,
)
train_history_a, val_history_a = trainer_a.train(num_epochs)
# 128 neurons per layer
neurons_per_layer = [128, 10]
model_b = SoftmaxModel(
neurons_per_layer,
use_improved_sigmoid,
use_improved_weight_init)
trainer_b = SoftmaxTrainer(
momentum_gamma, use_momentum,
model_b, learning_rate, batch_size, shuffle_data,
X_train, Y_train, X_val, Y_val,
)
# Load dataset
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
X_train = pre_process_images(X_train)
X_val = pre_process_images(X_val)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history, val_history = trainer.train(num_epochs)
# Example created in assignment text - Comparing with and without shuffling.
# FIRST CASE (weights)
use_improved_weight_init = True
model_weights = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer_weights = SoftmaxTrainer(
plt.xlabel("Training steps")
plt.legend(loc="lower right")
# First nothing
use_improved_weight_init = False
use_improved_sigmoid = False
use_momentum = False
model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history_nothing, val_history_nothing = trainer.train(num_epochs)
# Adding improved weights
use_improved_weight_init = True
use_improved_sigmoid = False
use_momentum = False
model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
# Load dataset
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
X_train = pre_process_images(X_train)
X_val = pre_process_images(X_val)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history, val_history = trainer.train(num_epochs)
# Example created in assignment text - Comparing with and without shuffling.
# YOU CAN DELETE EVERYTHING BELOW!
use_improved_weight_init = True
"""
model_improved_weights = SoftmaxModel(
neurons_per_layer,
use_improved_sigmoid,
use_improved_weight_init)
# Load dataset
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
X_train = pre_process_images(X_train)
X_val = pre_process_images(X_val)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
######Naked model - no improvments######
model_naked = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer_naked = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model_naked,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history_naked, val_history_naked = trainer_naked.train(num_epochs)
print("just basic")
print("Train accuracy:", calculate_accuracy(X_train, Y_train, model_naked))
print("Validation accuracy:", calculate_accuracy(X_val, Y_val,
model_naked))
print("Final Validation Cross Entropy Loss:",
cross_entropy_loss(Y_val, model_naked.forward(X_val)))
print("Final Train Cross Entropy Loss:",
cross_entropy_loss(Y_train, model_naked.forward(X_train)))
mean = np.mean(X_train)
std = np.std(X_train)
X_train = pre_process_images(X_train, mean, std)
X_val = pre_process_images(X_val, mean, std)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
model = SoftmaxModel(neurons_per_layer_small, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model,
learning_rate,
batch_size,
shuffle_data,
early_stopping,
X_train,
Y_train,
X_val,
Y_val,
)
train_history, val_history = trainer.train(num_epochs)
# Example created in assignment text - Comparing with and without shuffling.
# YOU CAN DELETE EVERYTHING BELOW!
shuffle_data = False
model_large = SoftmaxModel(neurons_per_layer_large, use_improved_sigmoid,
use_improved_weight_init)
trainer_shuffle = SoftmaxTrainer(
momentum_gamma,
# Load dataset
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
X_train = pre_process_images(X_train)
X_val = pre_process_images(X_val)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
print("Training standard model:\n")
model = SoftmaxModel(
neurons_per_layer,
use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma, use_momentum,
model, learning_rate, batch_size, shuffle_data,
X_train, Y_train, X_val, Y_val,
)
train_history, val_history = trainer.train(num_epochs)
print("Final Train Cross Entropy Loss:",
cross_entropy_loss(Y_train, model.forward(X_train)))
print("Final Validation Cross Entropy Loss:",
cross_entropy_loss(Y_val, model.forward(X_val)))
print("Train accuracy:", calculate_accuracy(X_train, Y_train, model))
print("Validation accuracy:", calculate_accuracy(X_val, Y_val, model))
print("\n\n")
# Example created in assignment text - Comparing with and without shuffling.
# YOU CAN DELETE EVERYTHING BELOW!
# Load dataset
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
X_train = pre_process_images(X_train)
X_val = pre_process_images(X_val)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
model_previous = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model_previous,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history, val_history = trainer.train(num_epochs)
# Example created in assignment text - Comparing with and without shuffling.
# YOU CAN DELETE EVERYTHING BELOW!
#Here you edit the new one!
use_improved_weight_init = True
use_improved_sigmoid = True
use_momentum = True
learning_rate = .02
neurons_per_layer = [64, 64, 10]
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
X_train = pre_process_images(X_train)
X_val = pre_process_images(X_val)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
# Task 4 point a and b
# Here is presented the model retrieved from the previous task3
model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
print(
f'Training the Original model {neurons_per_layer[0]} neurons in hidden layer'
)
train_history, val_history = trainer.train(num_epochs)
# Task 4
# TASK 4 - a) 32 neurons for hidden layer
neurons_per_layer = [32, 10]
model_32neu = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
X_train = pre_process_images(X_train)
X_val = pre_process_images(X_val)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
neurons_per_layer = [64, 10]
model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history_64, val_history_64 = trainer.train(num_epochs)
print("\n\n --------------------------")
print(neurons_per_layer)
print("Final Train Cross Entropy Loss:",
cross_entropy_loss(Y_train, model.forward(X_train)))
print("Final Validation Cross Entropy Loss:",
cross_entropy_loss(Y_val, model.forward(X_val)))
print("Train accuracy:", calculate_accuracy(X_train, Y_train, model))
print("Validation accuracy:", calculate_accuracy(X_val, Y_val, model))
print("\n\n --------------------------")
mean = np.mean(X_train)
std = np.std(X_train)
X_train = pre_process_images(X_train, mean, std)
X_val = pre_process_images(X_val, mean, std)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
# 64 hidden units
model64 = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer64 = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model64,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history64, val_history64 = trainer64.train(num_epochs)
# 32 hidden units
neurons_per_layer = [32, 10]
model32 = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer32 = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model32,
# Task 4e) #
###################
shuffle_data = True
use_improved_sigmoid = True
use_improved_weight_init = True
use_momentum = False
learning_rate_momentum = 0.02
neurons_per_layer = [64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 10]
model_64_layers = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer_shuffle = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model_64_layers,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history_64_layers, val_history_64_layers = trainer_shuffle.train(
num_epochs)
plt.figure(figsize=(20, 12))
plt.subplot(1, 2, 1)
plt.ylim([0.00, 1.00])
utils.plot_loss(train_history_64_layers["loss"],
"Training Loss 10 hidden layers",
npoints_to_average=10)
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
mean = np.mean(X_train)
std = np.std(X_train)
X_train = pre_process_images(X_train, mean, std)
X_val = pre_process_images(X_val, mean, std)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history, val_history = trainer.train(num_epochs)
# Example created in assignment text - Comparing with and without shuffling.
# YOU CAN DELETE EVERYTHING BELOW!
use_improved_weight_init = True
use_improved_sigmoid = True
use_momentum = True
learning_rate = 0.02
new_model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
plt.ylim([0.85, 1.01])
plt.ylabel("Accuracy")
plt.xlabel("Training steps")
plt.legend(loc="lower right")
# Single hidden layer
neurons_per_layer = [64, 10]
model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history_single, val_history_single = trainer.train(num_epochs)
print("---------", neurons_per_layer, "----------")
print("Final Train Cross Entropy Loss:",
cross_entropy_loss(Y_train, model.forward(X_train)))
print("Final Validation Cross Entropy Loss:",
cross_entropy_loss(Y_val, model.forward(X_val)))
print("Train accuracy:", calculate_accuracy(X_train, Y_train, model))
print("Validation accuracy:", calculate_accuracy(X_val, Y_val, model))
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
mean = X_train.mean()
sd = X_train.std()
X_train = pre_process_images(X_train, mean, sd)
X_val = pre_process_images(X_val, mean, sd)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history, val_history = trainer.train(num_epochs)
#two hidden layers
neurons_per_layer = [59, 59, 10]
model_two_hidden_layers = SoftmaxModel(neurons_per_layer,
use_improved_sigmoid,
use_improved_weight_init)
trainer_two_hidden_layers = SoftmaxTrainer(
momentum_gamma,
use_momentum,
X_val = pre_process_images(X_val)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
######1nd model - network from task 3######
use_improved_sigmoid = True
use_improved_weight_init = True
use_momentum = True
model = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model,
learning_rate,
batch_size,
shuffle_data,
X_train,
Y_train,
X_val,
Y_val,
)
train_history, val_history = trainer.train(num_epochs)
######2nd model - network from task 4d ######
neurons_per_layer = [60, 60, 10]
model1 = SoftmaxModel(neurons_per_layer, use_improved_sigmoid,
use_improved_weight_init)
trainer1 = SoftmaxTrainer(
momentum_gamma,
use_momentum,
model1,
X_train, Y_train, X_val, Y_val = utils.load_full_mnist()
mean = np.mean(X_train)
std = np.std(X_train)
X_train = pre_process_images(X_train, mean, std)
X_val = pre_process_images(X_val, mean, std)
Y_train = one_hot_encode(Y_train, 10)
Y_val = one_hot_encode(Y_val, 10)
model = SoftmaxModel(
neurons_per_layer,
use_improved_sigmoid,
use_improved_weight_init)
trainer = SoftmaxTrainer(
momentum_gamma, use_momentum,
model, learning_rate, batch_size, shuffle_data, early_stop,
X_train, Y_train, X_val, Y_val,
)
train_history, val_history = trainer.train(num_epochs)
## RUN WITH MOMENTUM
use_momentum = True
model_momentum = SoftmaxModel(
neurons_per_layer,
use_improved_sigmoid,
use_improved_weight_init)
trainer_momentum = SoftmaxTrainer(
momentum_gamma, use_momentum,
model_momentum, learning_rate, batch_size, shuffle_data, early_stop,
X_train, Y_train, X_val, Y_val,
)