I am training a CNN on face images and I want it to perform classification and regression tasks at the same time. I figured out how to train the CNN as below:
resnet = tf.keras.applications.ResNet50(
include_top=False ,
weights='imagenet' ,
input_shape=(96, 96, 3) ,
pooling="avg"
)
for layer in resnet.layers:
layer.trainable = True
inputs = Input(shape=(96, 96, 3), name='main_input')
main_branch = resnet(inputs)
main_branch = Flatten()(main_branch)
expr_branch = Dense(8, activation='softmax', name='expr_output')(main_branch)
va_branch = Dense(2, name='va_output')(main_branch)
model = Model(inputs = inputs,
outputs = [expr_branch, va_branch])
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.0001),
loss={'expr_output': 'sparse_categorical_crossentropy', 'va_output':
'mean_squared_error'})
I want after every epoch to prin the accuracy and the mse metrics for each task. So far I have written the below:
checkpoint = tf.keras.callbacks.ModelCheckpoint(
model_path,
save_weights_only=True,
verbose=1
)
history = model.fit_generator(
train_generator,
epochs=2,
steps_per_epoch=STEP_SIZE_TRAIN_resnet,
validation_data=test_generator,
validation_steps=STEP_SIZE_TEST_resnet,
max_queue_size=1,
shuffle=True,
callbacks=[checkpoint],
verbose=1
)
When I had only the classification task I would write
checkpoint = tf.keras.callbacks.ModelCheckpoint(
model_path,
monitor='val_accuracy',
save_best_only=True,
mode='max',
verbose=1
)
which printed the val_accuracy at every epoch and saved the weights. How can I do the same (print mse and accuracy and save the weights after every epoch) at a multitask problem?
Related
I am making a class-incremental learning multi-label classifier. Here the model first trains with 7 labels. After training, another dataset emerges that contains the same labels except one more. I want to automatically add an extra node to the trained network and continue training on this new dataset. How can I do this?
class FeedForewardNN(nn.Module):
def __init__(self, input_size, h1_size = 264, h2_size = 128, num_services=8):
super().__init__()
self.input_size = input_size
self.lin1 = nn.Linear(input_size, h1_size)
self.lin2 = nn.Linear(h1_size, h2_size)
self.lin3 = nn.Linear(h2_size, num_services)
self.relu = nn.ReLU()
self.sigmoid = nn.Sigmoid()
def forward(self, x):
x = self.lin1(x)
x = self.relu(x)
x = self.lin2(x)
x = self.relu(x)
x = self.lin3(x)
x = self.sigmoid(x)
return x
This is the architecture of the feedforward Neural Network.
Then I first train on the data set with only 7 classes.
#Create NN
input_size = len(x_columns)
net1 = FeedForewardNN(input_size, num_services=7)
alpha= 0.001
#Define optimizer
optimizer = optim.Adam(net.parameters(), lr=alpha)
criterion = nn.BCELoss()
running_loss = 0
#Training Loop
loss_list = []
auc_list = []
for i in range(len(train_data_x)):
optimizer.zero_grad()
outputs = net1(train_data_x[i])
loss = criterion(outputs, train_data_y[i])
loss.backward()
optimizer.step()
However then, I want to add one additional output node, define the new weights but maintain the old trained weights, and train on this new data set.
I suggest to replace layer with new one, having desired shape, and than partially assign its parameter values with old ones as follows:
def increaseClassifier( m: torch.nn.Linear ):
w = m.weight
b = m.bias
old_shape = m.weight.shape
m2 = nn.Linear( old_shape[1], old_shape[0] +1 )
m2.weight = nn.parameter.Parameter( torch.cat( (m.weight, m2.weight[0:1]) ) )
m2.bias = nn.parameter.Parameter( torch.cat( (m.bias, m2.bias[0:1]) ) )
return m2
class FeedForewardNN(nn.Module):
...
def incrHere(self):
self.lin3 = increaseClassifier( self.lin3 )
UPD:
Can you explain, how these additional weights that come with this new output node are initialized?
The initial weights for new channel come from new layer creation, layer constructor make new parameters with some random initialization, then we are replace part of it with trained weight, and remained part is ready for new training.
m2.weight = nn.parameter.Parameter( torch.cat( (m.weight, m2.weight[0:1]) ) )
I have been learning PyTorch for some weeks. While I was practicing with CIFAR-10 dataset from PyTorch datasets, I also thought of practicing with ImageFolder class, so I found a version of Cifar-10 from Kaggle, where the images were foldered.(I you don't remember PyTorch datasets are in tar.gz format, not in folder structure)
To my utter surprise, in spite of using the same loss function, learning rate and architecture, The Kaggle dataset test set accuracy starts from 0.18 and PyTorch dataset accuracy starts from 0.56 at epoch 1.
Finally after 20 epochs ,one almost saturates near 0.45 and the later one almost fixes near 0.86.
I have checked again and again,but not finding any big difference in those two codes.
I really want to know, if I have done anything deadly wrong, or there is anything fundamentally different about those two datasets.
To clarify, I am using this Pytorch dataset, and this Kaggle dataset .
The codes are too large to be provided here, so I am providing links my notebooks, you are welcome to take a look at my whole code, and also can run if necessary [You only need to use your Kaggle API key to download the dataset from kaggle, I can't make mine one public...sorry for the inconvinience]
Kaggle Dataset Notebook here and Pytorch Dataset Notebook here
I am also providing the chunk of code that I think , is mostly different.
Kaggle Dataset:
Epoch 1 score = 0.18
Epoch 20 score = 0.45
from torch.utils.data import DataLoader
def createVal(train_list, root_folder, classes, valid_split ):
try:
os.mkdir(os.path.join(root_folder, 'val'))
except FileExistsError:
pass
for cls in classes:
try:
os.mkdir(os.path.join(root_folder, 'val', cls))
except FileExistsError:
pass
np.random.shuffle(train_list)
valid_len = len(train_list) * valid_split
for i in tqdm(range(int(valid_len))):
shutil.move(train_list[i], train_list[i].replace('/train/', '/val/'))
valid_split = 0.2
batch_size = 32
num_workers = 4
root_folder = "/content/cifar10/cifar10"
train_folder = os.path.join(root_folder, "train")
test_folder = os.path.join(root_folder, "test")
if valid_split:
createVal(train_list, root_folder, classes, valid_split = valid_split)
val_folder = os.path.join(root_folder, "val")
val_data = datasets.ImageFolder(val_folder, transform = transform)
val_loader = DataLoader(val_data, batch_size = batch_size, num_workers = num_workers )
train_data = datasets.ImageFolder(train_folder, transform = transform)
train_loader = DataLoader(train_data, shuffle = True, batch_size = batch_size, num_workers = num_workers )
test_data = datasets.ImageFolder(test_folder, transform = transform)
test_loader = DataLoader(test_data, batch_size = batch_size, num_workers = num_workers )
Pytorch Dataset:
Epoch 1 score = 0.18
Epoch 20 score = 0.45
valid_split = 0.2
batch_size = 32
num_workers = 4
if valid_split:
num_train = len(train_data)
idx = list(range(num_train))
np.random.shuffle(idx)
train_idx = idx[int(valid_split*num_train):]
val_idx = idx[:int(valid_split*num_train)]
train_sampler = SubsetRandomSampler(train_idx)
val_sampler = SubsetRandomSampler(val_idx)
train_loader = DataLoader(train_data, sampler = train_sampler, batch_size = batch_size, num_workers = num_workers )
val_loader = DataLoader(train_data, sampler = val_sampler, batch_size = batch_size, num_workers = num_workers )
else:
train_loader = DataLoader(train_data, batch_size = batch_size, num_workers = num_workers )
test_loader = DataLoader(test_data, batch_size = batch_size, num_workers = num_workers )
I see , difference in method of shuffling the training dataset.
Kaggle dataset : train_loader > shuffle = True
Pytorch Dataset : train_loader > No shuffle
When using shuffle ==True , it will do RandomSampler function .
I'm using DNN to fit these data, and I use softmax to classify them into 2 class, and each of them has a demensity of 4040, can someone with experience tell me what's wrong with my nets.
It is strange that my initial loss is 7.6 and my initial error is 0.5524, and Basically they won't change anymore.
for train, test in kfold.split(data_pro, valence_labels):
model = keras.Sequential()
model.add(keras.layers.Dense(5000,activation='relu',input_shape=(4040,)))
model.add(keras.layers.Dropout(rate=0.25))
model.add(keras.layers.Dense(500, activation='relu'))
model.add(keras.layers.Dropout(rate=0.5))
model.add(keras.layers.Dense(1000, activation='relu'))
model.add(keras.layers.Dropout(rate=0.5))
model.add(keras.layers.Dense(2, activation='softmax'))
model.add(keras.layers.Dropout(rate=0.5))
model.compile(optimizer=tf.keras.optimizers.RMSprop(learning_rate=0.0001,rho=0.9),
loss='binary_crossentropy',
metrics=['accuracy'])
print('------------------------------------------------------------------------')
print(f'Training for fold {fold_no} ...')
log_dir="logs/fit/" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir, histogram_freq=1)
# Fit data to model
history = model.fit(data_pro[train], valence_labels[train],
batch_size=128,
epochs=50,
verbose=1,
callbacks=[tensorboard_callback]
)
# Generate generalization metrics
scores = model.evaluate(data_pro[test], valence_labels[test], verbose=0)
print(f'Score for fold {fold_no}: {model.metrics_names[0]} of {scores[0]}; {model.metrics_names[1]} of {scores[1]*100}%')
acc_per_fold.append(scores[1] * 100)
loss_per_fold.append(scores[0])
# Increase fold number
fold_no = fold_no + 1
# == Provide average scores ==
print('------------------------------------------------------------------------')
print('Score per fold')
for i in range(0, len(acc_per_fold)):
print('------------------------------------------------------------------------')
print(f'> Fold {i+1} - Loss: {loss_per_fold[i]} - Accuracy: {acc_per_fold[i]}%')
print('------------------------------------------------------------------------')
print('Average scores for all folds:')
print(f'> Accuracy: {np.mean(acc_per_fold)} (+- {np.std(acc_per_fold)})')
print(f'> Loss: {np.mean(loss_per_fold)}')
print('------------------------------------------------------------------------')
You shouldn't add Dropout after the final Dense , delete the model.add(keras.layers.Dropout(rate=0.5))
And I think your code may raise error because your labels's dim is 1 , But your final Dense's units is 2 . Change model.add(keras.layers.Dense(2, activation='softmax')) to model.add(keras.layers.Dense(1, activation='sigmoid'))
Read this to learn tensorflow
Update 1 :
Change
model.compile(optimizer= tf.keras.optimizers.SGD(learning_rate = 0.00001,momentum=0.9,nesterov=True),
loss=tf.keras.losses.CategoricalCrossentropy(),
metrics=['accuracy'])
to
model.compile(optimizer= tf.keras.optimizers.Adam(learning_rate=3e-4),
loss=tf.keras.losses.CategoricalCrossentropy(),
metrics=['accuracy'])
And change
accAll = []
for epoch in range(1, 50):
model.fit(train_data, train_labels,
batch_size=50,epochs=5,
validation_data = (val_data, val_labels))
val_loss, val_Accuracy = model.evaluate(val_data,val_labels,batch_size=1)
accAll.append(val_Accuracy)
to
accAll = model.fit(
train_data, train_labels,
batch_size=50,epochs=20,
validation_data = (val_data, val_labels)
)
I'm using neural nets for a regression problem where I have 3 features and I'm trying to predict one continuous value. I noticed that my neural net start learning good but after 10 epochs it get stuck on a high loss value and could not improve anymore.
I tried to use Adam and other adaptive optimizers instead of SGD but that didn't work. I tried a complex architectures like adding layers, neurons, batch normalization and other activations etc.. and that also didn't work.
I tried to debug and try to find out if something is wrong with the implementation but when I use only 10 examples of the data my model learn fast so there are no errors. I start to increase the examples of the data and monitoring my model results as I increase the data examples. when I reach 3000 data examples my model start to get stuck on a high value loss.
I tried to increase layers, neurons and also to try other activations, batch normalization. My data are also normalized between [-1, 1], my target value is not normalized since it is regression and I'm predicting a continuous value. I also tried using keras but I've got the same result.
My real dataset have 40000 data, I don't know what should I try, I almost try all things that I know for optimization but none of them worked. I would appreciate it if someone can guide me on this. I'll post my Code but maybe it is too messy to try to understand, I'm sure there is no problem with my implementation, I'm using skorch/pytorch and some SKlearn functions:
# take all features as an Independant variable except the bearing and distance
# here when I start small the model learn good but from 3000 data points as you can see the model stuck on a high value. I mean the start loss is 15 and it start to learn good but when it reach 9 it stucks there
# and if I try to use the whole dataset for training then the loss start at 47 and start decreasing until it reach 36 and then stucks there too
X = dataset.iloc[:3000, 0:-2].reset_index(drop=True).to_numpy().astype(np.float32)
# take distance and bearing as the output values:
y = dataset.iloc[:3000, -2:].reset_index(drop=True).to_numpy().astype(np.float32)
y_bearing = y[:, 0].reshape(-1, 1)
y_distance = y[:, 1].reshape(-1, 1)
# normalize the input values
scaler = StandardScaler()
X_norm = scaler.fit_transform(X, y)
X_br_train, X_br_test, y_br_train, y_br_test = train_test_split(X_norm,
y_bearing,
test_size=0.1,
random_state=42,
shuffle=True)
X_dis_train, X_dis_test, y_dis_train, y_dis_test = train_test_split(X_norm,
y_distance,
test_size=0.1,
random_state=42,
shuffle=True)
bearing_trainset = Dataset(X_br_train, y_br_train)
bearing_testset = Dataset(X_br_test, y_br_test)
distance_trainset = Dataset(X_dis_train, y_dis_train)
distance_testset = Dataset(X_dis_test, y_dis_test)
def root_mse(y_true, y_pred):
return np.sqrt(mean_squared_error(y_true, y_pred))
class RMSELoss(nn.Module):
def __init__(self):
super().__init__()
self.mse = nn.MSELoss()
def forward(self, yhat, y):
return torch.sqrt(self.mse(yhat, y))
class AED(nn.Module):
"""custom average euclidean distance loss"""
def __init__(self):
super().__init__()
def forward(self, yhat, y):
return torch.dist(yhat, y)
def train(on_target,
hidden_units,
batch_size,
epochs,
optimizer,
lr,
regularisation_factor,
train_shuffle):
network = None
trainset = distance_trainset if on_target.lower() == 'distance' else bearing_trainset
testset = distance_testset if on_target.lower() == 'distance' else bearing_testset
print(f"shape of trainset.X = {trainset.X.shape}, shape of trainset.y = {trainset.y.shape}")
print(f"shape of testset.X = {testset.X.shape}, shape of testset.y = {testset.y.shape}")
mse = EpochScoring(scoring=mean_squared_error, lower_is_better=True, name='MSE')
r2 = EpochScoring(scoring=r2_score, lower_is_better=False, name='R2')
rmse = EpochScoring(scoring=make_scorer(root_mse), lower_is_better=True, name='RMSE')
checkpoint = Checkpoint(dirname=f'results/{on_target}/checkpoints')
train_end_checkpoint = TrainEndCheckpoint(dirname=f'results/{on_target}/checkpoints')
if on_target.lower() == 'bearing':
network = BearingNetwork(n_features=X_norm.shape[1],
n_hidden=hidden_units,
n_out=y_distance.shape[1])
elif on_target.lower() == 'distance':
network = DistanceNetwork(n_features=X_norm.shape[1],
n_hidden=hidden_units,
n_out=1)
model = NeuralNetRegressor(
module=network,
criterion=RMSELoss,
device='cpu',
batch_size=batch_size,
lr=lr,
optimizer=optim.Adam if optimizer.lower() == 'adam' else optim.SGD,
optimizer__weight_decay=regularisation_factor,
max_epochs=epochs,
iterator_train__shuffle=train_shuffle,
train_split=predefined_split(testset),
callbacks=[mse, r2, rmse, checkpoint, train_end_checkpoint]
)
print(f"{'*' * 10} start training the {on_target} model {'*' * 10}")
history = model.fit(trainset, y=None)
print(f"{'*' * 10} End Training the {on_target} Model {'*' * 10}")
if __name__ == '__main__':
args = parser.parse_args()
train(on_target=args.on_target,
hidden_units=args.hidden_units,
batch_size=args.batch_size,
epochs=args.epochs,
optimizer=args.optimizer,
lr=args.learning_rate,
regularisation_factor=args.regularisation_lambda,
train_shuffle=args.shuffle)
and this is my network declaration:
class DistanceNetwork(nn.Module):
"""separate NN for predicting distance"""
def __init__(self, n_features=5, n_hidden=16, n_out=1):
super().__init__()
self.model = nn.Sequential(
nn.Linear(n_features, n_hidden),
nn.LeakyReLU(),
nn.Linear(n_hidden, 5),
nn.LeakyReLU(),
nn.Linear(5, n_out)
)
here is the log while training:
I am new to TensorFlow. Today I tried to implement my first model in TF but it returned strange results. I know that I am missing something here but I was not able to figure it out. Here is the story.
Model
I have a simple Multilayer Perceptron model with only a single hidden layer applied on MNIST databse. Layers are defined like [input(784) , hidden_layer(470) , output_layer(10)] with tanh as non-linearity for hidden layer and softmax as the loss for output layer. The optimizer I am using is Gradient Descent Algorithm with learning rate of 0.01. My mini batch size is 1 (I am training model with samples one by one).
My implementations :
First I implemented my model in C++ and got around 96% accuracy.Here is the repository : https://github.com/amin2ros/Artificog
I implemented the exact model in TensorFlow but surprisingly the model didn't converge at all. Here is the code.
Code:
import sys
import input_data
import matplotlib.pyplot as plt
from pylab import *
mnist = input_data.read_data_sets("/tmp/data/", one_hot=True)
import tensorflow as tf
# Parameters
learning_rate = 0.1
training_epochs = 1
batch_size = 1
display_step = 1
# Network Parameters
n_hidden_1 = 470 # 1st layer num features
n_input = 784 # MNIST data input (img shape: 28*28)
n_classes = 10 # MNIST total classes (0-9 digits)
# tf Graph input
x = tf.placeholder("float", [None, n_input])
y = tf.placeholder("float", [None, n_classes])
# Create model
def multilayer_perceptron(_X, _weights, _biases):
layer_1 = tf.tanh(tf.add(tf.matmul(_X, _weights['h1']), _biases['b1']))
return tf.matmul(layer_1, _weights['out']) + _biases['out']
# Store layers weight & bias
weights = {
'h1': tf.Variable(tf.random_normal([n_input, n_hidden_1])),
'out': tf.Variable(tf.random_normal([n_hidden_1, n_classes]))
}
biases = {
'b1': tf.Variable(tf.random_normal([n_hidden_1])),
'out': tf.Variable(tf.random_normal([n_classes]))
}
# Construct model
pred = multilayer_perceptron(x, weights, biases)
# Define loss and optimizer
cost = tf.reduce_mean(tf.nn.softmax(pred)) # Softmax loss
optimizer = tf.train.GradientDescentOptimizer(0.01).minimize(cost) #
# Initializing the variables
init = tf.initialize_all_variables()
# Launch the graph
with tf.Session() as sess:
sess.run(init)
# Training cycle
for epoch in range(training_epochs):
avg_cost = 0.
m= 0
total_batch = int(mnist.train.num_examples/batch_size)
counter=0
#print 'count = ' , total_batch
#sys.stdin.read(1)
# Loop over all batches
for i in range(total_batch):
batch_xs, batch_ys = mnist.train.next_batch(batch_size)
label = tf.argmax(batch_ys,1).eval()[0]
counter+=1
sess.run(optimizer, feed_dict={x: batch_xs, y: batch_ys})
wrong_prediction = tf.not_equal(tf.argmax(pred, 1), tf.argmax(y, 1))
missed=tf.cast(wrong_prediction, "float")
m += missed.eval({x: batch_xs, y: batch_ys})[0]
print "Sample #", counter , " - Label : " , label , " - Prediction :" , tf.argmax(pred, 1).eval({x: batch_xs, y: batch_ys})[0] ,\
"- Missed = " , m , " - Error Rate = " , 100 * float(m)/counter
print "Optimization Finished!"
I am very curious why this happens. Any help is appreciated.
Edit:
As commented below definition of cost function was incorrect so it should be like
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(pred,y))
Now model converges :)