simflow 0.0.1

simflow

SimFlow

Ultra portable Deep Learning framework in Numpy

Link to official Documentation : https://00arun00.github.io/SimFlow/

Currently supported features

Layers:

Actication Functions:

• ReLU
• Sigmoid
• Tanh
• LeakyReLU
• Softplus
• exp

Convolutional Layers:

• Convolutional Neural nets
• Dilated Convolutional Layer

Other Layers:

• Dense
• BN_mean (mean only batch norm)
• Batch Normalization

Losses:

• SoftmaxCrossEntropyLoss

Optimizers:

• SGD
• Momentum
• Nestrov Momentum
• RMSProp
• Adagrad
• Adadelta
• Adam

Iterators:

• Full batch
• Mini batch
• Stochastic

Data Loaders:

• MNIST

Installation steps

pip install -r requirements.txt

Sample network/ How to use

import simflow as sf
Data,Labels = sf.data_loader_mnist.load_normalized_mnist_data_flat()

inp_dim = 784
num_classes = 10

#create network
net = sf.Model()
net.add_layer(sf.layers.Dense(inp_dim,200))
net.add_layer(sf.layers.ReLU())
net.add_layer(sf.layers.BN_mean(200))
net.add_layer(sf.layers.Dense(200,num_classes))

#add loss function
net.set_loss_fn(sf.losses.SoftmaxCrossEntropyLoss())

# add optimizer
net.set_optimizer(sf.optimizers.SGD(lr=0.01,momentum=0.9,nestrov=True))

# add iterator
net.set_iterator(sf.iterators.minibatch_iterator())

# fit the training data for 5 epochs
net.fit(Data['train'],Labels['train'],epochs=5)

# pring scores after training
print("Final Accuracies after training :")
print("Train Accuracy: ",net.score(Data['train'],Labels['train'])[1],end=" ")
print("validation Accuracy: ",net.score(Data['val'],Labels['val'])[1],end =' ')
print("Test Accuracy: ",net.score(Data['test'],Labels['test'])[1])

Features currently worked on

Layers:

• dropout
• maxpool / average pool
• PReLU

Regularizers:

• L1
• L2
• elastic net

Optimizers

• Nadam
• Adamax

Testing Features

run the following command to check if all your layers are functional

python -m pytest -v

currently supports

• Dense Layer
• BN_mean Layer
• BN layer
• Conv Layer
• dilatedConv Layer