Torch-Keras

Recreating Keras with PyTorch

functional_api_v1 - An easy to use framework inspired by Tensorflow's keras but for PyTorch users.

Installing the library

pip install git+https://github.com/bipinKrishnan/torch_keras

Usage

• Training convolutional neural network on CIFAR-100 dataset using torch_keras

import torch
from torch import nn, optim
import torchvision
from torchvision import transforms
from torch.utils.data import DataLoader

from torch_keras import Input, Model
from torch_keras.layers import Dense, Conv2d, Flatten


bs = 128
device = 'cuda' if torch.cuda.is_available() else 'cpu'
transform = transforms.ToTensor()

# prepare train and test sets
trainset = torchvision.datasets.CIFAR100(
    root='./data', 
    train=True, 
    download=True, 
    transform=transform
    )
testset = torchvision.datasets.CIFAR100(
    root='./data', 
    train=False, 
    download=True, 
    transform=transform
    )

trainloader = DataLoader(trainset, batch_size=bs, shuffle=True)
testloader = DataLoader(testset, batch_size=bs)

# build the model architecture
input = Input((3, 32, 32))
x = Conv2d(3, 3, 1, 'same', 1, nn.ReLU())(input)
y = Conv2d(5, 3, 1, 0, 1, nn.ReLU())(x)
z = Conv2d(6, 3, 1, 'same', 1, nn.ReLU())(y)
a = Flatten()(z)
b = Dense(100, activation=nn.ReLU())(a)

model = Model(input, b, device)
model.compile(optim.Adam(model.parameters(), lr=0.001), nn.CrossEntropyLoss())

print(model.summary())

# train the model
model.fit_generator(trainloader, 3)

#### Output
# Epoch 1/3
# 391/391 [=========================] - 27s 69ms/step - train_loss: : 4.5869
# Epoch 2/3
# 391/391 [=========================] - 30s 76ms/step - train_loss: : 4.5388
# Epoch 3/3
# 391/391 [=========================] - 28s 72ms/step - train_loss: : 4.5173

# evaluate the model
model.evaluate_generator(testloader)

#### Output
# 79/79 [=========================] - 3s 42ms/step - eval_loss: 2.3131

# get the predictions
out = model.predict_generator(testloader)