Visual Studio Code TensorFlow Snippets
This extension includes a set of useful code snippets for developing TensorFlow models in Visual Studio Code.
See Getting started for a quick tutorial on how to use this extension.
Features
The extension includes three sets of snippets.
The first are used to generate entire python files:
tf:trainer: Generates an entire training framework based on learn API.tf:cnn-classifier: Generates a convolutional neural network classifier.tf:mnist: Generates data loader for mnist dataset.tf:cifar10: Generates data loader for CIFAR10 dataset.tf:cifar100: Generates data loader for CIFAR100 dataset.
The second set generate medium size snippets:
tf:resnet-block: Generates a ResNet block.
The third set generate common TensorFlow operations:
tf:import: Imports TensorFlow package.tf:conv2d: Defines a 2D convolution layer.tf:transposed-conv2d: Defines a transposed 2D convolution layer.tf:separable-conv2d: Defines a separable 2D convolution layer.tf:dense: Defines a dense (fully connected) layer.tf:max-pool2d: Defines a max pooling layer.tf:dropout: Defines a dropout layer.tf:batch-norm: Defines a batch normalization layer.tf:cross-entropy: Defines a softmax cross entropy loss.
Installation
Install Visual Studio Code from here.
Install this extension by selecting Install Extension in the command pallette (cmd-shift-p) and searching for "TensorFlow Snippets".
The generated code also relies on the following python dependencies:
pip install numpy
pip install tensorflow # or tensorflow-gpu
pip install six
Getting started
Make sure the requirements are installed.
Create a new directory for your project.
Create trainer.py, open the file and type: tf:trainer enter.
Create cnn.py, open the file and type: tf:cnn-classifier enter.
Create mnist.py, open the file and type: tf:mnist enter.
Edit trainer.py, and insert two lines at the top:
import cnn
import mnist
In the MODELS dictionary, add a new element "cnn": cnn, also in the DATASETS dictionary add a new element "mnist": mnist. Now run the trainer:
python -m trainer --model=cnn --dataset=mnist
Open another terminal window in the same directory, and type:
tensorboard --logdir=output
Point your browser at http://localhost:6006.
