What is brain4k?

brain4k, pronounced "brainfork", is a framework that intends to improve the reproducability and shareability of machine learning models. Think of it as MVC for machine learning, except with brain4k pipelines you have Data, Transforms, and Metrics.

The rules of brain4k:

1. A brain4k pipeline lives in version control, and so can be forked, reverted and managed like other code.
2. Each stage in the pipeline is deterministic and reproducible for a given commit
3. brain4k is framework and language agnostic - pipe from one language to another if your execution environment supports it
4. Every brain4k pipeline publishes performance metrics to encourage quality models to rise to the top
5. Contribute plugins for your preferred ML framework back to the community

Sample pipelines

• Extract image features from a convolutional neural network
• Train a classifier on image features extracted from a convolutional neural network

Wishlist

• Deep Q-Learning
  • Caffe implementation, usage instructions
  • Theano/Pylearn2/RLGlue
  • Cuda convnet2
  • Torch Arcade Learning Environment Wrapper, dataset
• Scene Recognition
  • Using Caffe's Places205-CNN model
• Object Localization
  • Using Overfeat
  • Using VGG team's Caffe model
• Semantic modelling using RNNs
• Deconvolutional Autoencoder
  • Using Caffe's deconvolution operation
• Text Understanding from Scratch
• Semantic modelling using RNNs
• Scene Recognition
  • Using Caffe's Places205-CNN model
• Semantic Segmentation
  • Using FCN-16s PASCAL
• [Facial Keypoints Detection]
  • Torch and dp
  • Lasange and Theano
• Analysis of financial time series data
• Text Understanding from Scratch
• Street View House Numbers classifier
  • Using Caffe's NiN model
  • Torch example
• Semantic Segmentation
  • Using FCN-16s PASCAL

Installing brain4k

# Install Python
# easy_install pip
pip install git+https://github.com/shuggiefisher/brain4k.git

Executing a pipeline

Clone one of our sample pipelines

git clone https://github.com/shuggiefisher/kitten-or-bear-classifier.git local-path-to-repo

then, ensuring you have installed the dependencies listed in the README of the repo you just cloned

brain4k local-path-to-repo

Publishing a pipeline

1. Push your repo somewhere public
2. If you want others to be able to reproduce your model, change any data blobs with a "local_filename" in favour of a "url" and sha1 hash for the file.
3. Tell us about it.