KeraSH is a shell-based neural network toolkit. KeraSH was implemented using beer-based programming and is distributed under the Beerware license.

• KeraSH run only on zsh
• Being brave enough to use it

A simple make check is all you need to run KeraSH's tests.

Here is the most simple way to run KeraSH:

$ # Source KeraSH
$ source ./kera.sh

$ # Init a filesystem for our project
$ init_fs

$ # Loading configuration and datafile into KeraSH
$ store_model "xor" ../test_data/test_model ../test_data/test_data ../test_data/test_label

$ # Compiling the model from configuration
$ create_genome "xor" "${MODEL}/xor.model"

$ # Training our Neural Network
$ fit 1000 "xor" 4 4 10 0.1

KeraSH has several interesting features:

• Train and evaluate FFN
• Evolve your topology of FFN to optimize training performance
• Evaluate CNN
• Expose a BLAS that allows you to manipulate matrix and 3D tensors with ZSH

You can get the slides of our KeraSH's lightning talk on the LSE Website

• Jean-Adrien Ducastaing [email protected]
• Audran Doublet [email protected]