Documentation

NORSAR Artificial Inteligence System (NAIS, yes, the acronym was chosen first)

This repository will eventually contain tools specific to aiding in AI research at NORSAR.

This repository is based on Tensorflow and Keras.

This repository will contain:

• Baseline models, for waveforms and spectrograms. Possibly pretrained for certain tasks.
• Standard datasets to test models. Development of new models can be difficult. Standard datasets eliminates errors in the data and lets you focus on developing the model.
  • Classification
  • Regression
  • Masking (eg. arrival picking)
• Augmentation methods.

Installation

pip install nais

If an error occurs when importing nais, likely sndfile library is not installed (check the error), and needs to be:

apt-get install libsndfile1-dev

Quick example

import numpy as np
from nais.Models import AlexNet1D

X = np.random.normal(size=(16,256,3)) #16 examples of three channel data.
y = np.random.randint(0,1,size=(16,)) #Labels

model = AlexNet1D(num_outputs=1) #binary 
model.compile('adam','binary_crossentropy')
model.fit(X,y)