Cerebro

A tool for NuPIC to run CLA models and visually examine the output of the Spatial Pooler. Click here for a video introduction and tutorial.

What is Cerebro?

• A DVR for the CLA algorithm
• A visual debugging tool for the CLA
• A dataset generator
• A local web app
• Another client of the OPF
• A tool for finding mutants

How do I use it?

To install (with Homebrew for Mac OS X):

brew install libevent
brew install mongodb
pip install -r requirements.txt

To launch:

mongod &
python cerebro.py <portnum>

In a browser, navigate to:

http://localhost:<portnum>

To build the docker image:

sudo docker build -t="nupic.cerebro" .

To run the docker image:

sudo docker run -p=1955:<portnum> <image name> 

Datasets

File based

• Need a description.py file as input
• Can load a base and sub description.py file
• Path to data file should be in the description.py file

Function based

• Write a function to procedurally generate datasets
• Python
• Provides a number of useful utilities for generating datasets
• Can save to CSV

Function Data Details

• Function written in python. Don’t need to include def funcname()
• Dictionary fields should be filled with data fields. Fields can be strings or scalars
• Utility variable history passed to function. Used to create sequential patterns. history[-1][‘foo’] retrieves the field value of ‘foo’ from 1 timestep ago
• Several useful libraries are imported on your behalf:
  • Random
  • Numpy
  • String
  • Math
• Provides some syntactic sugar for adding randomness to datasets
  • If a field is set to a list-like, values are randomly sampled from set.�Ex. fields[‘f’] = (‘A’, ‘B’, ‘C’, ‘D’) => Randomly sample ‘A’, ‘B’, ‘C’, D’�Ex. Fields[‘foo’] = list(string.uppercase) => randomly sample from uppercase letters
  • If a field is set to a dict-like, values are randomly sampled according to likelihood ratios.�Ex. fields[‘f’] = {‘A’: 1.0, ‘B’:2.0} => Randomly sample ‘A’, ‘B’, where ‘B’ is twice as likely.

Experiment Pane

• Show’s predicted vs. actual graphs
• If present, also shows anomaly score graph
• On right, shows predicted and active columns
• Shows encoder representations on the bottom
• At the very bottom, shows a whole bunch of textual output.
• Verbose output: captures everything from stdout while CLA processes a single record