Python 3.7 version of David Barber's MATLAB BRMLtoolbox
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
Demo
brml
check moved array, potential and const to prob_table package Aug 6, 2018
.gitignore Added gitignore file to not track changes to PyCharm environment file… Aug 1, 2018
LICENSE
README.md
google2b7259c3debd1a24.html
google9b26102d8bd84b8b.html

README.md

PythonBRMLtoolbox

What is PythonBRMLtoolbox?

BRMLtoolbox in Python 3.7. My goal with this repository is to:

  1. Learn BRMLtoolbox with an insider perspective
  2. Provide a Python package for BRMLtoolbox which is available in MATLAB, Python 2.7 and Julia.
  3. Improve my Python skills

Implementation details

Variables

When initializing a potential we can pass variables as arguments in different ways:

  • We can pass a scalar value (int or float) when we only have 1 variable. (We could also pass a list or a numpy array with only one value)
import numpy as np
from brml.array import Array
pot = Array(variables=1)
pot2 = Array(variables=1.0)
pot3 = Array(variables=[1])
pot4 = Array(variables=np.array([1]))
pot5 = Array(variables=np.array([[1]]))
  • We can pass a list or a numpy.array when we have multiple variables.
import numpy as np
from brml.array import Array
pot = Array(variables=[1, 2])
pot2 = Array(variables=np.array([1, 2]))
pot3 = Array(variables=np.array([[1, 2]]))
pot4 = Array(variables=np.array([[1], [2]]))

All of these calls are equivalent among them because they are all converted to a 1D numpy.array (notice that in MATLAB, because 1D arrays don't exist, this is different). It might be good to discuss whether it is better to convert all of these to 1D numpy.array or to a 2D array with shape (1, n) where n is the number of variables.

It is worth mentioning that you don't need to pass variables on initialization. You can initialize an array without variables and without table, and add the later. For example:

import numpy as np
from brml.array import Array
pot = Array()
pot.set_variables(np.array([1, 2]))

Notice, however, that you CANNOT initialize variables before initializing the table. That is, you cannot initialize an empty array pot = Array() and then call pot.set_table(np.array([0.4, 0.6])) because this will throw an error. You must first use set_variables method, and after use the set_table method. If, instead, you decide to pass variables and table on initialization, then everything will be fine.

Go to check/potential_variables.py and you'll find a test to check that the statements above are correct!

What is BRMLtoolbox?

The BRMLtoolbox is a MATLAB toolbox written by David Barber, Reader for the Computational Statistics and Machine Learning MSc at UCL. The BRML toolbox should be used together with the book Bayesian Reasoning and Machine Learning. It is a set of tools to help readers see how mathematical models translate into actual code.

Software Requirements

  1. Python (3.6 onwards, although only type hints would need to be removed for Python 3.5)
  2. Numpy

Mathematical Requirements

The mathematical requirements for understanding classes within BRML toolbox are:

  1. Calculus & Linear Algebra
  2. Probability
  3. Algorithms
  4. Statistics (Bayesian mainly) I am currenlty trying to write an introduction to Probability here in my Proofwiki account, although it is just a draft for now.

Contributions

Please feel free to contribute to this python version of the BRML toolbox! There are already a few incomplete implementations of the code, hopefully this version will be more up-to-date and efficient with your help!

Reference & Book

As I said earlier BRML toolbox comes with Bayesian Reasoning and Machine Learning book.

@BOOK{barberBRML2012,
author = {Barber, D.},
title= {{Bayesian Reasoning and Machine Learning}},
publisher = {{Cambridge University Press}},
year = 2012}