Sparse Distributed Memory for Python

What is this?

This is a fast implementation of Pentti Kanerva's Sparse Distributed Memory (SDM). sdmlib is written in Python using numpy.

Installation

The package is available through PyPi:

pip install sdmlib

Using sdmlib

sdmlib is designed to operate on numpy arrays.

The Memory class

This class provides the main functionality of SDM.

Example Usage

import numpy as np
from sdmlib import Memory

N = 256
M = 1000
U = 256
T = 100

addresses =  np.random.randint(low=0, high=2, size=(T, N), dtype=np.uint8)
data      =  np.random.randint(low=0, high=2, size=(T, U), dtype=np.uint8)

mem = Memory(N=N, M=M, U=U, d=None, T=T)

for t in range(T):
    mem.write(addresses[t], data[t])

error = 0
for t in range(T):
    error += np.mean(data[t] != mem.read(addresses[t]))/T
print(f'Reconstruction error: {100*error:.2f}%')

Reconstruction error: 0.51%

The write method

Write a binary string to memory.

The read method

Read a binary string from memory.