sdmlib 0.1.2

A fast, numpy-based implementation of Kanerva's Sparse Distributed Memory (SDM)

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.

Parameter	Description
N	Length of addresses in bits
M	Number of hard addresses
U	Length of data in bits
d	Hamming radius of addresses considered "near" for reading and writing. If None, then it is computed using T
T	(Default: None) Number of data points to be written. Ignored if d is provided. If T is None then d must be provided. d and T cannot both be None
seed	Seed the random number generator (for reproducability)

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.

Parameter	Description
x	Address. Binary numpy array of shape (N,)
w	Data. Binary numpy array of shape (U,)

The read method

Read a binary string from memory.

Parameter	Description
x	Address. Binary numpy array of shape (N,)

Returns	Description
z	Data. Binary numpy array of shape (U,)