sdwfrost

/* BEGIN node2umd PREFIX */
;(function(define) { define(function(require, exports, module) {
/* END node2umd PREFIX */
/**
 * gpu.js
 * http://gpu.rocks/
 *
 * GPU Accelerated JavaScript
 *
 * @version 1.0.0-rc.10

sdwfrost

/*
These are the helper functions to store and to restore a 2D vector with a custom 16 floating point precision in a texture.
The 16 bit are used as follows: 1 bit is for the sign, 4 bits are used for the exponent, the remaining 11 bit are for the mantissa.

The exponent bias is asymmetric so that the maximum representable number is 2047 (and bigger numbers will be cut)
the accuracy from 1024 - 2047 is one integer
512-1023 it's 1/2 int
256-511 it's 1/4 int and so forth...
between 0 and 1/16 the accuracy is the highest with 1/2048 (which makes 1/32768 the minimum representable number)

sdwfrost

/* This is a naive implementation of Jonathan McCabe's elaboration of
 * Alan Turing's model of morphogenesis, as described here:
 * http://www.jonathanmccabe.com/Cyclic_Symmetric_Multi-Scale_Turing_Patterns.pdf
 */

extern crate rand;

use rand::Rng;
use std::env;
use std::io::Write;

sdwfrost

sdwfrost

This gist compares the performance of Julia, Nim, C++ and R - the latter using either POMP, or LibBi in a simple simulation of an SIR epidemiological model. In addition to keeping track of susceptibles, infecteds and recovereds, I also store the cumulative number of infections. Time moves in discrete steps, and the algorithm avoids language-specific syntax features to make the comparison as fair as possible, including using the same algorithm for generating binomial random numbers and the same random number generator; the exception are the R versions, POMP uses the standard R Mersenne Twister for the random number generator; I'm not sure what LibBi uses. The algorithm for generating random binomial numbers is only really suitable for small np.

Benchmarks were run on a Mac Pro (Late 2013), with 3 Ghz 8-core Intel Xeon E3, 64GB 1866 Mhz RAM, running OSX v 10.11.3 (El Capitan

sdwfrost

sdwfrost

sdwfrost

# Example for my blog post at:
# https://danijar.com/introduction-to-recurrent-networks-in-tensorflow/
import functools
import sets
import tensorflow as tf


def lazy_property(function):
    attribute = '_' + function.__name__

sdwfrost

import numpy as np

from utils import orthogonal, tanh, sigmoid, dtanh, dsigmoid


class LSTM(object):
	"""Long Short Term Memory Unit
	
	Parameters
	----------

sdwfrost

###
### This is a batched LSTM forward and backward pass. Written by Andrej Karpathy (@karpathy)
### BSD License
### Re-written in R by @georgeblck
###

rm(list=ls(all=TRUE))

LSTM.init <- function(input_size, hidden_size, fancy_forget_bias_init = 3){
  # Initialize parameters of the LSTM (both weights and biases in one matrix)