hal-ml 0.2.0

HAL: a machine learning library that is able to run on Nvidia, OpenCL or CPU BLAS based compute backends. It currently provides stackable classical neural networks, RNN's and soon to be LSTM's. A differentiation of this package is that we are looking to implement RTRL (instead of just BPTT) for the recurrent layers in order to provide a solid framework for online learning. We will also (in the future) be implementing various layers such as unitary RNN's, NTM's and Adaptive Computation time based LSTM's. HAL also comes with the ability to plot and do many basic math operations on arrays.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32

extern crate arrayfire as af;
extern crate itertools;
extern crate rand;
extern crate csv;
extern crate num;
extern crate conv;
extern crate tar;
extern crate flate2;
extern crate hyper;
extern crate statistical;
extern crate rustc_serialize;

pub use layer::{Layer};
pub mod layer;

pub use model::{Model};
pub mod model;

pub use optimizer::{Optimizer};
pub mod optimizer;

pub use data::{DataSource, Data, DataParams, Normalize, Shuffle};
pub mod data;

pub mod params;
pub mod error;
pub mod loss;
pub mod activations;
pub mod initializations;
pub mod plot;
pub mod utils;
pub mod device;