Artificial Neural Network to study HMC



PRE-REQUISITE:
==============
numpy
scipy
python2.7


RUNNING:
========
python ann.py < input file path >



OPTIONS:
========

an input file with the following options needs to be specificed to run the program

input_vector: 
-------------
path to the file containint input unit values for each subject

output_vector: 
-------------
path to the expected output file

num_hidden_units: 
-----------------
number of hidden units to be used

hidden_weights:
---------------
can either supply the path to a file which can be loaded directly as a weights file for hidden units or provide variance of a gaussian from which the weights will be drawn.

hidden_biases:
-------------
same as hidden_weights

output_weights:
---------------
same as hidden_weights

output_biases:
--------------
same as hidden_weights

initialise:
----------
whether to run the initialise routine: possible values: "true" or "false"

initialise_steps:
----------------
number of iterations in the initialisation routine

initialise_eps:
--------------
step-size (epsilon) for the initialise routine

initialize_verbose:
-------------------
whether or not to print iteration-wise state of the neural network. [ not implemented in current version ] 

descent:
-------
what descent algorithm to use "log"(for log-like) or "error"(for RMSD)

hmc:
----
whether to call the HMC routine

hmc_steps:
---------
number of steps to run HMC for.
 	
hmc_eps:
-------
step-size(epsilon) value for HMC steps

hmc_verbose:
-----------
whether or not to print iteration-wise state of the neural network. [ not implemented in current version ] 

log_on:
-------
whether to include log-like contribution in the hamiltonian for HMC: Possible values being "true" or "false"

prior_on:
--------
whether to include prior contribution in the hamiltonian for HMC: Possible values being "true" or "false"

precision:
----------
machine precision for all values: Possible values are "single" or "double"

ard_prior_scale:
---------------
scale parameter for the ARD prior of the hidden layer

ard_prior_shape:
---------------
shape parameter for the ARD prior of the hidden layer

ard_init:
---------------
initial value of the  ARD prior of the hidden layer. [Implemented to imitate Andy's code ]