init.lua

require 'torch'   -- torch
require 'os'   --
require 'nn'      -- provides a normalization operator
-- require 'cunn'
require 'xlua'    -- xlua provides useful tools, like progress bars
require 'optim'   -- an optimization package, for online and batch methods
require "libhtktoth"

require 'libpreparedata'
require 'data'
require 'model'

if not (opt) then
    cmd = torch.CmdLine()
    cmd:text()
    cmd:text('Speaker Verification with DNN')
    cmd:text()
    cmd:text('Options:')
    -- filelist:
    cmd:option('-featfile', '', 'name a file storing all the filenames of data')
    cmd:option('-maxrows', 4000, 'max number of rows to be read from fbank file each time')
    cmd:option('-scpfile', '', 'name a file storing all the filenames of train or test data')
    cmd:option('-filenum', 20, 'max nb of fbank file each time')
    cmd:option('-labelfile', '', 'name a file storing the labels for each file in scp')
    cmd:option('-cvscpfile', '', 'name a file storing all the filenames of cv data')
    cmd:option('-globalnorm', '', 'normalization file contains the means and variances')
    -- global:
    cmd:option('-seed', 1, 'fixed input seed for repeatable experiments')
    cmd:option('-threads', 4, 'number of threads')
    -- data:
    cmd:option('-size', 'full', 'how many samples do we load: small | full | extra')
    -- model:
    cmd:option('-model', 'deepneunet', 'type of model to construct: linear | mlp | convnet | deepneunet')
    cmd:option('-ldmodel', 'model.net', 'name of the model to be loaded')
    cmd:option('-modelPara', '', 'model file which stores pretrained weights and bias format as DNN fintune')
    cmd:option('-hidlaynb', 0, 'nb of hidden layers')
    cmd:option('-noutputs',0,'nb of output neurons')
    cmd:option('-inputdim',0,'nb of the single input')
    cmd:option('-fext',5,'nb of frames which will be extended')
    -- loss:
    cmd:option('-loss', 'nll', 'type of loss function to minimize: nll | mse | margin')
    -- training:
    cmd:option('-save', 'results', 'subdirectory to save/log experiments in')
    cmd:option('-plot', false, 'live plot')
    cmd:option('-optimization', 'SGD', 'optimization method: SGD | ASGD | CG | LBFGS')
    cmd:option('-learningRate', 2, 'learning rate at t=0')
    cmd:option('-batchSize', 10, 'mini-batch size (1 = pure stochastic)')
    cmd:option('-weightDecay', 0, 'weight decay (SGD only)')
    cmd:option('-momentum', 0.7, 'momentum (SGD only)')
    cmd:option('-t0', 1, 'start averaging at t0 (ASGD only), in nb of epochs')
    cmd:option('-maxIter', 2, 'maximum nb of iterations for CG and LBFGS')
    cmd:option('-type', 'double', 'type: double | float | cuda')
    cmd:option('-crossvalid', 0, 'use test for cross validaton set which do not extract bottleneck feature and compute the accuracy,0 is false, 1 is true')
    cmd:text()
    opt = cmd:parse(arg or {})
end
print '==> processing options'
-- nb of threads and fixed seed (for repeatable experiments)
if opt.type == 'float' then
   print('==> switching to floats')
   torch.setdefaulttensortype('torch.FloatTensor')
elseif opt.type == 'cuda' then
   print('==> switching to CUDA')
   require 'cunn'
   torch.setdefaulttensortype('torch.FloatTensor')
end
torch.setnumthreads(opt.threads)
torch.manualSeed(opt.seed)

assert(opt.noutputs ~= 0,"Please define a number of outputs with -noutputs")
assert(opt.inputdim ~= 0,"Please define the (static) dimension of inputs with -inputdim")

print '==> define parameters'
-- hidden units (for creating new model or loading model from binary)
nstates = {1024,1024,1024,1024}
filtsizew = 11
filtsizeh = 3
poolsize = 2
-- number of units in output layer, but meaningless in loading model from binary file
noutputs = opt.noutputs
-- number of frame extension to each direction
frameExt = opt.fext
-- [Number of incorelated features], [Width and Height for each feature map(height is the extended frame)], [Number of units in input layer] (for creating new model only)

-- nfeats = 3
width = opt.inputdim
height = 2*frameExt+1
ninputs = width*height
-- number of hidden units (for MLP only):
nhiddens = ninputs / 2
-- number of hidden units for the output of Convolution and pooling layers(2 convolutional and pooling layers)
height2 = math.floor((math.floor((height-filtsizeh+1)/poolsize)-filtsizeh+1)/poolsize)
width2 = math.floor((math.floor((width-filtsizew+1)/poolsize)-filtsizew+1)/poolsize)




if (opt.modelPara~='') then
    modelfromparameterfile()
else
    local filename = paths.concat(opt.save, opt.ldmodel)
    model = io.open(filename, 'rb')
    if (model)then
        loadmodelfromfile(filename)
    else
        newmodel()
    end
end