#Copyright (c) 2014, Roland Memisevic
#All rights reserved.
#
#memisevr[at]iro[dot]umontreal[dot]ca
#
#Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
#
#    * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
#    * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
#
#            THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

import numpy 
import numpy.random
import pylab
from dispims_color import dispims_color
import zae 
import train
import theano
from theano.tensor.shared_randomstreams import RandomStreams

rng = numpy.random.RandomState(1)
theano_rng = RandomStreams(1)
SMALL = 0.001
patchsize = 12
numfeatures = 100


import os
HOME = os.environ['HOME']
CIFARDATADIR = HOME+'/research/data/cifar/cifar-10-batches-py'


def crop_patches_color(image, keypoints, patchsize):
    patches = numpy.zeros((len(keypoints), 3*patchsize**2))
    for i, k in enumerate(keypoints):
        patches[i, :] = image[k[0]-patchsize/2:k[0]+patchsize/2, k[1]-patchsize/2:k[1]+patchsize/2,:].flatten()
    return patches


def pca(data, var_fraction, whiten=True):
    """ principal components analysis of data (columnwise in array data), retaining as many components as required to retain var_fraction of the variance 
    """
    from numpy.linalg import eigh
    u, v = eigh(numpy.cov(data, rowvar=0, bias=1))
    v = v[:, numpy.argsort(u)[::-1]]
    u.sort()
    u = u[::-1]
    u = u[u.cumsum()<=(u.sum()*var_fraction)]
    numprincomps = u.shape[0]
    u[u<SMALL] = SMALL
    if whiten: 
        backward_mapping = ((u**(-0.5))[:numprincomps][numpy.newaxis,:]*v[:,:numprincomps]).T
        forward_mapping = (u**0.5)[:numprincomps][numpy.newaxis,:]*v[:,:numprincomps]
    else: 
        backward_mapping = v[:,:numprincomps].T
        forward_mapping = v[:,:numprincomps]
    return backward_mapping, forward_mapping, numpy.dot(v[:,:numprincomps], backward_mapping), numpy.dot(forward_mapping, v[:,:numprincomps].T)




#GET SOME CIFAR IMAGES 
trainimages = (numpy.concatenate([(numpy.load(CIFARDATADIR+'/data_batch_'+b)['data']) for b in ["1"]], 0).reshape(-1,3,32,32)/255.).astype("float32")[:1000]

#CROP PATCHES
print "cropping patches"
trainpatches = numpy.concatenate([crop_patches_color(im.reshape(3, 32, 32).transpose(1,2,0), numpy.array([numpy.random.randint(patchsize/2, 32-patchsize/2, 40), numpy.random.randint(patchsize/2, 32-patchsize/2, 40)]).T, patchsize) for im in trainimages])
R = rng.permutation(trainpatches.shape[0])
trainpatches = trainpatches[R, :]
print "numpatches: ", trainpatches.shape[0]
print "done"

#LEARN WHITENING MATRICES 
print "whitening"
meanstd = trainpatches.std()
trainpatches -= trainpatches.mean(1)[:,None]
trainpatches /= trainpatches.std(1)[:,None] + 0.1 * meanstd
trainpatches_mean = trainpatches.mean(0)[None,:]
trainpatches_std = trainpatches.std(0)[None,:] 
trainpatches -= trainpatches_mean
trainpatches /= trainpatches_std + 0.1 * meanstd
pca_backward, pca_forward, zca_backward, zca_forward = pca(trainpatches, 0.9, whiten=True)
trainpatches_whitened = numpy.dot(trainpatches, pca_backward.T).astype("float32")
trainpatches_theano = theano.shared(trainpatches_whitened)
print "done"

#INSTANTIATE THE ZERO-BIAS AUTOENCODER
model = zae.Zae(numvis=trainpatches_whitened.shape[1], numhid=numfeatures, vistype="real", init_features=0.1*trainpatches_whitened[:numfeatures].T, selectionthreshold=1.0)

assert False, "preprocessing is done, may train now"


#DO SOME STEPS WITH SMALL LEARNING RATE TO MAKE SURE THE INITIALIZATION IS IN A REASONABLE RANGE
trainer = train.GraddescentMinibatch(model, trainpatches_theano, 100, learningrate=0.0001, momentum=0.9)
trainer.step(); trainer.step(); trainer.step() 

#TRAIN THE MODEL FOR REAL, AND SHOW FILTERS 
trainer = train.GraddescentMinibatch(model, trainpatches_theano, 100, learningrate=0.01, momentum=0.9)


for epoch in xrange(100):
    trainer.step()
    if epoch % 10 == 0 and epoch > 0:
        trainer.set_learningrate(trainer.learningrate*0.8)
        dispims_color(numpy.dot(model.W.get_value().T, pca_forward.T).reshape(-1, patchsize, patchsize, 3), 1)
        pylab.draw(); pylab.show()