#!/usr/bin/env python

import dae_theano

#dae_theano = reload(dae_theano)

mydae = dae_theano.DAE(n_hiddens=56,

batch_size=25)

import debian_spiral

import numpy

n_spiral_samples = 2000

spiral_samples_noise = 0.01

(X,Y) = debian_spiral.sample(n_spiral_samples, spiral_samples_noise)

data = numpy.vstack((X,Y)).T

mydae.fit(data, verbose=True)

mydae.set_params_to_best_noisy()

# print "mydae.W is "

# print mydae.W

# print "mydae.b is "

# print mydae.b

# print "mydae.c is "

# print mydae.c

# print data

import os

# create a new directory to host the result files of this experiment

output_directory = '/u/alaingui/umontreal/cae.py/plots/experiment_%0.6d' % int(numpy.random.random() * 1.0e6)

if not os.path.exists(output_directory):

os.makedirs(output_directory)

import matplotlib

matplotlib.use('Agg')

import pylab

pylab.hold(True)

p1, = pylab.plot(mydae.logging['noisy']['mean_abs_loss'], label='noisy', c='#f9761d', linewidth = 2)

for s in [-1.0, 1.0]:

pylab.plot(mydae.logging['noisy']['mean_abs_loss']

+ s * numpy.sqrt(mydae.logging['noisy']['var_abs_loss']),

c='#f9a21d', linestyle='dashed')

p2, = pylab.plot(mydae.logging['noiseless']['mean_abs_loss'], label='noiseless', c='#9418cd', linewidth = 2)

for s in [-1.0, 1.0]:

pylab.plot(mydae.logging['noiseless']['mean_abs_loss']

+ s * numpy.sqrt(mydae.logging['noiseless']['var_abs_loss']),

c='#d91986', linestyle='dashed')

pylab.title('Absolute Losses')

pylab.legend([p1,p2], ["noisy", "noiseless"])

pylab.draw()

pylab.savefig(os.path.join(output_directory, 'absolute_losses.png'), dpi=300)

pylab.close()

################################

## ##

## spiral reconstruction grid ##

## ##

################################

plotgrid_N_buckets = 30

window_width = 1.0

(plotgrid_X, plotgrid_Y) = numpy.meshgrid(numpy.arange(- window_width,

window_width,

2 * window_width / plotgrid_N_buckets),

numpy.arange(- window_width,

window_width,

2 * window_width / plotgrid_N_buckets))

plotgrid = numpy.vstack([numpy.hstack(plotgrid_X), numpy.hstack(plotgrid_Y)]).T

D = numpy.sqrt(plotgrid[:,0]**2 + plotgrid[:,1]**2)

plotgrid = plotgrid[D<0.7]

print plotgrid_X.shape

print plotgrid_Y.shape

print "Will keep only %d points on the plotting grid after starting from %d." % (plotgrid.shape[0], plotgrid_X.shape[0])

print "Making predictions for the grid."

grid_pred = mydae.encode_decode(plotgrid)

#grid_pred = predict(plotgrid, W, grid, (lambda X, xi: kernel(X,xi,sigma)))

grid_error = numpy.sqrt(((grid_pred - plotgrid)**2).sum(axis=1)).mean()

print "grid_error = %0.6f" % grid_error

print "Generating plot."

# print only one point in 100

# pylab.scatter(data[0:-1:100,0], data[0:-1:100,1], c='#f9a21d')

pylab.scatter(data[:,0], data[:,1], c='#f9a21d')

pylab.hold(True)

arrows_scaling = 1.0

pylab.quiver(plotgrid[:,0],

plotgrid[:,1],

arrows_scaling * (grid_pred[:,0] - plotgrid[:,0]),

arrows_scaling * (grid_pred[:,1] - plotgrid[:,1]))

pylab.draw()

#pylab.axis([-0.6, 0.6, -0.6, 0.6])

pylab.axis([-0.7, 0.7, -0.7, 0.7])

# pylab.axis([-window_width*1.5, window_width*1.5, -window_width*1.5, window_width*1.5])

pylab.savefig(os.path.join(output_directory, 'spiral_reconstruction_grid.png'), dpi=300)

pylab.close()

###################################

## ##

## html file for showing results ##

## ##

###################################

html_file_path = os.path.join(output_directory, 'results.html')

f = open(html_file_path, "w")

hyperparams_contents = """

mydae.jacobi_penalty)

#n_spiral_samples = 30

#spiral_samples_noise = 0.0

params_contents = ""

graphs_contents = """

<img src='%s' width='600px'/>

<img src='%s' width='600px'/>

""" % ('absolute_losses.png', 'spiral_reconstruction_grid.png')

contents = """

graphs_contents)

f.write(contents)

f.close()