#!/usr/bin/env python3

# Miscellaneous utility functions

import os

import scipy.io as sio

import numpy as np

import glob

def nsubplots(n):

'''Returns [ynum, xnum], which how many plots in the y and x directions to

cover n in total while keeping the aspect ratio close to rectangular'''

if n==2:

ynum = 2; xnum = 2;

else:

xnum = np.ceil(np.sqrt(n));

ynum = np.ceil(n/np.float(xnum));

return ynum, xnum

def add_ones(X):

"""Add a column of ones to X"""

n = len(X)

return np.column_stack((np.ones(n), X))

def degexpand(X, deg, add_ones=False):

"""Expand input vectors to contain powers of the input features"""

n = len(X)

xx = X

for i in xrange(1, deg):

xx = np.column_stack((xx, np.power(X, i + 1)))

if add_ones:

xx = np.column_stack((np.ones(n), xx))

return xx

def rescale_data(X, min_val=-1, max_val=1, minx=None, rangex=None):

"""

Rescale columns to lie in the range

[min_val, max_val] (defaults to [-1,1]])

"""

if minx is None:

minx = X.min(axis=0)

if rangex is None:

rangex = X.max(axis=0) - X.min(axis=0)

return (max_val - min_val) * (X - minx) / rangex + min_val

def center_cols(X, mu=None):

"""

Make each column be zero mean

"""

if mu is None:

mu = X.mean(axis=0)

return X - mu, mu

def mk_unit_variance(X, s=None):

"""

Make each column of X be variance 1

"""

if s is None:

s = X.std(axis=0)

try:

len(s)

s[s < np.spacing(1)] = 1

except TypeError:

s = s if s > np.spacing(1) else 1

return X / s, s

class preprocessor_create():

def __init__(self, standardize_X=False, rescale_X=False, kernel_fn=None,

poly=None, add_ones=False):

self.standardize_X = standardize_X

self.rescale_X = rescale_X

self.kernel_fn = kernel_fn

self.poly = poly

self.add_ones = add_ones

def poly_data_make(sampling="sparse", deg=3, n=21):

"""

Create an artificial dataset

"""

np.random.seed(0)

if sampling == "irregular":

xtrain = np.concatenate(

(np.arange(-1, -0.5, 0.1), np.arange(3, 3.5, 0.1)))

elif sampling == "sparse":

xtrain = np.array([-3, -2, 0, 2, 3])

elif sampling == "dense":

xtrain = np.arange(-5, 5, 0.6)

elif sampling == "thibaux":

xtrain = np.linspace(0, 20, n)

xtest = np.arange(0, 20, 0.1)

sigma2 = 4

w = np.array([-1.5, 1/9.])

fun = lambda x: w[0]*x + w[1]*np.square(x)

if sampling != "thibaux":

assert deg < 4, "bad degree, dude %d" % deg

xtest = np.arange(-7, 7, 0.1)

if deg == 2:

fun = lambda x: (10 + x + np.square(x))

else:

fun = lambda x: (10 + x + np.power(x, 3))

sigma2 = np.square(5)

ytrain = fun(xtrain) + np.random.normal(0, 1, xtrain.shape) * \

np.sqrt(sigma2)

ytestNoisefree = fun(xtest)

ytestNoisy = ytestNoisefree + np.random.normal(0, 1, xtest.shape) * \

np.sqrt(sigma2)

return xtrain, ytrain, xtest, ytestNoisefree, ytestNoisy, sigma2

def load_mat(matName):

"""look for the .mat file in pmtk3/pmtkdataCopy/

currently only support .mat files create by Matlab 5,6,7~7.2,

"""

print 'looking for ', matName, ' in ', DATA_DIR

try:

data = sio.loadmat(os.path.join(DATA_DIR, matName))

except NotImplementedError:

raise

except FileNotFoundError:

raise

return data

def generate_rst():

"""generate chX.rst in current working directory"""

cwd = os.getcwd()

demo_dir = os.path.join(cwd, 'demos')

chapters = os.listdir(demo_dir)

for chapter in chapters:

if not os.path.isdir(os.path.join(demo_dir, chapter)):

continue

reg_py = os.path.join(demo_dir, chapter, '*.py')

scripts = glob.glob(reg_py)

rst_file = chapter + '.rst'

rst_file = os.path.join(demo_dir, chapter, rst_file)

with open(rst_file, 'w') as f:

f.write(chapter)

f.write('\n========================================\n')

for script in scripts:

script_name = os.path.basename(script)

f.write('\n' + script_name[:-3])

f.write('\n----------------------------------------\n')

reg_png = os.path.join(demo_dir,

chapter,

script_name[:-3] + '*.png')

for img in glob.glob(reg_png):

img_name = os.path.basename(img)

f.write(".. image:: " + img_name + "\n")

f.write(".. literalinclude:: " + script_name + "\n")

if __name__ == '__main__':

generate_rst()

print("Finished generate chX.rst!")