#!/usr/bin/env python

import scipy

from matplotlib import pyplot

import sklearn

import numpy as np

np.seterr(all='raise')

import lr

import logistic

def add_x2_y2(a):

return logistic.vstack([a.T, a[:,0]**2, a[:,1]**2]).T

def gen_sample(p, n):

return (pos, neg,) + logistic.sample_positive(c, pos, neg)

if __name__ == '__main__':

cs = np.linspace(0.05, 1, 20)

validation_fractions = [0.01, 0.05, 0.10, 0.30, 0.50, 1.0]

table = []

if 'FULL_GRAPH' in locals() and FULL_GRAPH:

speed_multiple = 1

else:

speed_multiple = 5

n_pos = 500 / speed_multiple

mean_pos = [2, 2]

cov_pos = [[1, 1], [1, 4]]

n_neg = 1000 / speed_multiple

mean_neg = [-2, -3]

cov_neg = [[4, -1], [-1, 4]]

gaussian = np.random.multivariate_normal

cs = [0.20,]

print cs

for c in cs:

vf = 0.2

#for vf in validation_fractions:

pos, neg, pos_sample, unlabeled = gen_sample(n_pos, n_neg)

# validation set:

_, _, v_p, v_u = gen_sample(int(vf * n_pos), int(vf * n_neg))

# pos only

X = np.vstack([pos_sample, unlabeled])

y = np.hstack([np.array([1] * pos_sample.shape[0]),

np.array([0] * unlabeled.shape[0]),])

X, y = sklearn.utils.shuffle(X, y)

scaler = sklearn.preprocessing.StandardScaler()

scaler.fit(X)

X = scaler.transform(X)

X = scipy.sparse.csr_matrix(X)

posonly = lr.SGDPosonlyMultinomialLogisticRegression(n_iter=1000, eta0=0.01, c=None)

posonly.fit(X, y)

print 'posonly c:', posonly.final_c()

pos_sample = scipy.sparse.csr_matrix(pos_sample)

unlabeled = scipy.sparse.csr_matrix(unlabeled)

testX = np.vstack([pos, neg])

testy = np.hstack([np.array([1] * pos.shape[0]),

np.array([0] * neg.shape[0]),])

scaler = sklearn.preprocessing.Scaler()

scaler.fit(testX)

testX = scaler.transform(testX)

data = (pos_sample, unlabeled, v_p, v_u)

#data, fixers = logistic.normalize_pu_data(*data)

params, estimators = logistic.calculate_estimators(*data, max_iter=1000)

theta, thetaM, b = params

t = ('vf:', vf, 'c:', c, ) + estimators

print t

table.append(t)

# run the LR on the true data

(thetaTrue, _, _), _ = logistic.calculate_estimators(*(pos, neg, v_p, v_u), max_iter=1000)

# unit area ellipse

fig = pyplot.figure()

ax = fig.add_subplot(111)

ax.scatter(pos[:,0], pos[:,1], s=6, c='b', marker='+')

ax.scatter(neg[:,0], neg[:,1], s=6, c='r', marker='o', lw=0)

delta = 0.01 * speed_multiple

x, y = np.arange(-8, 8, delta), np.arange(-10, 10, delta)

X, Y = np.meshgrid(x, y)

assert X.shape == Y.shape

shape = X.shape

data = np.hstack([X.flatten().reshape(-1, 1), Y.flatten().reshape(-1,1)])

assert data.shape[0] == (shape[0] * shape[1]) and data.shape[1] == 2

data = add_x2_y2(data)

scaled_data = scaler.transform(data)

# plot the LR on the true labels

labels = logistic.label_data(data, thetaTrue, normalizer=0.0, binarize=False)

labels.shape = shape

CS = pyplot.contour(X, Y, labels, [0.50,], colors='#0000FF')

CS.collections[0].set_label('LR True Labels')

labels = logistic.label_data(data, theta, normalizer=0.0, binarize=False)

labels.shape = shape

CS = pyplot.contour(X, Y, labels, [0.10,], colors='#AAAAFF')

CS.collections[0].set_label('LR Pos-only Labels')

labels = posonly.predict_proba(scaled_data)[:,1]

labels.shape = shape

CS = pyplot.contour(X, Y, labels, [0.50,], colors='#00FF00')

CS.collections[0].set_label('POLR Pos-only Labels')

print 'b: ', b

print 'c ~ ', 1.0 / (1.0 + b*b)

labels = logistic.label_data(data, thetaM, normalizer=(b*b), binarize=False)

labels.shape = shape

CS = pyplot.contour(X, Y, labels, [0.10,], colors='r')

CS.collections[0].set_label('CLR Pos-only Labels')

pyplot.title('Logistic regression on synthetic data')

handles, labels = ax.get_legend_handles_labels()

ax.legend(handles, labels, loc=3)

name = 'syntheticlr'

fig.savefig('pdf/%s.png' % name)

if speed_multiple > 1:

fig.savefig('pdf/%s-fast.png' % name)

else:

fig.savefig('pdf/%s-full.png' % name)

if 'SUPPRESS_PLOT' not in locals() or not SUPPRESS_PLOT:

pyplot.show()