'''

Testing out SGDClassifier's parameters, especially how to use 'fit' and

'partial_fit'

'''

import numpy as np

import scipy

from sklearn.linear_model import SGDClassifier

from sklearn.metrics import accuracy_score

if __name__ == '__main__':

############### Example 1 ###############

# Example 1: Generate fake data (static, small, super dumb)

print "Example 1"

X = np.array([[-1, -1], [-2, -1], [1, 1], [2, 1]]) # np.ndarray (4L, 2L)

y = np.array([1, 1, 2, 2]) # np.ndarray (4L,)

# fit with example 1 (fit)

clf_1 = SGDClassifier(alpha=.0001, loss='log', penalty='l2', n_jobs=-1,

#shuffle=True, n_iter=10,

verbose=1)

clf_1.fit(X, y)

print "Prediction is: ", clf_1.predict([[-0.8, -1]])

############### Example 2 ###############

# Example 2: Generate fake data (complex, large)

print "\nExample 2"

np.random.seed(1234)

nsamples = 100

nfeatures = 1000

numOnesInX = np.int(0.01 * nsamples * nfeatures) # 1000 # <type 'int'>

nlabels = 4

row_array = np.random.randint(0, nsamples, size=numOnesInX)

# row_array is: [47 83 38 53 ... ] <type 'numpy.ndarray'> (1000L,)

col_array = np.random.randint(0, nsamples, size=numOnesInX)

# col_array is: [82 44 48 26 ... ] <type 'numpy.ndarray'> (1000L,)

data = np.ones_like(row_array)

# data is: [1 1 1 1 1 1 ... ] <type 'numpy.ndarray'> (1000L,)

# data is converted into 'Compressed Sparse Rows representation'

# csr_matrix((data, (row_ind, col_ind)), [shape=(M, N)])

X = scipy.sparse.csc_matrix((data, (row_array, col_array)),

shape=(nsamples, nfeatures))

# X is: <class 'scipy.sparse.csc.csc_matrix'>

# (5, 0) 1

# (11,0) 1

y = np.random.randint(0, 2, size=[nsamples, nlabels])

# Y is: <type 'numpy.ndarray' (100L, 4L)

# [[0 1 1 0]

# [1 1 0 1]

# ... [1 0 0 0] ]

# partial_fit with partial_fit(X, y[, classes, sample_weight])

clf_2 = SGDClassifier(alpha=.0001, loss='log', penalty='l2', n_jobs=-1,

#shuffle=True, n_iter=10,

verbose=1)

print "Running Partial Fit 1"

clf_2.partial_fit(X[:99, :], y[:99, 0], classes=[0, 1])

pred = clf_2.predict(X[:10, :])

print "Prediction is: ", pred # [1 1 0 0 1 1 0 0 1 1]

print "Accuracy Score is: ", accuracy_score(pred, y[:10, 0]) # gives 0.9

print "Running Partial Fit 2"

clf_2.partial_fit(X[:99, :], y[:99, 0]) # fit again on same data!

pred = clf_2.predict(X[:10, :])

print "Prediction is: ", pred # [0 1 0 0 0 1 0 0 1 0]

print "Accuracy Score is: ", accuracy_score(pred, y[:10, 0]) # now 0.8