1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
|
# Test for conditional models
# Ed Schofield, 2006
from numpy import *
from scipy.maxentropy import *
# Two contexts W, four labels x
# E_p f_0(W, X) = 0.4
# where f_0(w, x) = indicator func "is the label x=0 and is the context w=0?"
# So we want the distribution:
# x \ w 0 1
# 0 0.4 0.25
# 1 0.2 0.25
# 2 0.2 0.25
# 3 0.2 0.25
# We can achieve this by creating a feature matrix with one row per constraint,
# as follows:
F = array([[1, 0, 0, 0, 0, 0, 0, 0]])
# Each column represents one (w, x) pair. The number of columns is the product
# |w| * |x|, here 8. The order is (w0,x0), (w0,x1), (w0, x2), ..., (w1, x0),
# etc.
numcontexts = 2
numlabels = 4
# OLD:
# These can be in any order. The indices_context parameter to the
# conditionalmodel constructor records this order, so indices_context[0] is an
# array of indices all labels x in context w=0. The simplest ordering is:
# (w0, x0), (w0, x1), ..., (w0, x{n-1}), (w1, x0), ...
# in which case the parameter is:
# indices_context = array([[0, 1, 2, 3], [4, 5, 6, 7]])
# The counts of each (w, x) pair, estimated from a corpus or training dataset, is
# stored as an array with |w| * |x| elements in same order as before.
counts = array([4, 3, 2, 1, 4, 3, 2, 1])
# Note that, since the only active feature was for the first element (w0, x0),
# only the first value is relevant. The others are subject to no constraints,
# and will be chosen to maximize entropy.
model = conditionalmodel(F, counts, numcontexts)
model.verbose = True
model.fit()
# Do it again, since the CG algorithm gets stuck sometimes. Why is this??
model.fit()
# Note: to use the bound-constrained limited memory BFGS algorithm instead, we
# would use:
# model.fit(algorithm='LBFGSB')
# Display the fitted model
pmf = model.pmf()
# The elements of this are flatted like the rows of F and p_tilde. We display
# them nicely:
print "x \ w \t 0 \t 1",
for x in range(4):
print '\n' + str(x),
for w in range(2):
print ' \t %.3f' % pmf[w*numlabels + x],
# print ' \t %.3f' % pmf[indices_context[w]][x],
print
|
