Skip to content
Navigation Menu
{{ message }}
forked from aimacode/aima-python
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathtext.py
More file actions
383 lines (293 loc) · 13.6 KB
/
Copy pathtext.py
File metadata and controls
383 lines (293 loc) · 13.6 KB
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
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
"""Statistical Language Processing tools. (Chapter 22)
We define Unigram and Ngram text models, use them to generate random text,
and show the Viterbi algorithm for segmentatioon of letters into words.
Then we show a very simple Information Retrieval system, and an example
working on a tiny sample of Unix manual pages."""
from utils import * # noqa
from learning import CountingProbDist
import search
from math import log, exp
from collections import defaultdict
import re
class UnigramTextModel(CountingProbDist):
"""This is a discrete probability distribution over words, so you
can add, sample, or get P[word], just like with CountingProbDist. You can
also generate a random text n words long with P.samples(n)"""
def samples(self, n):
"Return a string of n words, random according to the model."
return ' '.join(self.sample() for i in range(n))
class NgramTextModel(CountingProbDist):
"""This is a discrete probability distribution over n-tuples of words.
You can add, sample or get P[(word1, ..., wordn)]. The method P.samples(n)
builds up an n-word sequence; P.add and P.add_sequence add data."""
def __init__(self, n, observation_sequence=[]):
# In addition to the dictionary of n-tuples, cond_prob is a
# mapping from (w1, ..., wn-1) to P(wn | w1, ... wn-1)
CountingProbDist.__init__(self)
self.n = n
self.cond_prob = defaultdict()
self.add_sequence(observation_sequence)
# __getitem__, top, sample inherited from CountingProbDist
# Note they deal with tuples, not strings, as inputs
def add(self, ngram):
"""Count 1 for P[(w1, ..., wn)] and for P(wn | (w1, ..., wn-1)"""
CountingProbDist.add(self, ngram)
if ngram[:-1] not in self.cond_prob:
self.cond_prob[ngram[:-1]] = CountingProbDist()
self.cond_prob[ngram[:-1]].add(ngram[-1])
def add_sequence(self, words):
"""Add each of the tuple words[i:i+n], using a sliding window.
Prefix some copies of the empty word, '', to make the start work."""
n = self.n
words = ['', ] * (n-1) + words
for i in range(len(words)-n):
self.add(tuple(words[i:i+n]))
def samples(self, nwords):
"""Build up a random sample of text nwords words long, using
the conditional probability given the n-1 preceding words."""
n = self.n
nminus1gram = ('',) * (n-1)
output = []
for i in range(nwords):
if nminus1gram not in self.cond_prob:
nminus1gram = ('',) * (n-1) # Cannot continue, so restart.
wn = self.cond_prob[nminus1gram].sample()
output.append(wn)
nminus1gram = nminus1gram[1:] + (wn,)
return ' '.join(output)
# ______________________________________________________________________________
def viterbi_segment(text, P):
"""Find the best segmentation of the string of characters, given the
UnigramTextModel P."""
# best[i] = best probability for text[0:i]
# words[i] = best word ending at position i
n = len(text)
words = [''] + list(text)
best = [1.0] + [0.0] * n
# Fill in the vectors best, words via dynamic programming
for i in range(n+1):
for j in range(0, i):
w = text[j:i]
if P[w] * best[i - len(w)] >= best[i]:
best[i] = P[w] * best[i - len(w)]
words[i] = w
# Now recover the sequence of best words
sequence = []
i = len(words)-1
while i > 0:
sequence[0:0] = [words[i]]
i = i - len(words[i])
# Return sequence of best words and overall probability
return sequence, best[-1]
# ______________________________________________________________________________
# TODO(tmrts): Expose raw index
class IRSystem:
"""A very simple Information Retrieval System, as discussed in Sect. 23.2.
The constructor s = IRSystem('the a') builds an empty system with two
stopwords. Next, index several documents with s.index_document(text, url).
Then ask queries with s.query('query words', n) to retrieve the top n
matching documents. Queries are literal words from the document,
except that stopwords are ignored, and there is one special syntax:
The query "learn: man cat", for example, runs "man cat" and indexes it."""
def __init__(self, stopwords='the a of'):
"""Create an IR System. Optionally specify stopwords."""
# index is a map of {word: {docid: count}}, where docid is an int,
# indicating the index into the documents list.
update(self, index=defaultdict(lambda: defaultdict(int)),
stopwords=set(words(stopwords)), documents=[])
def index_collection(self, filenames):
"Index a whole collection of files."
prefix = os.path.dirname(__file__)
for filename in filenames:
self.index_document(open(filename).read(),
os.path.relpath(filename, prefix))
def index_document(self, text, url):
"Index the text of a document."
# For now, use first line for title
title = text[:text.index('\n')].strip()
docwords = words(text)
docid = len(self.documents)
self.documents.append(Document(title, url, len(docwords)))
for word in docwords:
if word not in self.stopwords:
self.index[word][docid] += 1
def query(self, query_text, n=10):
"""Return a list of n (score, docid) pairs for the best matches.
Also handle the special syntax for 'learn: command'."""
if query_text.startswith("learn:"):
doctext = os.popen(query_text[len("learn:"):], 'r').read()
self.index_document(doctext, query_text)
return []
qwords = [w for w in words(query_text) if w not in self.stopwords]
shortest = argmin(qwords, lambda w: len(self.index[w]))
docs = self.index[shortest]
results = [(sum([self.score(w, d) for w in qwords]), d) for d in docs]
results.sort()
results.reverse()
return results[:n]
def score(self, word, docid):
"Compute a score for this word on this docid."
# There are many options; here we take a very simple approach
return (math.log(1 + self.index[word][docid]) /
math.log(1 + self.documents[docid].nwords))
def present(self, results):
"Present the results as a list."
for (score, d) in results:
doc = self.documents[d]
print(
("{:5.2}|{:25} | {}".format(100 * score, doc.url,
doc.title[:45].expandtabs())))
def present_results(self, query_text, n=10):
"Get results for the query and present them."
self.present(self.query(query_text, n))
class UnixConsultant(IRSystem):
"""A trivial IR system over a small collection of Unix man pages."""
def __init__(self):
IRSystem.__init__(self, stopwords="how do i the a of")
import os
aima_root = os.path.dirname(__file__)
mandir = os.path.join(aima_root, 'aima-data/MAN/')
man_files = [mandir + f for f in os.listdir(mandir)
if f.endswith('.txt')]
self.index_collection(man_files)
class Document:
"""Metadata for a document: title and url; maybe add others later."""
def __init__(self, title, url, nwords):
update(self, title=title, url=url, nwords=nwords)
def words(text, reg=re.compile('[a-z0-9]+')):
"""Return a list of the words in text, ignoring punctuation and
converting everything to lowercase (to canonicalize).
>>> words("``EGAD!'' Edgar cried.")
['egad', 'edgar', 'cried']
"""
return reg.findall(text.lower())
def canonicalize(text):
"""Return a canonical text: only lowercase letters and blanks.
>>> canonicalize("``EGAD!'' Edgar cried.")
'egad edgar cried'
"""
return ' '.join(words(text))
# ______________________________________________________________________________
# Example application (not in book): decode a cipher.
# A cipher is a code that substitutes one character for another.
# A shift cipher is a rotation of the letters in the alphabet,
# such as the famous rot13, which maps A to N, B to M, etc.
alphabet = 'abcdefghijklmnopqrstuvwxyz'
# Encoding
def shift_encode(plaintext, n):
"""Encode text with a shift cipher that moves each letter up by n letters.
>>> shift_encode('abc z', 1)
'bcd a'
"""
return encode(plaintext, alphabet[n:] + alphabet[:n])
def rot13(plaintext):
"""Encode text by rotating letters by 13 spaces in the alphabet.
>>> rot13('hello')
'uryyb'
>>> rot13(rot13('hello'))
'hello'
"""
return shift_encode(plaintext, 13)
def translate(plaintext, function):
"""Translate chars of a plaintext with the given function."""
result = ""
for char in plaintext:
result += function(char)
return result
def maketrans(from_, to_):
"""Create a translation table and return the proper function."""
trans_table = {}
for n, char in enumerate(from_):
trans_table[char] = to_[n]
return lambda char: trans_table.get(char, char)
def encode(plaintext, code):
"Encodes text, using a code which is a permutation of the alphabet."
trans = maketrans(alphabet + alphabet.upper(), code + code.upper())
return translate(plaintext, trans)
def bigrams(text):
"""Return a list of pairs in text (a sequence of letters or words).
>>> bigrams('this')
['th', 'hi', 'is']
>>> bigrams(['this', 'is', 'a', 'test'])
[['this', 'is'], ['is', 'a'], ['a', 'test']]
"""
return [text[i:i+2] for i in range(len(text) - 1)]
# Decoding a Shift (or Caesar) Cipher
class ShiftDecoder:
"""There are only 26 possible encodings, so we can try all of them,
and return the one with the highest probability, according to a
bigram probability distribution."""
def __init__(self, training_text):
training_text = canonicalize(training_text)
self.P2 = CountingProbDist(bigrams(training_text), default=1)
def score(self, plaintext):
"Return a score for text based on how common letters pairs are."
s = 1.0
for bi in bigrams(plaintext):
s = s * self.P2[bi]
return s
def decode(self, ciphertext):
"Return the shift decoding of text with the best score."
list_ = [(self.score(shift), shift)
for shift in all_shifts(ciphertext)]
return max(list_, key=lambda elm: elm[0])[1]
def all_shifts(text):
"Return a list of all 26 possible encodings of text by a shift cipher."
yield from (shift_encode(text, i) for i, _ in enumerate(alphabet))
# Decoding a General Permutation Cipher
class PermutationDecoder:
"""This is a much harder problem than the shift decoder. There are 26!
permutations, so we can't try them all. Instead we have to search.
We want to search well, but there are many things to consider:
Unigram probabilities (E is the most common letter); Bigram probabilities
(TH is the most common bigram); word probabilities (I and A are the most
common one-letter words, etc.); etc.
We could represent a search state as a permutation of the 26 letters,
and alter the solution through hill climbing. With an initial guess
based on unigram probabilities, this would probably fare well. However,
I chose instead to have an incremental representation. A state is
represented as a letter-to-letter map; for example {'z': 'e'} to
represent that 'z' will be translated to 'e'.
"""
def __init__(self, training_text, ciphertext=None):
self.Pwords = UnigramTextModel(words(training_text))
self.P1 = UnigramTextModel(training_text) # By letter
self.P2 = NgramTextModel(2, training_text) # By letter pair
def decode(self, ciphertext):
"Search for a decoding of the ciphertext."
self.ciphertext = ciphertext
problem = PermutationDecoderProblem(decoder=self)
return search.best_first_tree_search(
problem, lambda node: self.score(node.state))
def score(self, code):
"""Score is product of word scores, unigram scores, and bigram scores.
This can get very small, so we use logs and exp."""
text = permutation_decode(self.ciphertext, code)
logP = (sum([log(self.Pwords[word]) for word in words(text)]) +
sum([log(self.P1[c]) for c in text]) +
sum([log(self.P2[b]) for b in bigrams(text)]))
return exp(logP)
class PermutationDecoderProblem(search.Problem):
def __init__(self, initial=None, goal=None, decoder=None):
self.initial = initial or {}
self.decoder = decoder
def actions(self, state):
# Find the best
p, plainchar = max([(self.decoder.P1[c], c)
for c in alphabet if c not in state])
succs = [extend(state, plainchar, cipherchar)] # ???? # noqa
def goal_test(self, state):
"We're done when we get all 26 letters assigned."
return len(state) >= 26
# ______________________________________________________________________________
# TODO(tmrts): Set RNG seed to test random functions
__doc__ += """
Random tests:
## Generate random text from the N-gram models # noqa
>>> P1.samples(20)
'you thought known but were insides of see in depend by us dodecahedrons just but i words are instead degrees'
>>> P2.samples(20)
'flatland well then can anything else more into the total destruction and circles teach others confine women must be added'
>>> P3.samples(20)
'flatland by edwin a abbott 1884 to the wake of a certificate from nature herself proving the equal sided triangle'
"""
You can’t perform that action at this time.