def build(self, text=None):
"""
Take text (for example as returned by corpus() ) and build a trie. The
maximum n-gram length is self.max_chars characters or self.max_words
words, whichever is shorter.
"""###
if text == None: text = self.read_corpus()
t = TreeStructure.trie()
istop = len(text)
pr = progress(istop, 'building language model')
for i in range(istop):
substr = ''
j = i
nwords = 0
lastchar = ' '
while j < istop:
char = text[j]
substr += char
endword = (char == ' ' and lastchar != ' ')
lastchar = char
j += 1
if endword:
nwords += 1
if len(substr
) >= self.max_chars or nwords >= self.max_words:
break
#print '"'+substr+'"'
t.add(substr)
if i % 5000 == 0: pr.update(i, '(%d nodes)' % len(t.nodes))
self.prefixtree = t
self.corpus = text
self.timestamp = time.localtime()[:6]
return self
def clean(self, text):
text = self.match_case(text)
if '\n' not in self.alphabet: text = text.replace('\n', ' ')
text = ['. '] + list(text) + [' ']
pr = progress(len(text), 'cleaning')
for i in range(len(text)):
if i > 0: prevx = text[i - 1]
else: prevx = ''
if i + 1 < len(text): nextx = text[i + 1]
else: nextx = ''
x = text[i]
if not x in self.alphabet: x = self.translations.get(x, '')
if self.space_delimited:
if x in '.,?!;:' and nextx == ' ': x = ' ' + x
elif x in '"()[]{}' and prevx == ' ': x = x + ' '
# In space_delimited mode, space now unambiguously denotes the end of a word. Ending the sentence
# is a separate issue (only likely to happen after the end of a word). So effectively, each punctuation
# concept ("sentence ends now", "sentence ends now as a question", etc...) has been made a "word". For
# now, the default is *not* to do things in space_delimited mode, but in future it would be desirable.
# However, TODO: how to undo this at the prediction stage??
if prevx.endswith(' ') or prevx.endswith('\n'):
x = x.lstrip(' ') # collapse multiple spaces
text[i] = x
if i % 5000 == 0: pr.update(i)
pr.done()
text = ''.join(text)
return text
def build(self, text=None): """ Take text (for example as returned by corpus() ) and build a trie. The maximum n-gram length is self.max_chars characters or self.max_words words, whichever is shorter. """### if text == None: text = self.read_corpus() t = TreeStructure.trie() istop = len(text) pr = progress(istop, 'building language model') for i in range(istop): substr = ''; j = i; nwords = 0; lastchar = ' ' while j<istop: char = text[j] substr += char endword = (char == ' ' and lastchar != ' ') lastchar = char j += 1 if endword: nwords += 1 if len(substr) >= self.max_chars or nwords >= self.max_words: break #print '"'+substr+'"' t.add(substr) if i%5000 == 0: pr.update(i, '(%d nodes)'%len(t.nodes)) self.prefixtree = t self.corpus = text self.timestamp = time.localtime()[:6] return self
def clean(self, text):
text = self.match_case(text)
if '\n' not in self.alphabet: text = text.replace('\n', ' ')
text = ['. '] + list(text) + [' ']
pr = progress(len(text), 'cleaning')
for i in range(len(text)):
if i > 0: prevx = text[i-1]
else: prevx = ''
if i+1 < len(text): nextx = text[i+1]
else: nextx = ''
x = text[i]
if not x in self.alphabet: x = self.translations.get(x,'')
if self.space_delimited:
if x in '.,?!;:' and nextx == ' ': x = ' ' + x
elif x in '"()[]{}' and prevx == ' ': x = x + ' '
# In space_delimited mode, space now unambiguously denotes the end of a word. Ending the sentence
# is a separate issue (only likely to happen after the end of a word). So effectively, each punctuation
# concept ("sentence ends now", "sentence ends now as a question", etc...) has been made a "word". For
# now, the default is *not* to do things in space_delimited mode, but in future it would be desirable.
# However, TODO: how to undo this at the prediction stage??
if prevx.endswith(' ') or prevx.endswith('\n'): x = x.lstrip(' ') # collapse multiple spaces
text[i] = x
if i%5000 == 0: pr.update(i)
pr.done()
text = ''.join(text)
return text
def packtrie(t): """ Helper function for trie.pack() """### if t.ispacked(): raise TrieError, 'already packed' x = '' lut = [0] for node in t.nodes[:-1]: lut.append(lut[-1]+node.packed_size()) if progress != None: pr = progress(len(t.nodes), ' packing') for i in range(len(t.nodes)): x += packnode(t.nodes[i], lut=lut) if i%5000 == 0 and progress != None: pr.update(i, '(%d nodes)'%i) if progress != None: pr.done() return x
def packtrie(t):
"""
Helper function for trie.pack()
"""
###
if t.ispacked(): raise TrieError, 'already packed'
x = ''
lut = [0]
for node in t.nodes[:-1]:
lut.append(lut[-1] + node.packed_size())
if progress != None: pr = progress(len(t.nodes), ' packing')
for i in range(len(t.nodes)):
x += packnode(t.nodes[i], lut=lut)
if i % 5000 == 0 and progress != None: pr.update(i, '(%d nodes)' % i)
if progress != None: pr.done()
return x
def unpacktrie(x):
"""
Helper function for trie.unpack()
"""###
t = trie()
t.nodes = []
ind = 0
rlut = {}
if progress != None: pr = progress(len(x), 'unpacking')
while ind < len(x):
rlut[ind] = len(t.nodes)
node,ind = unpacknode(x,ind)
t.nodes.append(node)
nnodes = len(t.nodes)
if nnodes % 5000 == 0 and progress != None: pr.update(ind, '(%d nodes)'%nnodes)
for node in t.nodes:
node.parent = rlut[node.parent]
for key,val in node.children.items(): node.children[key] = rlut[val]
return t
def unpacktrie(x):
"""
Helper function for trie.unpack()
"""
###
t = trie()
t.nodes = []
ind = 0
rlut = {}
if progress != None: pr = progress(len(x), 'unpacking')
while ind < len(x):
rlut[ind] = len(t.nodes)
node, ind = unpacknode(x, ind)
t.nodes.append(node)
nnodes = len(t.nodes)
if nnodes % 5000 == 0 and progress != None:
pr.update(ind, '(%d nodes)' % nnodes)
for node in t.nodes:
node.parent = rlut[node.parent]
for key, val in node.children.items():
node.children[key] = rlut[val]
return t
