%matplotlib inline
!ls divine_comedy
divinacommedia_cleaned.txt divina_commedia_with_copyright_notice.txt
text_file = "./divine_comedy/divinacommedia_cleaned.txt"
# can I read the data
with open(text_file, 'r', encoding='utf8') as infile:
for line in infile:
pass
how many couples do I need?
I have 30 characters possible (26 letters + space, newline, full stop, exclamation point)
# first order chain
30**2
900
# second order chain
30**3
27000
# third order chain
30**4
810000
how many letters do we have in our corpus?
letters = 0
with open(text_file, 'r', encoding='utf8') as infile:
for line in infile:
for letter in line:
letters += 1
print(letters)
529771
# how many and which letters do I have in my text?
from collections import Counter
observed = Counter()
with open(text_file, 'r', encoding='utf8') as infile:
for line in infile:
for letter in line:
observed[letter] += 1
observed
Counter({'\ufeff': 1,
'N': 258,
'e': 46094,
'l': 23087,
' ': 83019,
'm': 11681,
'z': 1848,
'o': 37254,
'd': 14599,
'c': 20267,
'a': 42035,
'i': 39200,
'n': 26229,
's': 22188,
't': 22478,
'r': 25805,
'v': 7946,
'\n': 19053,
'p': 10790,
'u': 13408,
',': 8513,
'h': 7109,
'é': 903,
'.': 3275,
'A': 377,
'q': 3025,
'è': 925,
'g': 7121,
'f': 4947,
'!': 232,
'T': 283,
'’': 7623,
'ù': 1080,
';': 1628,
'b': 2758,
'ò': 938,
'I': 359,
'M': 405,
'à': 855,
'E': 586,
'ì': 1383,
'P': 473,
'Q': 328,
'«': 1062,
'»': 1062,
'R': 117,
':': 988,
'ï': 427,
'ó': 30,
'?': 278,
'O': 356,
'V': 222,
'D': 407,
'C': 554,
'L': 411,
'S': 443,
'ë': 83,
'“': 51,
'B': 223,
'”': 51,
'—': 18,
'‘': 109,
'G': 199,
'F': 180,
'Z': 7,
'ü': 55,
'U': 43,
'H': 2,
'ö': 1,
'ä': 8,
'x': 3,
'y': 1,
'(': 3,
')': 3,
'È': 2,
'j': 2,
'Ë': 2,
'Ï': 1,
'-': 1})
# how many and which letters do I have in my text?
# try with a different encoding
from collections import Counter
observed = Counter()
with open(text_file, 'r', encoding='utf-8-sig') as infile:
for line in infile:
for letter in line:
observed[letter] += 1
observed
Counter({'N': 258,
'e': 46094,
'l': 23087,
' ': 83019,
'm': 11681,
'z': 1848,
'o': 37254,
'd': 14599,
'c': 20267,
'a': 42035,
'i': 39200,
'n': 26229,
's': 22188,
't': 22478,
'r': 25805,
'v': 7946,
'\n': 19053,
'p': 10790,
'u': 13408,
',': 8513,
'h': 7109,
'é': 903,
'.': 3275,
'A': 377,
'q': 3025,
'è': 925,
'g': 7121,
'f': 4947,
'!': 232,
'T': 283,
'’': 7623,
'ù': 1080,
';': 1628,
'b': 2758,
'ò': 938,
'I': 359,
'M': 405,
'à': 855,
'E': 586,
'ì': 1383,
'P': 473,
'Q': 328,
'«': 1062,
'»': 1062,
'R': 117,
':': 988,
'ï': 427,
'ó': 30,
'?': 278,
'O': 356,
'V': 222,
'D': 407,
'C': 554,
'L': 411,
'S': 443,
'ë': 83,
'“': 51,
'B': 223,
'”': 51,
'—': 18,
'‘': 109,
'G': 199,
'F': 180,
'Z': 7,
'ü': 55,
'U': 43,
'H': 2,
'ö': 1,
'ä': 8,
'x': 3,
'y': 1,
'(': 3,
')': 3,
'È': 2,
'j': 2,
'Ë': 2,
'Ï': 1,
'-': 1})
# we decide to normalize the uncommon letters
to_replace = {'Ë': 'E', 'Ï': 'I',
'ö': 'o', 'ä': 'a',
'ü': 'u', 'ë': 'e',
'ï': 'i'}
def letter_normalization_naive(letter):
if letter in to_replace:
return to_replace[letter]
return letter
def letter_normalization_short(letter):
return to_replace.get(letter, letter)
letter_normalization_naive('Ë')
'E'
letter_normalization_short('Ë')
'E'
%timeit letter_normalization_naive('Ë')
116 ns ± 2.36 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)
%timeit letter_normalization_short('Ë')
158 ns ± 2.01 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)
letter_normalization = letter_normalization_naive
from collections import Counter
observed = Counter()
with open(text_file, 'r', encoding='utf-8-sig') as infile:
for line in infile:
for letter in line:
modified_letter = letter_normalization(letter)
observed[modified_letter] += 1
observed
Counter({'N': 258,
'e': 46177,
'l': 23087,
' ': 83019,
'm': 11681,
'z': 1848,
'o': 37255,
'd': 14599,
'c': 20267,
'a': 42043,
'i': 39627,
'n': 26229,
's': 22188,
't': 22478,
'r': 25805,
'v': 7946,
'\n': 19053,
'p': 10790,
'u': 13463,
',': 8513,
'h': 7109,
'é': 903,
'.': 3275,
'A': 377,
'q': 3025,
'è': 925,
'g': 7121,
'f': 4947,
'!': 232,
'T': 283,
'’': 7623,
'ù': 1080,
';': 1628,
'b': 2758,
'ò': 938,
'I': 360,
'M': 405,
'à': 855,
'E': 588,
'ì': 1383,
'P': 473,
'Q': 328,
'«': 1062,
'»': 1062,
'R': 117,
':': 988,
'ó': 30,
'?': 278,
'O': 356,
'V': 222,
'D': 407,
'C': 554,
'L': 411,
'S': 443,
'“': 51,
'B': 223,
'”': 51,
'—': 18,
'‘': 109,
'G': 199,
'F': 180,
'Z': 7,
'U': 43,
'H': 2,
'x': 3,
'y': 1,
'(': 3,
')': 3,
'È': 2,
'j': 2,
'-': 1})
from random import choices
letters = list(observed.keys())
occurrences = list(observed.values())
generated = choices(letters, occurrences, k=20)
collated = "".join(generated)
print(collated)
a ,v suì trdt r i ac
we need to take our letters in couples
if we have a text such "home" we want to obtain the couples:
(h, o) (o, m) (m, e)
list_1 = ['a', 'b', 'c']
list_2 = [1, 2, 3]
list(zip(list_1, list_2))
[('a', 1), ('b', 2), ('c', 3)]
text = "home"
list(zip(text, text[1:]))
[('h', 'o'), ('o', 'm'), ('m', 'e')]
def couples_from_seq(seq):
"""`seq` is a list of characters"""
return zip(seq, seq[1:])
list(couples_from_seq('home'))
[('h', 'o'), ('o', 'm'), ('m', 'e')]
with open(text_file, 'r', encoding='utf-8-sig') as infile:
whole_text = "".join(line for line in infile)
print(whole_text[:120])
Nel mezzo del cammin di nostra vita mi ritrovai per una selva oscura, ché la diritta via era smarrita. Ahi quanto a dir
all_couples = list(couples_from_seq(whole_text))
all_couples[:10]
[('N', 'e'),
('e', 'l'),
('l', ' '),
(' ', 'm'),
('m', 'e'),
('e', 'z'),
('z', 'z'),
('z', 'o'),
('o', ' '),
(' ', 'd')]
from collections import defaultdict
foo = defaultdict(Counter)
foo['b']['a'] += 1
foo
defaultdict(collections.Counter, {'b': Counter({'a': 1})})
foo['b']
Counter({'a': 1})
# defaultdict allow me to avoid this check:
if 'c' not in foo:
foo['c'] = Counter()
# and just leave this command
foo['c'] ['a'] += 1
from collections import defaultdict
# Counter ~= defaultdict(int)
couples_counter = defaultdict(Counter)
for first_letter, second_letter in all_couples:
counter_of_first_letter = couples_counter[first_letter]
counter_of_first_letter[second_letter] += 1
# too big for visualization?
# couples_counter
sum(sum(counts.values()) for counts in couples_counter.values())
529769
letter = 'N'
letter_counts = couples_counter[letter]
possible_letters = list(letter_counts.keys())
counts = list(letter_counts.values())
next_letter = choices(possible_letters, counts)[0]
next_letter
'o'
text = ['N']
for i in range(200):
last_letter = text[-1]
letter_counts = couples_counter[last_letter]
possible_letters = list(letter_counts.keys())
counts = list(letter_counts.values())
next_letter = choices(possible_letters, counts)[0]
text.append(next_letter)
print("".join(text))
No ial’i: co veterondonn sesa is’a pa norarni nce draiere piolì che qu fomoronte: ppe ’ fuià dè e pi, atiagi vi s’lli ’ cit’ ch’ fe coruto figua sso pienco v’a be ge Ver eropuan de, Qu qumer ce so, so
write a code that generates few hundred letters using arbitrary long markov chains, such as triples and quadruplets, and discuss the resulting text
text = "hello world"
[('h', 'e', 'l'), ('e', 'l', 'l'), ...]def groups_from_seq(seq, order=2):
"""`seq` is a list of characters
horrible version number 1
"""
if order==2:
return zip(seq[0:], seq[1:])
elif order==3:
return zip(seq[0:], seq[1:], seq[2:])
elif order==4:
return zip(seq[0:], seq[1:], seq[2:], seq[3:])
else:
raise ValueError("only order 2 or 3 accepted")
list(groups_from_seq('home sweet home', 3))
[('h', 'o', 'm'),
('o', 'm', 'e'),
('m', 'e', ' '),
('e', ' ', 's'),
(' ', 's', 'w'),
('s', 'w', 'e'),
('w', 'e', 'e'),
('e', 'e', 't'),
('e', 't', ' '),
('t', ' ', 'h'),
(' ', 'h', 'o'),
('h', 'o', 'm'),
('o', 'm', 'e')]
def groups_from_seq(seq, order=2):
"""`seq` is a list of characters
appropriate for short sequences
"""
truncated_lists = [seq[i:] for i in range(order)]
return zip(*truncated_lists)
list(groups_from_seq('home sw', 3))
[('h', 'o', 'm'),
('o', 'm', 'e'),
('m', 'e', ' '),
('e', ' ', 's'),
(' ', 's', 'w')]
*previous_letters, last_letter = "home"
print(previous_letters, last_letter)
['h', 'o', 'm'] e
*previous_letters, last_letter = "h"
print(previous_letters, last_letter)
[] h
from collections import defaultdict, Counter
# Counter ~= defaultdict(int)
all_groups = list(groups_from_seq('home sweet home', 2))
groups_counter = defaultdict(Counter)
for *previous_letters, last_letter in all_groups:
previous_letters = tuple(previous_letters)
counter_of_previous_letters = groups_counter[previous_letters]
counter_of_previous_letters[last_letter] += 1
groups_counter
defaultdict(collections.Counter,
{('h',): Counter({'o': 2}),
('o',): Counter({'m': 2}),
('m',): Counter({'e': 2}),
('e',): Counter({' ': 1, 'e': 1, 't': 1}),
(' ',): Counter({'s': 1, 'h': 1}),
('s',): Counter({'w': 1}),
('w',): Counter({'e': 1}),
('t',): Counter({' ': 1})})
text = ['N', 'e', 'l', ' ', 'm', 'e']
print(text[-1])
print(text[-1:])
print(text[-2:])
e ['e'] ['m', 'e']
from random import choices
text = ['h', 'o']
order = 2
for i in range(50):
last_letters = tuple(text[-order+1:])
letter_counts = groups_counter[last_letters]
possible_letters = list(letter_counts.keys())
counts = list(letter_counts.values())
next_letter = choices(possible_letters, counts)[0]
text.append(next_letter)
print("".join(text))
homee homet home home home swe swee swet homet homee
let's transform these pieces of code in functions
def counts_from_groups(group_seq):
groups_counter = defaultdict(Counter)
for *previous_letters, last_letter in group_seq:
previous_letters = tuple(previous_letters)
count_of_previous_letters = groups_counter[previous_letters]
count_of_previous_letters[last_letter] += 1
return groups_counter
def generate_text(groups_counter, text_seed, order=2, lenght=200):
text = list(text_seed)
for i in range(lenght):
last_letters = tuple(text[-order+1:])
letter_counts = groups_counter[last_letters]
possible_letters = list(letter_counts.keys())
counts = list(letter_counts.values())
next_letter = choices(possible_letters, counts)[0]
text.append(next_letter)
return "".join(text)
source_text = "home sweet home"
order = 3
all_groups = groups_from_seq(source_text, order)
groups_counter = counts_from_groups(all_groups)
output_text = generate_text(groups_counter, "home", order=order)
print(output_text)
home sweet home sweet home sweet home sweet home sweet
with open(text_file, 'r', encoding='utf-8-sig') as infile:
source_text = "".join(line for line in infile)
order = 5
all_groups = groups_from_seq(source_text, order)
groups_counter = counts_from_groups(all_groups)
seed_text = "Nel mezzo"
output_text = generate_text(groups_counter, seed_text,
order=order, lenght=200)
print(output_text)
Nel mezzo ’l qual era forse che fossa farai che passi aspetto in là ’ve s’interno Augusto perché seguire a vendue oscura cornuta ne fretta». «Né creatura ch’uscì de la dier suo ne rito a cotesta: esce disio
whole_text = load_source()
generator = TextGenerator(order=3, split='letters')
generator.learn(whole_text)
generator.generate(seed='Nel', lenght=400)
class TextGenerator:
def __init__(self, order):
self.order = order
self.data = defaultdict(Counter)
def learn(self, text):
all_groups = groups_from_seq(text, self.order)
self.data = counts_from_groups(all_groups)
def generate(self, seed, lenght):
output_text = generate_text(self.data,
seed,
order=self.order,
lenght=lenght)
return output_text
generator = TextGenerator(order=3)
generator.learn(source_text)
output_text = generator.generate(seed='Nel', lenght=400)
print(output_text)
Nelo inò morselliuntor ch’enimanta nonve quellor figliento, e ri nongualtrissa, sì chietta, cesti cogli etè lo solucello so, con pre se vid’ultra. Ellassanna, però ci pornai: Ma vedispedi a in or fu visello e vel salzio lumi co’ el pre tarra fa, offal be ’ngor dissaliand’ iradavolsemperà; s’erte lorgo, e s’ioterso giun non sandi ’l ser dimalte. luma la ché s’afincommio tar se fu che distrà diesti
class NotLearnedError(Exception):
pass
class TextGenerator:
def __init__(self, order):
self.order = order
self.data = defaultdict(Counter)
# have I ever alled the learn function
self._learned = False
def learn(self, text):
all_groups = groups_from_seq(text, self.order)
self.data = counts_from_groups(all_groups)
# we have learned something
self._learned = True
# fluent interface
return self
def generate(self, seed, lenght):
"""
raises:
-------
`NotLearnedError` if the `self.learn` function
has not been called before
"""
# check that we performed the learn function
if not self._learned:
raise NotLearnedError()
output_text = generate_text(self.data,
seed,
order=self.order,
lenght=lenght)
return output_text
output_text = (TextGenerator(order=3)
.learn(source_text)
.generate(seed='Nel', lenght=400))
print(output_text)
Nel foste più dessen diserga pice, e qualsi pinfacciai e anivivintente si cannece ’nciei». Ma di a v’ amon que mi lia l’affio mia verratto ve for, sce lacella gricesir sù alle dicernonnellanda ti Balietro piedo cal me la lonoi ma re piati a sa che que: «Tutto. E queste che pre. Lo ch’aspo ve imone sa, po e pare ne, ch’ il fatericun lucestra. I’ enti, da. «Vatrovia però vinsibra di?». Qui obbito
output_text = (TextGenerator(order=3)
.generate(seed='Nel', lenght=400))
print(output_text)
--------------------------------------------------------------------------- NotLearnedError Traceback (most recent call last) <ipython-input-55-6d3227991905> in <module> 1 output_text = (TextGenerator(order=3) ----> 2 .generate(seed='Nel', lenght=400)) 3 print(output_text) <ipython-input-53-314a6658a7c3> in generate(self, seed, lenght) 20 # check that we performed the learn function 21 if not self._learned: ---> 22 raise NotLearnedError() 23 output_text = generate_text(self.data, 24 seed, NotLearnedError:
text_1 = load_divina_commedia()
text_2 = load_decameron()
generator = TextGenerator(order=3)
generator.learn(text_1)
generator.learn(text_2)
output_text = generator.generate(seed='Nel', lenght=400)
print(output_text)
class NotLearnedError(Exception):
pass
class TextGenerator:
def __init__(self, order):
self.order = order
self.data = defaultdict(Counter)
# have I ever alled the learn function
self._learned = False
def _counts_from_groups(self, group_seq):
for *previous_letters, last_letter in group_seq:
previous_letters = tuple(previous_letters)
count_previous = self.data[previous_letters]
count_previous[last_letter] += 1
def learn(self, text):
all_groups = groups_from_seq(text, self.order)
self._counts_from_groups(all_groups)
# we have learned something
self._learned = True
# fluent interface
return self
def generate(self, seed, lenght):
"""
raises:
-------
`NotLearnedError` if the `self.learn` function
has not been called before
"""
# check that we performed the learn function
if not self._learned:
raise NotLearnedError()
output_text = generate_text(self.data,
seed,
order=self.order,
lenght=lenght)
return output_text
source = "home"
list(groups_from_seq(source, 2))
[('h', 'o'), ('o', 'm'), ('m', 'e')]
source = ["home", "sweet", "home", "nice", "home"]
list(groups_from_seq(source, 2))
[('home', 'sweet'), ('sweet', 'home'), ('home', 'nice'), ('nice', 'home')]
class NotLearnedError(Exception):
pass
class TextGenerator:
def __init__(self, order, split='letter'):
self.order = order
self.data = defaultdict(Counter)
self.split = split
# have I ever alled the learn function
self._learned = False
def _counts_from_groups(self, group_seq):
for *previous_letters, last_letter in group_seq:
previous_letters = tuple(previous_letters)
count_previous = self.data[previous_letters]
count_previous[last_letter] += 1
def _generate_text_tokens(self, text):
if self.split == 'letter':
return list(text)
elif self.split == 'word': # TODO: how to manage commas
return text.split(' ')
else:
raise ValueError("splitting method not known")
def learn(self, text):
tokens = self._generate_text_tokens(text)
all_groups = groups_from_seq(tokens, self.order)
self._counts_from_groups(all_groups)
# we have learned something
self._learned = True
# fluent interface
return self
def _generate_text(self, text_seed, lenght=200):
groups_counter = self.data
order = self.order
text = list(text_seed)
for i in range(lenght):
last_letters = tuple(text[-order+1:])
letter_counts = groups_counter[last_letters]
possible_letters = list(letter_counts.keys())
counts = list(letter_counts.values())
next_letter = choices(possible_letters, counts)[0]
text.append(next_letter)
return text
def _collate_text(self, text_seq):
if self.split == 'letter':
return "".join(text_seq)
elif self.split == 'word': # TODO: how to manage commas
return " ".join(text_seq)
else:
raise ValueError("splitting method not known")
def generate(self, seed, lenght):
"""
raises:
-------
`NotLearnedError` if the `self.learn` function
has not been called before
"""
# check that we performed the learn function
if not self._learned:
raise NotLearnedError()
output_seq = self._generate_text(seed, lenght=lenght)
output_text = self._collate_text(output_seq)
return output_text
output_text = (TextGenerator(order=2, split='word')
.learn(source_text)
.generate(seed='Nel', lenght=50))
print(output_text)
N e l ’altro polo, là onde ogne cor mi si paia che tu se’ in pianta silvestra: l’Arpie, pascendo poi che vuo’ veder, li occhi suoi nemici non ne rimarita, come se’ tu riguardi là giù ponesse mente al fondamento che la vista al piè de le guance lagrimose; ivi mi fec’ io contemplo, adora per tua pur su l’erbetta