def __init__(self):
os.environ['STANFORD_PARSER'] = './stanford-postagger-full-2014-08-27'
os.environ['CLASSPATH'] = './stanford-postagger-full-2014-08-27'
os.environ['STANFORD_MODELS'] = './stanford-postagger-full-2014-08-27'
self.pos = nltk.tag.stanford.StanfordPOSTagger('models/english-bidirectional-distsim.tagger')
self.inspirationSet = []
self.stopWords = get_stop_words('en')
self.ctr = syllableCounter.syllableCounter()
self.rhymeDictionary = RhymeDictionary()
self.rhymeDictionary.loadRhymeFiles()
self.robotBrain = RobotBrain()
self.nameList = nameList()
self.eval = Evaluate.Evaluator()
class Alvin:
def __init__(self):
os.environ['STANFORD_PARSER'] = './stanford-postagger-full-2014-08-27'
os.environ['CLASSPATH'] = './stanford-postagger-full-2014-08-27'
os.environ['STANFORD_MODELS'] = './stanford-postagger-full-2014-08-27'
self.pos = nltk.tag.stanford.StanfordPOSTagger('models/english-bidirectional-distsim.tagger')
self.inspirationSet = []
self.stopWords = get_stop_words('en')
self.ctr = syllableCounter.syllableCounter()
self.rhymeDictionary = RhymeDictionary()
self.rhymeDictionary.loadRhymeFiles()
self.robotBrain = RobotBrain()
self.nameList = nameList()
self.eval = Evaluate.Evaluator()
def get_cosine_similarity(self, word1, word2):
try:
return self.robotBrain.get_cosine_similarity(word1, word2)
except KeyError:
return -1
def LoadInspirationSet(self):
inspirationDirectory = "./InspirationSet/"
for file in os.listdir(inspirationDirectory):
with open(inspirationDirectory+file) as f:
for line in f:
out = line.translate(string.maketrans("",""), string.punctuation)
out = out.strip()
for word in out.split(" "):
if word:
self.inspirationSet.append(word)
def LoadOriginal(self):
inspirationFile = "./original.txt"
finalData = []
with open(inspirationFile) as f:
for line in f:
out = line.translate(string.maketrans("",""), string.punctuation)
out = out.strip()
finalData.append(out)
return finalData
def getTheme(self, data):
temp = [data[i].split() for i in range(len(data))]
temp = [temp[i][j] for i in range(len(temp)) for j in range(len(temp[i]))]
temp2 = []
for word in temp:
if self.isWordImportant(word) and word in self.robotBrain.model.vocab:
temp2.append(word)
indexes, sim = self.robotBrain.model.analogy(pos=temp2, neg=[])
theme = self.robotBrain.model.vocab[indexes[random.randint(0,9)]]
print(theme)
return theme
def transformTheme(self, theme):
numberOfThemes = 1
allwords = self.ctr.getAllWords()
themes = []
while(len(themes) < numberOfThemes):
try:
#perhaps make this a little more robust so we don't just keep getting errors forever
indexes, sim = self.robotBrain.model.analogy(pos=[allwords[random.randint(0, len(allwords)-1)]], neg=[theme])
for word in self.robotBrain.model.vocab[indexes]:
if self.isWordImportant(word) and self.robotBrain.get_popularity(word) > 125000:
themes.append(word)
break
except KeyError:
print("FINE: Error on theme")
return themes
def getRhymeScheme(self, data):
lastWords = [line.split()[len(line.split()) - 1] for line in data]
rhymeScheme = [i for i in range(len(lastWords))]
for i in range(len(lastWords)):
if rhymeScheme[i] == i:
for j in range(i+1, len(lastWords)):
if self.ctr.doWordsRhyme(lastWords[i], lastWords[j]):
rhymeScheme[j] = i
return rhymeScheme
def getMeter(self, data):
a = []
for word in data.split():
a.append(self.ctr.getEmphasisOf(word))
return a
def isWordImportant(self, data):
if data in self.stopWords:
return False
else:
return True
def evaluateLine(self, line):
print "Evaluating", line
words = line.split(' ')
currentOrder = 1
index = 0
while (index + currentOrder) < len(words):
phrase = ""
for i in xrange(index, currentOrder + index):
phrase = phrase + " " + words[i]
phrase = phrase.strip()
possibleWord = words[i+1]
#print phrase, possibleWord, currentOrder+1
nGramMap = NGrams.getProbabilities(phrase, [possibleWord], currentOrder+1) #returns an array of maps -- need to convert
nGramProbability = nGramMap[0]["probability"]
#print phrase, possibleWord, nGramProbability
if nGramProbability < -3.75:
return False
if currentOrder == 4:
index += 1
if currentOrder < 4:
currentOrder += 1
return True
def Evaluate(self, transformedText, rhymeScheme):
# first score each line according to how it conformed to its ryhme scheme
linesThatRhymed = 0
s = set(rhymeScheme)
print 1.0 * len(s) / len(transformedText)
for i, line in enumerate(transformedText):
text = line.split(' ')
# find the original line that this line needs to ryhme with
seed = rhymeScheme[i]
# this is the word we are testing to see if it ryhmes with the seed line
wordThatShouldRhyme = text[-1]
# seed word
seedWord = transformedText[seed].split(' ')[-1]
# if the word is the seed line, we count it as a rhymed line and move on to the next word
if i == seed:
linesThatRhymed += 1
continue
# we will also allow the words to be the same
if wordThatShouldRhyme == seedWord:
linesThatRhymed += 1
continue
# get the set of ryhmed words that ryhme with the original line
# eg
# 0 hi you
# 1 hello world
# if seed == 0, then rhymedWords = ['true', 'blue', ...]
rhymedWords = []
try:
rhymedWords = self.rhymeDictionary.getRhymes(seedWord)
except:
# the seed word was not found in the rhyme dictionary
continue
if wordThatShouldRhyme in rhymedWords:
linesThatRhymed += 1
def getNewLine(self, PoS, editedLine, transformedText, rhymeScheme, meter, newTheme, oldTheme, currentLineNumber): #magic happens
newLine = ""
allwords = self.ctr.getAllWords()
originalPoS = ["_"+PoS[i][1] for i in range(len(PoS))]
newLine = ""
i = 0
for word, part in zip(editedLine.split(), originalPoS):
newWord = ""
if word == "_":
if part == "_NNP":
part = "_NN"
elif part == "_NNPS":
part = "_NNS"
allwords = self.ctr.getWordsWithEmphasis(meter[i])
tempWords = [word for word in allwords if self.robotBrain.get_popularity(word) > 175000 and word not in self.nameList.names and self.robotBrain.get_most_likely_POS_tag(word) == part]
if len(tempWords) != 0:
allwords = tempWords
else:
print("failed", meter[i], part)
newLine = newLine + allwords[random.randint(0, len(allwords)-1)]
i = i + 1
continue
# if we are at the last word and the current line is not the first rhyming line in the series. ie [0,0,2,2] index != 1 || index != 3
if i == len(editedLine.split())-1 and rhymeScheme[currentLineNumber] != currentLineNumber:
# we retrieve the first line in the current ryhme series. ie [['hi', 'guys'],['burgers', 'fries'], ['spies', 'lies']], we would retrieve ['hi', 'guys']
transformedLineBefore = transformedText[rhymeScheme[currentLineNumber]]
# retrieve the last word from transformedLineBefore
wordToRhyme = transformedLineBefore.split()[len(transformedLineBefore.split())-1]
# retrieve all the rhyming words
rhymes = self.rhymeDictionary.getRhymes(wordToRhyme)
rhymes.append(wordToRhyme)
if len(set(allwords) & set(rhymes)) > 0:
# combine the rhyming words with the words that have the proper meter
rhymes = list(set(allwords) & set(rhymes))
# randomly select a word from this concatenated list; may not necessarily rhyme
similarity = []
for word in rhymes:
similarity.append(self.get_cosine_similarity(word, newTheme))
similarity = np.asarray(similarity)
if similarity.min() < 0:
similarity = similarity - similarity.min()
if sum(similarity) == 0:
similarity = np.ones(similarity.shape)
similarity = similarity / sum(similarity)
if newLine != "" or currentLineNumber == 0:
probabilities = NGrams.getNGramProbabilities(newLine, rhymes)
else:
probabilities = NGrams.getNGramProbabilities(transformedText[currentLineNumber-1], rhymes)
similarity = (similarity + 4 * probabilities) / 5.
index = np.argmax(np.random.multinomial(1, similarity))
while similarity[index] < 1./len(similarity):
index = np.argmax(np.random.multinomial(1, similarity))
newWord = rhymes[index]
elif i == len(editedLine.split()) - 1:
newWord = ""
possibleWords = [w for w in allwords if self.rhymeDictionary.wordList.has_key(w)]
probabilities = np.asarray(NGrams.getNGramProbabilities(newLine, possibleWords))
index = np.argmax(np.random.multinomial(1, probabilities))
while probabilities[index] < 1./len(probabilities):
index = np.argmax(np.random.multinomial(1, probabilities))
newWord = possibleWords[index]
else:
similarity = []
for word in allwords:
similarity.append(self.get_cosine_similarity(word, newTheme))
similarity = np.asarray(similarity)
if similarity.min() < 0:
similarity = similarity - similarity.min()
if sum(similarity) == 0:
similarity = np.ones(similarity.shape)
similarity = similarity / sum(similarity)
if newLine != "" or currentLineNumber == 0:
probabilities = NGrams.getNGramProbabilities(newLine, allwords)
else:
probabilities = NGrams.getNGramProbabilities(transformedText[currentLineNumber-1], allwords)
similarity = (similarity + 4 * probabilities) / 5.
index = np.argmax(np.random.multinomial(1, similarity))
while similarity[index] < 1./len(similarity):
index = np.argmax(np.random.multinomial(1, similarity))
newWord = allwords[index]
newLine = newLine + " " + newWord
else:
newLine = newLine + " " + word
i = i + 1
return newLine.strip()
def getPartOfSpeechTags(self, line):
return self.pos.tag(line.split())
def run(self):
self.LoadInspirationSet()
data = self.LoadOriginal()
theme = self.getTheme(data)
newTheme = self.transformTheme(theme)
rhymeScheme = self.getRhymeScheme(data)
for currentTheme in newTheme:
transformedText = []
for line, lineNumber in zip(data, range(len(data))):
wordNumber = 0
PoS = self.getPartOfSpeechTags(line)
meter = self.getMeter(line)
editedLine = ""
for word in line.split():
# if a stop word and not the last word of the line or is the last word of a line and is the seed rhyme word
if False and not self.isWordImportant(word) and (wordNumber != len(line.split()) - 1 or rhymeScheme[lineNumber] == lineNumber):
# we add the stop word or a word in the case that it is the last word of the line
editedLine = editedLine + " " + word
else:
# mark the space as needing to be replaced in the future
editedLine = editedLine + " _"
wordNumber = wordNumber + 1
evaluation = False
iteration = 0
while (not evaluation) and (iteration <= 10):
line = self.getNewLine(PoS, editedLine.strip(), transformedText, rhymeScheme, meter, currentTheme, theme, lineNumber)
evalution = self.eval.evaluateLine(line)
iteration += 1
print(iteration)
transformedText.append(line)
print(line)
print(currentTheme)
for line in transformedText:
print(line)
print("")
