def test(self, testing_data):
"""
Tests the precision/recall of the model
Parameters:
testing_data (array): data on which to test the model
Returns:
Null
"""
for row in testing_data:
u_priors = dict(self.priors)
tokenized_turn1 = tokenize(row["turn1"]["text"])
tokenized_turn2 = tokenize(row["turn2"]["text"])
tokenized_turn3 = tokenize(row["turn3"]["text"])
conv = tokenized_turn1 + tokenized_turn2 + tokenized_turn3
parsed_message = flatten([
ngrams_and_remove_stop_words(x, self.ngram_choice)
for x in [tokenized_turn1, tokenized_turn2, tokenized_turn3]
])
for var in self.variables:
classification = normalize(
self._classify(self.ngrams[var], parsed_message, conv,
u_priors[var], var))
for i, e in enumerate(self.variable_dimensions):
u_priors[var][e] = classification[i]
parsed_message = ngrams_and_remove_stop_words(
tokenized_turn1, self.ngram_choice)
for var in self.variables:
classification = normalize(
self._classify(self.ngrams[var], parsed_message,
tokenized_turn1, u_priors[var], var))
for i, e in enumerate(self.variable_dimensions):
u_priors[var][e] = classification[i]
parsed_message = ngrams_and_remove_stop_words(
tokenized_turn3, self.ngram_choice)
var_classification = {dim: "" for dim in self.variables}
for var in self.variables:
var_classification[var] = self._classify(
self.ngrams[var], parsed_message, tokenized_turn3,
u_priors[var], var, False)
self.true.append(row["turn3"]["emotion"])
emo_class = self.__map_to_emotion(var_classification)
self.pred.append(emo_class[0])
super(AVModel, self).calculate_scores()
def test(self, testing_data):
"""
Tests the precision/recall of the model
Parameters:
testing_data (array): data on which to test the model
Returns:
Null
"""
for var in self.variables:
self.true[var] = []
self.pred[var] = []
for row in testing_data:
u_priors = dict(self.priors)
tokenized_turn1 = tokenize(row["turn1"]["text"])
tokenized_turn2 = tokenize(row["turn2"]["text"])
tokenized_turn3 = tokenize(row["turn3"]["text"])
conv = tokenized_turn1 + tokenized_turn2 + tokenized_turn3
parsed_message = flatten([ngrams_and_remove_stop_words(x, self.ngram_choice) for x in [
tokenized_turn1, tokenized_turn2, tokenized_turn3]])
for var in self.variables:
classification = normalize(self.__classify(
self.ngrams[var], parsed_message, conv, u_priors[var], var))
for i, e in enumerate(self.variable_dimensions):
u_priors[var][e] = classification[i]
parsed_message = ngrams_and_remove_stop_words(
tokenized_turn1, self.ngram_choice)
for var in self.variables:
classification = normalize(self.__classify(
self.ngrams[var], parsed_message, tokenized_turn1, u_priors[var], var))
for i, e in enumerate(self.variable_dimensions):
u_priors[var][e] = classification[i]
parsed_message = ngrams_and_remove_stop_words(
tokenized_turn3, self.ngram_choice)
for var in self.variables:
weight = self.variable_dimensions[int(
row["turn3"]["appraisals"][var])]
self.true[var].append(weight)
classification = self.__classify(
self.ngrams[var], parsed_message, tokenized_turn3, u_priors[var], var, False)
self.pred[var].append(classification)
self.calculate_scores()
def test(self, testing_data):
"""
Tests the precision/recall of the model
Parameters:
testing_data (array): data on which to test the model
Returns:
Null
"""
for row in testing_data:
u_priors = dict(self.priors)
tokenized_turn1 = tokenize(row["turn1"]["text"])
tokenized_turn2 = tokenize(row["turn2"]["text"])
tokenized_turn3 = tokenize(row["turn3"]["text"])
conv = tokenized_turn1 + tokenized_turn2 + tokenized_turn3
parsed_message = flatten([
ngrams_and_remove_stop_words(x, self.ngram_choice)
for x in [tokenized_turn1, tokenized_turn2, tokenized_turn3]
])
classification = normalize(
self.__classify(self.ngrams, parsed_message, conv, u_priors))
for i, e in enumerate(self.emotions):
u_priors[e] = classification[i]
parsed_message = ngrams_and_remove_stop_words(
tokenized_turn1, self.ngram_choice)
classification = normalize(
self.__classify(self.ngrams, parsed_message, tokenized_turn1,
u_priors))
for i, e in enumerate(self.emotions):
u_priors[e] = classification[i]
emotion = row["turn3"]["emotion"]
self.true.append(emotion)
parsed_message = ngrams_and_remove_stop_words(
tokenized_turn3, self.ngram_choice)
classification = self.__classify(self.ngrams, parsed_message,
tokenized_turn3, u_priors, False)
self.pred.append(str(classification))
self.calculate_scores()
def _train_by_variable(self, training_set, variable, data_points={}):
"""
Calculates the counts for each unigram and priors for each classification
Parameters:
training_set (array): training data used to train the model
variable (string): variable in use in training
Returns:
Object: ngrams with associated counts
Object: sums for each classification
Object: priors for each classification
"""
words = {}
words_totals = {dim: 0 for dim in self.variable_dimensions}
tense_totals = {dim: 0 for dim in self.variable_dimensions}
pronoun_totals = {dim: 0 for dim in self.variable_dimensions}
words_vocab = set()
tense_vocab = set()
pronoun_vocab = set()
for row in training_set:
for turn in ["turn1", "turn2", "turn3"]:
true_dim = self.variable_dimensions[int(
row[turn]["appraisals"][variable])]
tokenized_res = tokenize(row[turn]["text"])
pos = parts_of_speech(tokenized_res)
tense_vocab, tense_totals, pronoun_vocab, pronoun_totals = self.__build_pos_counts(
pos, tense_vocab, variable, true_dim, tense_totals, pronoun_vocab, pronoun_totals)
res = ngrams_and_remove_stop_words(
tokenized_res, self.ngram_choice)
words_vocab, words, words_totals = self.__build_word_counts(
res, words_vocab, words, true_dim, words_totals)
denom = sum(words_totals.values())
self.priors[variable] = {dim: float(
words_totals[dim])/float(denom) for dim in self.variable_dimensions}
self.__calculate_probabilities(
words, words_totals, words_vocab, tense_totals, tense_vocab, pronoun_totals, pronoun_vocab, variable)
def train(self, training_data):
"""
Builds a trained Emotions model
Parameters:
training_data (array): training data used to train the model
Returns:
Model: a trained model
"""
words = {}
words_vocab = set()
tense_vocab = set()
pronoun_vocab = set()
words_totals = {emotion: 0 for emotion in self.emotions}
tense_totals = {emotion: 0 for emotion in self.emotions}
pronoun_totals = {emotion: 0 for emotion in self.emotions}
for row in training_data:
for turn in ["turn1", "turn2", "turn3"]:
true_emotion = row[turn]["emotion"]
tokenized_res = tokenize(row[turn]["text"])
pos = parts_of_speech(tokenized_res)
tense_vocab, tense_totals, pronoun_vocab, pronoun_totals = self.__build_pos_counts(
pos, tense_vocab, true_emotion, tense_totals,
pronoun_vocab, pronoun_totals)
res = ngrams_and_remove_stop_words(tokenized_res,
self.ngram_choice)
words_vocab, words, words_totals = self.__build_word_counts(
res, words_vocab, words, true_emotion, words_totals)
sum_totals = sum(words_totals.values())
self.priors = {
emotion: float(words_totals[emotion]) / float(sum_totals)
for emotion in self.emotions
}
self.__calculate_probabilities(words, words_totals, words_vocab,
tense_totals, tense_vocab,
pronoun_totals, pronoun_vocab)