class NaiveBayes: def __init__(self, feature, vocab_size, positive_y, negative_y, positive_x, negative_x): self.parser = Parser() self.feature = feature # string... (ex) 'k1' 'w1' self.vocab_size = vocab_size self.total = float(len(positive_y) + len(negative_y)) self.positive_y = float(len(positive_y)) # (Y = +1) self.negative_y = float(len(negative_y)) # (Y = -1) self.positive_x = positive_x # (X = xi | Y = +1) self.negative_x = negative_x # (X = xi | Y = -1) self.positive_sum = float(sum(positive_x.values())) self.negative_sum = float(sum(negative_x.values())) def classify(self, example): e = example if type(example) is str: e = self.parser.stem_sentence_porter(example) elif type(example) is list: pass else: print "example should be of type str or list of str" positive_score = self.score(e, True) negative_score = self.score(e, False) if positive_score > negative_score: return 1.0 return -1.0 # return positive_score # result = -1.0*(positive_score - negative_score)#/ (abs(positive_score) + abs(negative_score)) # if result > 0: # return math.log10(result) # return -1.0 def score(self, example, positive): py, px, tw = None, None, None if positive: py = self.positive_y / self.total px = self.positive_x tw = self.positive_sum else: py = self.negative_y / self.total px = self.negative_x tw = self.negative_sum # s = math.log10(py) s = py for e in example: if e in px: s *= float(px[e])/float(tw) # s += math.log10(float(px[e])/float(tw)) # s += math.log10(float(1+px[e]) / float(self.vocab_size+tw)) return s
class Node: def __init__(self, left=None, right=None, criterion=None, label=None, depth=None): # in our case, it is on whether the document has a word -- a string self.criterion = criterion self.left = left self.right = right self.label = label self.depth = depth self.parser = None def get_label(self, tweet, tweet_cleaned=False): if self.parser is None: self.parser = Parser() if not tweet_cleaned: tweet = self.parser.stem_sentence_porter(tweet) tweet_cleaned = True if self.criterion: if self.criterion in tweet: # if has, go right, else left return self.right.get_label(tweet, tweet_cleaned) else: return self.left.get_label(tweet, tweet_cleaned) return self.label
class SVM:
def __init__(self):
self.parser = Parser()
self.weather_models = []
self.time_models = []
self.is_weather_model = None
self.default_data_features = []
self.data = None
self.index = None
self.index_map = None
self.threshold = 0.7
self.weather_labels = ["clouds", "cold", "dry", "hot", "humid", "hurricane",
"I can't tell", "ice", "other", "rain", "snow", "storms",
"sun", "tornado", "wind"]
def initialize_svm(self):
# get file path, depending on the location from which the class is called
cwd = os.getcwd()
cwd = cwd.split('/')
if cwd[len(cwd)-1] == 'src':
index_file_path = '../data/svm/data.index'
map_file_path = '../data/svm/data.map'
models_file_path = '../data/svm/models/'
else:
index_file_path = 'data/svm/data.index'
map_file_path = 'data/svm/data.map'
models_file_path = 'data/svm/models/'
self.load_all_models(models_file_path)
if self.index is None:
index = self.parser.load_pickled_data(index_file_path)
index_map = self.parser.load_pickled_data(map_file_path)
self.index = index
self.index_map = index_map
def load_all_models(self, path):
filepath = path + 's5.model0.01'
model = self.read_model(filepath)
self.is_weather_model = model
for i in range(4):
filepath = path + 'new_c_w{}.model1'.format(i+1)
model = self.read_model(filepath)
self.time_models.append(model)
for i in range(15):
# filepath = path + 'new_c_k{}.model0.1'.format(i+1)
filepath = path + 'k{}.model0.1'.format(i+1)
model = self.read_model(filepath)
self.weather_models.append(model)
def load_data(self, rel_path):
'''
Loads data from a SVMLight file using the svmlight_loader
library: https://github.com/mblondel/svmlight-loader
Returns a list of the dataset and the labels
'''
abs_path = os.path.abspath(rel_path)
(x_train, labels) = svml.load_svmlight_file(abs_path)
return [x_train, labels]
def combine_data(self, data):
'''
Returns a list that combines the point coordinates
and their labels
'''
print 'Combining data...'
combined_data = []
labels = data[1]
data_list = np.array(data[0].todense()).tolist()
for i in range(len(labels)):
combined_data.append([labels[i], data_list[i]])
if i%100 == 0:
print 'Combined {} data'.format(i)
return combined_data
def format_data(self, data):
formatted_data = []
print 'Formatting data...'
default_data_features = []
for i in range(len(data[0][1])):
default_data_features.append((i+1, 0))
data_num = 0
for datum in data:
nonzero_elements = np.nonzero(datum[1])[0]
data_features = default_data_features[:]
# pdb.set_trace()
for e in nonzero_elements:
data_features[e-1] = (e+1, datum[1][e])
if data_num%100 == 0:
print 'Formatted {} data'.format(data_num)
data_num += 1
formatted_data.append((datum[0], data_features))
return formatted_data
def format_for_svmlight(self, data):
combined_data = self.combine_data(data)
formatted_data = self.format_data(combined_data)
return formatted_data
def format_tweet_for_svmlight(self, tweet):
data_features = []
word_dict = {}
for word in tweet:
try:
word_dict[word] += 1
except:
word_dict[word] = 1
for word in tweet:
try:
idx = self.index_map[word]
data_features.append((idx, word_dict[word]))
except:
pass
return [(1, data_features)]
def read_model(self, rel_path):
abs_path = os.path.abspath(rel_path)
model = svmlight.read_model(abs_path)
return model
def train(self, data, t=0, C=1.0):
model = svmlight.learn(data, type="classifier", t=t, C=C)
return model
def get_weather_tweets(self, tweets):
weather_tweets = []
if not isinstance(tweets, list):
tweets = [tweets]
count = 0
for tweet in tweets:
count += 1
formatted_tweet = self.parser.stem_sentence_porter(tweet)
formatted_tweet = self.format_tweet_for_svmlight(formatted_tweet)
c = svmlight.classify(self.is_weather_model, formatted_tweet)
if count%100 == 0:
print count
if c[0] < 0:
weather_tweets.append(tweet)
return weather_tweets
def classify(self, model, data):
classifications = svmlight.classify(model, data)
return classifications
def classify_tweet(self, tweet):
try:
tweet = self.parser.stem_sentence_porter(tweet)
formatted_tweet = self.format_tweet_for_svmlight(tweet)
time_class = []
weather_class = []
for model in self.time_models:
time_class.append(self.classify(model, formatted_tweet)[0])
for model in self.weather_models:
weather_class.append(self.classify(model, formatted_tweet)[0])
return weather_class, time_class
except:
print 'You have yet to load the models.'
print 'Please load all models with load_all_models()'
return None
def classify_tweets(self, tweets, formatted_tweets):
weather_class = []
tweet_dict = {}
count = 0
for model in self.weather_models:
scores = self.classify(model, formatted_tweets)
weather_class.append(scores)
for i in range(len(scores)):
if scores[i] > self.threshold:
try:
tweet_dict[self.weather_labels[count]].append(tweets[i])
except:
tweet_dict[self.weather_labels[count]] = [tweets[i]]
count += 1
results = []
for i in range(len(weather_class)):
results.append([sum(weather_class[i]), self.weather_labels[i]])
return results, tweet_dict
