def __init__(self, num_features, min_df=10):
self.priors = {}
self.condprobs = defaultdict(lambda: defaultdict(float))
self.class_label_counts = defaultdict(int)
self.token_tf_counts = defaultdict(lambda: defaultdict(int))
self.token_df_counts = defaultdict(lambda: defaultdict(int)) # Info-Gain requires knowing document frequency counts
self.total_tf = 0
self.feature_selection = IGFeatureSelection(num_features, min_df)
class NBTrain:
def __init__(self, num_features, min_df=10):
self.priors = {}
self.condprobs = defaultdict(lambda: defaultdict(float))
self.class_label_counts = defaultdict(int)
self.token_tf_counts = defaultdict(lambda: defaultdict(int))
self.token_df_counts = defaultdict(lambda: defaultdict(int)) # Info-Gain requires knowing document frequency counts
self.total_tf = 0
self.feature_selection = IGFeatureSelection(num_features, min_df)
def addDocument(self, tokens, class_label):
self.total_tf+=len(tokens)
self.class_label_counts[class_label]+=1
for token in tokens:
self.token_tf_counts[token][class_label]+=1 # Term frequencies
for tokens in set(tokens):
self.token_df_counts[token][class_label]+=1 # Document Frequencies
def train(self):
#calculate class priors
doc_count = sum(self.class_label_counts.values())
for class_label, count in self.class_label_counts.items():
self.priors[class_label] = count/doc_count
#calcualte count of each token in everey class. If token never seen in class then token_counts[token][class_label] = 0
token_tf_counts = defaultdict(lambda: defaultdict(int))
token_df_counts = defaultdict(lambda: defaultdict(int))
for token, class_counts in self.token_tf_counts.items():
for class_label in self.class_label_counts.keys():
try:
token_tf_counts[token][class_label] = self.token_tf_counts[token][class_label]
token_df_counts[token][class_label] = self.token_df_counts[token][class_label]
except KeyError:
token_tf_counts[token][class_label] = 0
token_df_counts[token][class_label] = 0
selected_tokens = self.feature_selection.selectFeatures(token_df_counts, self.class_label_counts)
#calculate conditional probabilities for selected tokens
vocabulary_size = len(self.token_tf_counts.keys())
for token in selected_tokens.keys():
for class_label in self.class_label_counts.keys():
self.condprobs[token][class_label] = (token_tf_counts[token][class_label] + 1)/(self.total_tf + vocabulary_size)
return self.priors, self.condprobs
def __init__(self, num_features, min_df=10):
self.priors = {}
self.condprobs = defaultdict(lambda: defaultdict(float))
self.class_label_counts = defaultdict(int)
self.token_df_counts = defaultdict(lambda: defaultdict(int))
self.feature_selection = IGFeatureSelection(num_features, min_df)