train.py

# create our training data
training = []
output = []
# create an empty array for our output
output_empty = [0] * len(classes)
# training set, bag of words for each sentence
for doc in documents:
    # initialize our bag of words
    bag = []
    # list of tokenized words for the pattern
    pattern_words = doc[0]
    # stem each word
    pattern_words = [lemmatizer.lemmatize(word.lower()) for word in pattern_words]
    #lemmatizer.lemmatize(w.lower()) for w in words if w
    # create our bag of words array
    for w in words:
        bag.append(1) if w in pattern_words else bag.append(0)
    # output is a '0' for each tag and '1' for current tag
    output_row = list(output_empty)
    output_row[classes.index(doc[1])] = 1
    training.append([bag, output_row])
# shuffle our features and turn into np.array
random.shuffle(training)
training = np.array(training)
# create train and test lists
train_x = list(training[:,0])
train_y = list(training[:,1])