pos_gradients.py

import numpy as np
from scipy.stats import linregress
import nltk
from sklearn.svm import SVC


pos_index = ['CC', 'CD', 'DT', 'EX', 'FW', 'IN', 'JJ', 'JJR', 'JJS', 'LS', 'MD', 'NN', 'NNP', 'NNPS', 'NNS', 'PDT', 'POS', 'PRP', 'PRP$', 'RB', 'RBR', 'RBS', 'RP', 'SYM', 'TO', 'UH', 'VB', 'VBD', 'VBG', 'VBN', 'VBP', 'VBZ', 'WDT', 'WP', 'WP$', 'WRB']


train = [('i am a banana', 'declarative'),
         ('the weather sure is nice today', 'declarative'),
         ('the mitochondria is the powerhouse of the cell', 'declarative'),
         ('i cannot hold all these feelings because reasons', 'declarative'),
         ('coastal waters tend to be shallow', 'declarative'),

         ('who are you and how did you get here', 'interrogative'),
         ('what are you doing', 'interrogative'),
         ('would you please go to the market and get me a bagel', 'interrogative'),
         ('what does the mitochondria do', 'interrogative'),
         ('what do you call coastal water', 'interrogative'),

         ('get away from me right now', 'imperative'),
         ('do not ever do that again', 'imperative'),
         ('put that gun down', 'imperative'),
         ('never go outside by yourself', 'imperative'),
         ('turn off the toaster and look at me', 'imperative')]

test = [('what time is it', 'interrogative'),
        ('can you get me a beer', 'interrogative'),
        ('what exactly do you think you are doing', 'interrogative'),
        ('how do you know your own name', 'interrogative'),

        ('I do not believe you', 'declarative'),
        ('you cannot possibly be alive', 'declarative'),
        ('this clock is quite broken', 'declarative'),
        ('You are the kindest person I know', 'declarative'),

        ('stop that', 'imperative'),
        ('clean this up', 'imperative'),
        ('cancel your subscription to netflix', 'imperative'),
        ('drive the nail through the board', 'imperative')]


def find_gradient(pos, tokens):
    x = []
    for i in tokens:
        if i[1] == pos:
            x.append(1)
        else:
            x.append(0)
    y = range(len(x))
    line = linregress(x, y)
    #print(line)
    slope = line[0]
    if np.isnan(slope):
        return 0
    else:
        return slope


def sent2grad(sentence):
    t = nltk.word_tokenize(sentence)
    p = nltk.pos_tag(t)
    gradients = []
    for i in pos_index:
        grad = find_gradient(i, p)
        gradients.append(grad)
    return np.asarray(gradients)


if __name__ == '__main__':
    train_x = []
    train_y = []
    for i in train:
        train_x.append(sent2grad(i[0]))
        train_y.append(i[1])
    model = SVC()
    train_x = np.asarray(train_x)
    model.fit(train_x, train_y)
    for i in test:
        vector = sent2grad(i[0])
        vector = np.asarray(vector).reshape(1, -1)
        pred, = model.predict(vector)
        print('Predicted', pred, '   Actual', i[1])