ptm/utils.py - master - SeregKa/python-topic-model

ptm/utils.py
124 строки3 KB
Dongwoo Kim
Add example of author-topic model with cora dataset #4
15 фев 2016, 01:03
15 фев 2016, 01:0374de3de
import numpy as np

from six.moves import xrange


def sampling_from_dist(prob):
    """ Sample index from a list of unnormalised probability distribution
        same as np.random.multinomial(1, prob/np.sum(prob)).argmax()

    Parameters
    ----------
    prob: ndarray
        array of unnormalised probability distribution

    Returns
    -------
    new_topic: return a sampled index
    """
    thr = prob.sum() * np.random.rand()
    new_topic = 0
    tmp = prob[new_topic]
    while tmp < thr:
        new_topic += 1
        tmp += prob[new_topic]
    return new_topic


def sampling_from_dict(prob):
    """ sample key from dictionary `prob` where values are unnormalised probability distribution

    Parameters
    ----------
    prob: dict
        key = topic
        value = unnormalised probability of the topic

    Returns
    -------
    key: int
        sampled key
    """
    prob_sum = sum(prob.values())

    thr = prob_sum * np.random.rand()
    tmp = 0
    for key, p in prob.items():
        tmp += p
        if tmp < thr:
            new_topic = key
    return new_topic


def isfloat(value):
    """
    Check the value is convertable to float value
    """
    try:
        float(value)
        return True
    except ValueError:
        return False


def read_voca(path):
    """
    open file from path and read each line to return the word list
    """
    with open(path, 'r') as f:
        return [word.strip() for word in f.readlines()]


def word_cnt_to_bow_list(word_ids, word_cnt):
    corpus_list = list()
    for di in xrange(len(word_ids)):
        doc_list = list()
        for wi in xrange(len(word_ids[di])):
            word = word_ids[di][wi]
            for c in xrange(word_cnt[di][wi]):
                doc_list.append(word)
        corpus_list.append(doc_list)
    return corpus_list


def log_normalize(log_prob_vector):
    """
    returns a probability vector of log probability vector
    """
    max_v = log_prob_vector.max()
    log_prob_vector += max_v
    log_prob_vector = np.exp(log_prob_vector)
    log_prob_vector /= log_prob_vector.sum()
    return log_prob_vector


def convert_cnt_to_list(word_ids, word_cnt):
    corpus = list()

    for di in xrange(len(word_ids)):
        doc = list()
        doc_ids = word_ids[di]
        doc_cnt = word_cnt[di]
        for wi in xrange(len(doc_ids)):
            word_id = doc_ids[wi]
            for si in xrange(doc_cnt[wi]):
                doc.append(word_id)
        corpus.append(doc)
    return corpus


def write_top_words(topic_word_matrix, vocab, filepath, n_words=20, delimiter=',', newline='\n'):
    with open(filepath, 'w') as f:
        for ti in xrange(topic_word_matrix.shape[0]):
            top_words = vocab[topic_word_matrix[ti, :].argsort()[::-1][:n_words]]
            f.write('%d' % (ti))
            for word in top_words:
                f.write(delimiter + word)
            f.write(newline)


def get_top_words(topic_word_matrix, vocab, topic, n_words=20):
    if not isinstance(vocab, np.ndarray):
        vocab = np.array(vocab)
    top_words = vocab[topic_word_matrix[topic].argsort()[::-1][:n_words]]
    return top_words