python类logaddexp()的实例源码

def test_logaddexp_range(self):
        x = [1000000, -1000000, 1000200, -1000200]
        y = [1000200, -1000200, 1000000, -1000000]
        z = [1000200, -1000000, 1000200, -1000000]
        for dt in ['f', 'd', 'g']:
            logxf = np.array(x, dtype=dt)
            logyf = np.array(y, dtype=dt)
            logzf = np.array(z, dtype=dt)
            assert_almost_equal(np.logaddexp(logxf, logyf), logzf)

def test_inf(self):
        inf = np.inf
        x = [inf, -inf,  inf, -inf, inf, 1,  -inf,  1]
        y = [inf,  inf, -inf, -inf, 1,   inf, 1,   -inf]
        z = [inf,  inf,  inf, -inf, inf, inf, 1,    1]
        with np.errstate(invalid='raise'):
            for dt in ['f', 'd', 'g']:
                logxf = np.array(x, dtype=dt)
                logyf = np.array(y, dtype=dt)
                logzf = np.array(z, dtype=dt)
                assert_equal(np.logaddexp(logxf, logyf), logzf)

def test_nan(self):
        assert_(np.isnan(np.logaddexp(np.nan, np.inf)))
        assert_(np.isnan(np.logaddexp(np.inf, np.nan)))
        assert_(np.isnan(np.logaddexp(np.nan, 0)))
        assert_(np.isnan(np.logaddexp(0, np.nan)))
        assert_(np.isnan(np.logaddexp(np.nan, np.nan)))

def forward(self, x):
        """
        Implementation of softplus. Overflow avoided by use of the logaddexp function.
        self._lower is added before returning.
        """
        return np.logaddexp(0, x) + self._lower

def ctc_loss(label, prob, remainder, seq_length, batch_size, num_gpu=1, big_num=1e10):
    label_ = [0, 0]
    prob[prob < 1 / big_num] = 1 / big_num
    log_prob = np.log(prob)

    l = len(label)
    for i in range(l):
        label_.append(int(label[i]))
        label_.append(0)

    l_ = 2 * l + 1
    a = np.full((seq_length, l_ + 1), -big_num)
    a[0][1] = log_prob[remainder][0]
    a[0][2] = log_prob[remainder][label_[2]]
    for i in range(1, seq_length):
        row = i * int(batch_size / num_gpu) + remainder
        a[i][1] = a[i - 1][1] + log_prob[row][0]
        a[i][2] = np.logaddexp(a[i - 1][2], a[i - 1][1]) + log_prob[row][label_[2]]
        for j in range(3, l_ + 1):
            a[i][j] = np.logaddexp(a[i - 1][j], a[i - 1][j - 1])
            if label_[j] != 0 and label_[j] != label_[j - 2]:
                a[i][j] = np.logaddexp(a[i][j], a[i - 1][j - 2])
            a[i][j] += log_prob[row][label_[j]]

    return -np.logaddexp(a[seq_length - 1][l_], a[seq_length - 1][l_ - 1])


# label is done with remove_blank
# pred is got from pred_best

def _forward_cpu_one(self, x, t, W):
        begin = self.begins[t]
        end = self.begins[t + 1]

        w = W[self.paths[begin:end]]
        wxy = w.dot(x) * self.codes[begin:end]
        loss = numpy.logaddexp(0.0, -wxy)  # == log(1 + exp(-wxy))
        return numpy.sum(loss)

def check_forward(self, x_data, t_data, use_cudnn=True):
        x = chainer.Variable(x_data)
        t = chainer.Variable(t_data)
        loss = functions.softmax_cross_entropy(
            x, t, use_cudnn=use_cudnn, normalize=self.normalize,
            cache_score=self.cache_score)
        self.assertEqual(loss.data.shape, ())
        self.assertEqual(loss.data.dtype, self.dtype)
        self.assertEqual(hasattr(loss.creator, 'y'), self.cache_score)
        loss_value = float(cuda.to_cpu(loss.data))

        # Compute expected value
        loss_expect = 0.0
        count = 0
        x = numpy.rollaxis(self.x, 1, self.x.ndim).reshape(
            (self.t.size, self.x.shape[1]))
        t = self.t.ravel()
        for xi, ti in six.moves.zip(x, t):
            if ti == -1:
                continue
            log_z = numpy.ufunc.reduce(numpy.logaddexp, xi)
            loss_expect -= (xi - log_z)[ti]
            count += 1

        if self.normalize:
            if count == 0:
                loss_expect = 0.0
            else:
                loss_expect /= count
        else:
            loss_expect /= len(t_data)

        gradient_check.assert_allclose(
            loss_expect, loss_value, **self.check_forward_options)

def check_forward(self, x_data, use_cudnn=True):
        x = chainer.Variable(x_data)
        y = functions.log_softmax(x, use_cudnn)
        self.assertEqual(y.data.dtype, self.dtype)

        log_z = numpy.ufunc.reduce(
            numpy.logaddexp, self.x, axis=1, keepdims=True)
        y_expect = self.x - log_z

        gradient_check.assert_allclose(
            y_expect, y.data, **self.check_forward_options)

def js_with(self, p):
        log_p = np.array([p.log_likelihood(ngram) for ngram in p.unique_ngrams()])
        log_q = np.array([self.log_likelihood(ngram) for ngram in p.unique_ngrams()])
        log_m = np.logaddexp(log_p - np.log(2), log_q - np.log(2))
        kl_p_m = np.sum(np.exp(log_p) * (log_p - log_m))

        log_p = np.array([p.log_likelihood(ngram) for ngram in self.unique_ngrams()])
        log_q = np.array([self.log_likelihood(ngram) for ngram in self.unique_ngrams()])
        log_m = np.logaddexp(log_p - np.log(2), log_q - np.log(2))
        kl_q_m = np.sum(np.exp(log_q) * (log_q - log_m))

        return 0.5*(kl_p_m + kl_q_m) / np.log(2)

def test_logaddexp_values(self):
        x = [1, 2, 3, 4, 5]
        y = [5, 4, 3, 2, 1]
        z = [6, 6, 6, 6, 6]
        for dt, dec_ in zip(['f', 'd', 'g'], [6, 15, 15]):
            xf = np.log(np.array(x, dtype=dt))
            yf = np.log(np.array(y, dtype=dt))
            zf = np.log(np.array(z, dtype=dt))
            assert_almost_equal(np.logaddexp(xf, yf), zf, decimal=dec_)

def test_logaddexp_range(self):
        x = [1000000, -1000000, 1000200, -1000200]
        y = [1000200, -1000200, 1000000, -1000000]
        z = [1000200, -1000000, 1000200, -1000000]
        for dt in ['f', 'd', 'g']:
            logxf = np.array(x, dtype=dt)
            logyf = np.array(y, dtype=dt)
            logzf = np.array(z, dtype=dt)
            assert_almost_equal(np.logaddexp(logxf, logyf), logzf)

def test_inf(self):
        inf = np.inf
        x = [inf, -inf,  inf, -inf, inf, 1,  -inf,  1]
        y = [inf,  inf, -inf, -inf, 1,   inf, 1,   -inf]
        z = [inf,  inf,  inf, -inf, inf, inf, 1,    1]
        with np.errstate(invalid='raise'):
            for dt in ['f', 'd', 'g']:
                logxf = np.array(x, dtype=dt)
                logyf = np.array(y, dtype=dt)
                logzf = np.array(z, dtype=dt)
                assert_equal(np.logaddexp(logxf, logyf), logzf)

def test_nan(self):
        assert_(np.isnan(np.logaddexp(np.nan, np.inf)))
        assert_(np.isnan(np.logaddexp(np.inf, np.nan)))
        assert_(np.isnan(np.logaddexp(np.nan, 0)))
        assert_(np.isnan(np.logaddexp(0, np.nan)))
        assert_(np.isnan(np.logaddexp(np.nan, np.nan)))

def js_with(self, p):
        log_p = np.array([p.log_likelihood(ngram) for ngram in p.unique_ngrams()])
        log_q = np.array([self.log_likelihood(ngram) for ngram in p.unique_ngrams()])
        log_m = np.logaddexp(log_p - np.log(2), log_q - np.log(2))
        kl_p_m = np.sum(np.exp(log_p) * (log_p - log_m))

        log_p = np.array([p.log_likelihood(ngram) for ngram in self.unique_ngrams()])
        log_q = np.array([self.log_likelihood(ngram) for ngram in self.unique_ngrams()])
        log_m = np.logaddexp(log_p - np.log(2), log_q - np.log(2))
        kl_q_m = np.sum(np.exp(log_q) * (log_q - log_m))

        return 0.5 * (kl_p_m + kl_q_m) / np.log(2)

def js_with(self, p):
        log_p = np.array([p.log_likelihood(ngram) for ngram in p.unique_ngrams()])
        log_q = np.array([self.log_likelihood(ngram) for ngram in p.unique_ngrams()])
        log_m = np.logaddexp(log_p - np.log(2), log_q - np.log(2))
        kl_p_m = np.sum(np.exp(log_p) * (log_p - log_m))

        log_p = np.array([p.log_likelihood(ngram) for ngram in self.unique_ngrams()])
        log_q = np.array([self.log_likelihood(ngram) for ngram in self.unique_ngrams()])
        log_m = np.logaddexp(log_p - np.log(2), log_q - np.log(2))
        kl_q_m = np.sum(np.exp(log_q) * (log_q - log_m))

        return 0.5*(kl_p_m + kl_q_m) / np.log(2)

def evalObjectiveFunction(clean_biascount,Pword_plus,Pword_minus,ratio,removing_words):
    mlog(evalObjectiveFunction.__name__,"call")
    import numpy as np
    import math
    #Obj = np.log(1)
    Obj = 0
    for doc in clean_biascount:
        Pdoc = calcProbabilityDocument(Pword_plus,Pword_minus,doc[1],ratio,removing_words)
        i = clean_biascount.index(doc) 
        if i == 0:
            mlog(evalObjectiveFunction.__name__,'Pdoc 0 ' + str(Pdoc) + ' type ' + str(type(Pdoc)))
            t00 = np.exp(np.float64(Pdoc[0]))
            t01 = np.exp(np.float64(Pdoc[1]))
            t1 = t00 - t01
            #print str(t1)
            t2 = abs(t1)
            Obj = np.log(t2)
            #print str(Obj)
        if i % 100 == 0:
            mlog(evalObjectiveFunction.__name__,"Pdoc + " + str(i) + "= " + str(Pdoc))
        if doc[0] == True:
            Obj = np.logaddexp(Obj,Pdoc[0])
            if i % 100 == 1:
                mlog(evalObjectiveFunction.__name__,"Obj+ " + str(i) + " after += " + str(Obj))
            Obj = np.log(np.exp(Obj) - np.exp(Pdoc[1]))
        elif doc[0] == False:
            Obj = np.log(np.exp(np.float64(Obj)) + np.exp(np.float64(Pdoc[1])))
            if i % 100 == 2:
                mlog(evalObjectiveFunction.__name__,"Obj- " + str(i) + " after += " + str(Obj))
            Obj = np.log(np.exp(Obj) - np.exp(Pdoc[0]))
        if Obj == 0.0:
            mlog(evalObjectiveFunction.__name__, "Obj=0 fuck " + str(i) + "")
    mlog(evalObjectiveFunction.__name__,"Obj = " + str(np.exp(Obj)))
    if math.isnan(np.exp(Obj))==False:
        print 'J = ' + str(np.exp(Obj))
    return Obj #type np.log

def sigmoid(x):
    x = np.array(x)
    return np.exp(-np.logaddexp(0, -x))

def update_phi(self, doc_number, time):
        """
        Update variational multinomial parameters, based on a document and a time-slice.
        This is done based on the original Blei-LDA paper, where:
        log_phi := beta * exp(?(gamma)), over every topic for every word.

        TODO: incorporate lee-sueng trick used in **Lee, Seung: Algorithms for non-negative matrix factorization, NIPS 2001**.
        """
        num_topics = self.lda.num_topics
        # digamma values
        dig = np.zeros(num_topics)

        for k in range(0, num_topics):
            dig[k] = digamma(self.gamma[k])

        n = 0   # keep track of iterations for phi, log_phi
        for word_id, count in self.doc:
            for k in range(0, num_topics):
                self.log_phi[n][k] = dig[k] + self.lda.topics[word_id][k]

            log_phi_row = self.log_phi[n]
            phi_row = self.phi[n]

            # log normalize
            v = log_phi_row[0]
            for i in range(1, len(log_phi_row)):
                v = np.logaddexp(v, log_phi_row[i])

            # subtract every element by v
            log_phi_row = log_phi_row - v 
            phi_row = np.exp(log_phi_row)
            self.log_phi[n] = log_phi_row
            self.phi[n] = phi_row
            n +=1 # increase iteration

        return self.phi, self.log_phi

def test_logaddexp_values(self):
        x = [1, 2, 3, 4, 5]
        y = [5, 4, 3, 2, 1]
        z = [6, 6, 6, 6, 6]
        for dt, dec_ in zip(['f', 'd', 'g'], [6, 15, 15]):
            xf = np.log(np.array(x, dtype=dt))
            yf = np.log(np.array(y, dtype=dt))
            zf = np.log(np.array(z, dtype=dt))
            assert_almost_equal(np.logaddexp(xf, yf), zf, decimal=dec_)

def test_logaddexp_range(self):
        x = [1000000, -1000000, 1000200, -1000200]
        y = [1000200, -1000200, 1000000, -1000000]
        z = [1000200, -1000000, 1000200, -1000000]
        for dt in ['f', 'd', 'g']:
            logxf = np.array(x, dtype=dt)
            logyf = np.array(y, dtype=dt)
            logzf = np.array(z, dtype=dt)
            assert_almost_equal(np.logaddexp(logxf, logyf), logzf)