Python theano.function方法代码示例

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def batch_normalization(x, mean, var, beta, gamma, epsilon=1e-3):
    '''Apply batch normalization on x given mean, var, beta and gamma.
    '''
    # TODO remove this if statement when Theano without
    # T.nnet.bn.batch_normalization_test is deprecated
    if not hasattr(T.nnet.bn, 'batch_normalization_test'):
        return _old_batch_normalization(x, mean, var, beta, gamma, epsilon)

    if mean.ndim == 1:
        # based on TensorFlow's default: normalize along rightmost dimension
        reduction_axes = range(x.ndim - 1)
    else:
        reduction_axes = [i for i in range(x.ndim) if mean.broadcastable[i]]

    return T.nnet.bn.batch_normalization_test(
        x, gamma, beta, mean, var, reduction_axes, epsilon)


# TODO remove this function when Theano without
# T.nnet.bn.batch_normalization_train is deprecated 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def test_cmrnorm():
    from theano.tests.unittest_tools import verify_grad

    xtest = np.random.rand(2,8,3,4)
    xtest = xtest.astype(theano.config.floatX)

    x = T.tensor4('x', dtype=theano.config.floatX)
    x.tag.test_value = xtest

    y = cmrnorm(x, input_shape=xtest.shape[1:])
    f = theano.function([x], y, mode='DEBUG_MODE')
    f(xtest)

    f = theano.function([x], gpu_from_host(T.grad(T.sum(y), wrt=x)),
                        mode='DEBUG_MODE')
    f(xtest)
    theano.printing.debugprint(f)

    T.verify_grad(lambda x: cmrnorm(x, input_shape=xtest.shape[1:]),
                  (xtest,),
                  rng=np.random.RandomState(0))

    print 'cmrnorm passed' 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def build_cost(logits, targets):
    """
    Build a classification cost function.
    """
    # Clip gradients coming from the cost function.
    logits = theano.gradient.grad_clip(
        logits, -1. * FLAGS.clipping_max_value, FLAGS.clipping_max_value)

    predicted_dist = T.nnet.softmax(logits)

    costs = T.nnet.categorical_crossentropy(predicted_dist, targets)
    cost = costs.mean()

    pred = T.argmax(logits, axis=1)
    acc = 1. - T.mean(T.cast(T.neq(pred, targets), theano.config.floatX))

    return cost, acc 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def test_speed(self):
        top = self.stack.final_stack[-self.batch_size:]
        cost = self._make_cost(top)
        error_signal = T.grad(cost, top)

        # Build automatic backprop function.
        self.stack.make_backprop_scan(error_signal, [self.y],
                                      compute_embedding_gradients=False)
        f = theano.function(
            [self.X, self.transitions, self.y],
            [cost] + self.stack.gradients.values(),
            updates=self.stack.scan_updates + self.stack.bscan_updates)
        theano.printing.debugprint(f.maker.fgraph.outputs[1])

        for t in range(10):
            self._run_batch(f) 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def get_corrupted_input(self, input, corruption_level):
        """This function keeps ``1-corruption_level`` entries of the inputs the
        same and zero-out randomly selected subset of size ``coruption_level``
        Note : first argument of theano.rng.binomial is the shape(size) of
               random numbers that it should produce
               second argument is the number of trials
               third argument is the probability of success of any trial

                this will produce an array of 0s and 1s where 1 has a
                probability of 1 - ``corruption_level`` and 0 with
                ``corruption_level``

                The binomial function return int64 data type by
                default.  int64 multiplicated by the input
                type(floatX) always return float64.  To keep all data
                in floatX when floatX is float32, we set the dtype of
                the binomial to floatX. As in our case the value of
                the binomial is always 0 or 1, this don't change the
                result. This is needed to allow the gpu to work
                correctly as it only support float32 for now.

        """
        return self.theano_rng.binomial(size=input.shape, n=1,
                                        p=1 - corruption_level,
                                        dtype=theano.config.floatX) * input 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def save(self, training_losses, save_dir, step):
        ''' Save the current network parameters to the save_dir and make a
        symlink to the latest param so that the training function can easily
        load the latest model'''
        save_path = os.path.join(save_dir, 'weights.%d' % (step))
        self.net.save(save_path)

        # Make a symlink for weights.npy
        symlink_path = os.path.join(save_dir, 'weights.npy')
        if os.path.lexists(symlink_path):
            os.remove(symlink_path)

        # Make a symlink to the latest network params
        os.symlink("%s.npy" % os.path.abspath(save_path), symlink_path)

        # Write the losses
        with open(os.path.join(save_dir, 'loss.%d.txt' % step), 'w') as f:
            f.write('\n'.join([str(l) for l in training_losses])) 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def return_network(self):
        '''This function returns weight matrix and bias vectors of each hidden layer in the 
        final network after training.'''

        weights_all_layer = []
        bias_all_layer = []
        bias_prime_all_layer = []

        for dA_layer in self.dA_layers:
            weight = dA_layer.W.get_value(borrow = True)
            bias = dA_layer.b.get_value(borrow = True)
            bias_prime = dA_layer.b_prime.get_value(borrow = True)
            weights_all_layer.append(weight)
            bias_all_layer.append(bias)
            bias_prime_all_layer.append(bias_prime)

        return weights_all_layer, bias_all_layer, bias_prime_all_layer 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def test_softmax():

    from keras.activations import softmax as s

    # Test using a reference implementation of softmax
    def softmax(values):
        m = max(values)
        values = numpy.array(values)
        e = numpy.exp(values - m)
        dist = list(e / numpy.sum(e))

        return dist

    x = T.vector()
    exp = s(x)
    f = theano.function([x], exp)
    test_values=get_standard_values()

    result = f(test_values)
    expected = softmax(test_values)

    print(str(result))
    print(str(expected))

    list_assert_equal(result, expected) 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def test_tanh():

    from keras.activations import tanh as t
    test_values = get_standard_values()

    x = T.vector()
    exp = t(x)
    f = theano.function([x], exp)

    result = f(test_values)
    expected = [math.tanh(v) for v in test_values]

    print(result)
    print(expected)

    list_assert_equal(result, expected) 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def adadelta(lr, tparams, grads, inp, cost):
    running_up2 = [theano.shared(p.get_value() * numpy.float32(0.), name='%s_rup2'%k) for k, p in tparams.iteritems()]
    running_grads2 = [theano.shared(p.get_value() * numpy.float32(0.), name='%s_rgrad2'%k) for k, p in tparams.iteritems()]

    rg2_new = [0.95 * rg2 + 0.05 * (g ** 2) for rg2, g in zip(running_grads2, grads)]
    rg2up = [(rg2, r_n) for rg2, r_n in zip(running_grads2, rg2_new)]
    
    
    updir = [-tensor.sqrt(ru2 + 1e-6) / tensor.sqrt(rg2 + 1e-6) * zg for zg, ru2, rg2 in zip(grads, running_up2, rg2_new)]
    ru2up = [(ru2, 0.95 * ru2 + 0.05 * (ud ** 2)) for ru2, ud in zip(running_up2, updir)]
    param_up = [(p, p + ud) for p, ud in zip(itemlist(tparams), updir)]

    inp += [lr]
    f_update = theano.function(inp, cost, updates=rg2up+ru2up+param_up, on_unused_input='ignore', profile=profile)

    return f_update 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def debugging_adadelta(lr, tparams, grads, inp, cost):
    zipped_grads = [theano.shared(p.get_value() * numpy.float32(0.), name='%s_grad'%k) for k, p in tparams.iteritems()]
    running_up2 = [theano.shared(p.get_value() * numpy.float32(0.), name='%s_rup2'%k) for k, p in tparams.iteritems()]
    running_grads2 = [theano.shared(p.get_value() * numpy.float32(0.), name='%s_rgrad2'%k) for k, p in tparams.iteritems()]

    zgup = [(zg, g) for zg, g in zip(zipped_grads, grads)]
    rg2up = [(rg2, 0.95 * rg2 + 0.05 * (g ** 2)) for rg2, g in zip(running_grads2, grads)]

    f_grad_shared = theano.function(inp, cost, updates=zgup+rg2up, profile=profile)
    
    
    updir = [-tensor.sqrt(ru2 + 1e-6) / tensor.sqrt(rg2 + 1e-6) * zg for zg, ru2, rg2 in zip(zipped_grads, running_up2, running_grads2)]
    ru2up = [(ru2, 0.95 * ru2 + 0.05 * (ud ** 2)) for ru2, ud in zip(running_up2, updir)]
    param_up = [(p, p + ud) for p, ud in zip(itemlist(tparams), updir)]

    f_update = theano.function([lr], [], updates=ru2up+param_up, on_unused_input='ignore', profile=profile)

    return f_grad_shared, f_update 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def rmsprop(lr, tparams, grads, inp, cost):
    zipped_grads = [theano.shared(p.get_value() * numpy.float32(0.), name='%s_grad'%k) for k, p in tparams.iteritems()]
    running_grads = [theano.shared(p.get_value() * numpy.float32(0.), name='%s_rgrad'%k) for k, p in tparams.iteritems()]
    running_grads2 = [theano.shared(p.get_value() * numpy.float32(0.), name='%s_rgrad2'%k) for k, p in tparams.iteritems()]

    zgup = [(zg, g) for zg, g in zip(zipped_grads, grads)]
    rgup = [(rg, 0.95 * rg + 0.05 * g) for rg, g in zip(running_grads, grads)]
    rg2up = [(rg2, 0.95 * rg2 + 0.05 * (g ** 2)) for rg2, g in zip(running_grads2, grads)]

    f_grad_shared = theano.function(inp, cost, updates=zgup+rgup+rg2up, profile=profile)

    updir = [theano.shared(p.get_value() * numpy.float32(0.), name='%s_updir'%k) for k, p in tparams.iteritems()]
    updir_new = [(ud, 0.9 * ud - 1e-4 * zg / tensor.sqrt(rg2 - rg ** 2 + 1e-4)) for ud, zg, rg, rg2 in zip(updir, zipped_grads, running_grads, running_grads2)]
    param_up = [(p, p + udn[1]) for p, udn in zip(itemlist(tparams), updir_new)]
    f_update = theano.function([lr], [], updates=updir_new+param_up, on_unused_input='ignore', profile=profile)

    return f_grad_shared, f_update 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def modelScore(self,tag_ids,scores,s_len):
    #{{{
        """
            ATTENTATION THIS FUNCTION IS SYMBOL PROGRAMMING
            this function is to return the score of our model at a fixed sentence label 
        @param:
            scores:        the scores matrix ,the output of our model
            tag:           a numpy array, which represent one sentence label 
            sent_lens:     a scalar number, the length of sentence.
                because our sentence label will be expand to max sentence length,
                so we will use this to get the original sentence label. 
        @return: 
            a scalar number ,the score;
        """
    #{{{
        n_tags=self.output_dim;
        transitions=self.transitions;
        #score from tags_scores
        real_path_score = scores[T.arange(s_len), tag_ids].sum()

        # Score from transitions
        b_id = theano.shared(value=np.array([n_tags], dtype=np.int32))
        e_id = theano.shared(value=np.array([n_tags + 1], dtype=np.int32))
        padded_tags_ids = T.concatenate([b_id, tag_ids, e_id], axis=0)
        real_path_score += transitions[
                padded_tags_ids[T.arange(s_len + 1)],
                padded_tags_ids[T.arange(s_len + 1) + 1]
            ].sum()
        #to prevent T.exp(real_path_score) to be inf 
        #return real_path_score;
        return real_path_score/s_len;
    #}}}
    #}}} 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def arange(start, stop=None, step=1, dtype='int32'):
    '''Creates a 1-D tensor containing a sequence of integers.

    The function arguments use the same convention as
    Theano's arange: if only one argument is provided,
    it is in fact the "stop" argument.

    The default type of the returned tensor is 'int32' to
    match TensorFlow's default.
    '''
    return T.arange(start, stop=stop, step=step, dtype=dtype) 

# 需要导入模块: import theano [as 别名]
# 或者: from theano import function [as 别名]
def __init__(self, inputs, outputs, updates=[], **kwargs):
        unique_variables_to_update = {}
        for v, nv in updates:
            if v not in unique_variables_to_update:
                unique_variables_to_update[v] = nv
        updates = unique_variables_to_update.items()
        self.function = theano.function(inputs, outputs, updates=updates,
                                        allow_input_downcast=True,
                                        on_unused_input='ignore',
                                        **kwargs)