Python theano.ifelse.ifelse() Examples

def __init__(self, input, centerbias = None, alpha=1.0):
        self.input = input
        if centerbias is None:
            centerbias = np.ones(12)
        self.alpha = theano.shared(value = np.array(alpha).astype(theano.config.floatX), name='alpha')
        self.centerbias_ys = theano.shared(value=np.array(centerbias, dtype=theano.config.floatX), name='centerbias_ys')
        self.centerbias_xs = theano.shared(value=np.linspace(0, 1, len(centerbias), dtype=theano.config.floatX), name='centerbias_xs')

        height = T.cast(input.shape[0], theano.config.floatX)
        width = T.cast(input.shape[1], theano.config.floatX)
        x_coords = (T.arange(width) - 0.5*width) / (0.5*width)
        y_coords = (T.arange(height) - 0.5*height) / (0.5*height) + 0.0001  # We cannot have zeros in there because of grad

        x_coords = x_coords.dimshuffle('x', 0)
        y_coords = y_coords.dimshuffle(0, 'x')

        dists = T.sqrt(T.square(x_coords) + self.alpha*T.square(y_coords))
        self.max_dist = T.sqrt(1 + self.alpha)
        self.dists = dists/self.max_dist

        self.factors = nonlinearity(self.dists, self.centerbias_xs, self.centerbias_ys, len(centerbias))

        apply_centerbias = T.gt(self.centerbias_ys.shape[0], 2)
        self.output = ifelse(apply_centerbias, self.input+self.factors, self.input)
        self.params = [self.centerbias_ys, self.alpha] 

def test_sparse_tensor_error(self):
        import theano.sparse
        if not theano.sparse.enable_sparse:
            raise SkipTest("Optimization temporarily disabled")
        rng = numpy.random.RandomState(utt.fetch_seed())
        data = rng.rand(2, 3).astype(self.dtype)
        x = self.shared(data)
        y = theano.sparse.matrix('csc', dtype=self.dtype, name='y')
        z = theano.sparse.matrix('csr', dtype=self.dtype, name='z')
        cond = theano.tensor.iscalar('cond')

        self.assertRaises(TypeError, ifelse, cond, x, y)
        self.assertRaises(TypeError, ifelse, cond, y, x)
        self.assertRaises(TypeError, ifelse, cond, x, z)
        self.assertRaises(TypeError, ifelse, cond, z, x)
        self.assertRaises(TypeError, ifelse, cond, y, z)
        self.assertRaises(TypeError, ifelse, cond, z, y) 

def __init__(self, input, centerbias = None, alpha=1.0):
        self.input = input
        if centerbias is None:
            centerbias = np.ones(12)
        self.alpha = theano.shared(value = np.array(alpha).astype(theano.config.floatX), name='alpha')
        self.centerbias_ys = theano.shared(value=np.array(centerbias, dtype=theano.config.floatX), name='centerbias_ys')
        self.centerbias_xs = theano.shared(value=np.linspace(0, 1, len(centerbias), dtype=theano.config.floatX), name='centerbias_xs')

        height = T.cast(input.shape[0], theano.config.floatX)
        width = T.cast(input.shape[1], theano.config.floatX)
        x_coords = (T.arange(width) - 0.5*width) / (0.5*width)
        y_coords = (T.arange(height) - 0.5*height) / (0.5*height) + 0.0001  # We cannot have zeros in there because of grad

        x_coords = x_coords.dimshuffle('x', 0)
        y_coords = y_coords.dimshuffle(0, 'x')

        dists = T.sqrt(T.square(x_coords) + self.alpha*T.square(y_coords))
        self.max_dist = T.sqrt(1 + self.alpha)
        self.dists = dists/self.max_dist

        self.factors = nonlinearity(self.dists, self.centerbias_xs, self.centerbias_ys, len(centerbias))

        apply_centerbias = T.gt(self.centerbias_ys.shape[0], 2)
        self.output = ifelse(apply_centerbias, self.input*self.factors, self.input)
        self.params = [self.centerbias_ys, self.alpha] 

def test_c_thunks():
    a = tensor.scalars('a')
    b, c = tensor.vectors('bc')
    cases = [False]
    if theano.config.cxx:
        cases.append(True)
    for c_thunks in cases:
        f = function([a, b, c], ifelse(a, a * b, b * c),
                     mode=Mode(
                         optimizer=None,
                         linker=vm.VM_Linker(c_thunks=c_thunks,
                                             use_cloop=False)))
        f(1, [2], [3, 2])
        from nose.tools import assert_raises
        assert_raises(ValueError, f, 0, [2], [3, 4])
        assert any([hasattr(t, 'cthunk') for t in f.fn.thunks]) == c_thunks 

def __init__(self, rng, input, n_in, n_out, is_train,
                 activation, dropout_rate, mask=None, W=None, b=None):
        super(DropoutHiddenLayer, self).__init__(
                rng=rng, input=input, n_in=n_in, n_out=n_out, W=W, b=b,
                activation=activation)

        self.dropout_rate = dropout_rate
        self.srng = T.shared_randomstreams.RandomStreams(rng.randint(999999))
        self.mask = mask
        self.layer = self.output

        # Computes outputs for train and test phase applying dropout when needed.
        train_output = self.layer * T.cast(self.mask, theano.config.floatX)
        test_output = self.output * (1 - dropout_rate)
        self.output = ifelse(T.eq(is_train, 1), train_output, test_output)
        return 

def test_lazy_if(self):
        # Tests that lazy if works .. even if the two results have different
        # shapes but the same type (i.e. both vectors, or matrices or
        # whatnot of same dtype)
        x = tensor.vector('x', dtype=self.dtype)
        y = tensor.vector('y', dtype=self.dtype)
        c = tensor.iscalar('c')
        f = theano.function([c, x, y], ifelse(c, x, y), mode=self.mode)
        self.assertFunctionContains1(f, self.get_ifelse(1))
        rng = numpy.random.RandomState(utt.fetch_seed())

        xlen = rng.randint(200)
        ylen = rng.randint(200)

        vx = numpy.asarray(rng.uniform(size=(xlen,)), self.dtype)
        vy = numpy.asarray(rng.uniform(size=(ylen,)), self.dtype)

        assert numpy.allclose(vx, f(1, vx, vy))
        assert numpy.allclose(vy, f(0, vx, vy)) 

def get_aggregator(self):
        initialized = shared_like(0.)
        expression_acc = shared_like(self.expression)

        # Dummy default expression to use as the previously-accumulated
        # value, that has the same shape as the new result
        expression_zeros = tensor.as_tensor(self.expression).zeros_like()

        conditional_update_expr = self.expression + ifelse(initialized,
                                                           expression_acc,
                                                           expression_zeros)

        initialization_updates = [(expression_acc,
                                   tensor.zeros_like(expression_acc)),
                                  (initialized, 0.)]
        accumulation_updates = [(expression_acc,
                                 conditional_update_expr),
                                (initialized, 1.)]
        aggregator = Aggregator(aggregation_scheme=self,
                                initialization_updates=initialization_updates,
                                accumulation_updates=accumulation_updates,
                                readout_variable=(expression_acc))
        return aggregator 

def test_sparse_tensor_error(self):
        import theano.sparse
        if not theano.sparse.enable_sparse:
            raise SkipTest("Optimization temporarily disabled")
        rng = numpy.random.RandomState(utt.fetch_seed())
        data = rng.rand(2, 3).astype(self.dtype)
        x = self.shared(data)
        y = theano.sparse.matrix('csc', dtype=self.dtype, name='y')
        z = theano.sparse.matrix('csr', dtype=self.dtype, name='z')
        cond = theano.tensor.iscalar('cond')

        self.assertRaises(TypeError, ifelse, cond, x, y)
        self.assertRaises(TypeError, ifelse, cond, y, x)
        self.assertRaises(TypeError, ifelse, cond, x, z)
        self.assertRaises(TypeError, ifelse, cond, z, x)
        self.assertRaises(TypeError, ifelse, cond, y, z)
        self.assertRaises(TypeError, ifelse, cond, z, y) 

def test_lazy_if(self):
        # Tests that lazy if works .. even if the two results have different
        # shapes but the same type (i.e. both vectors, or matrices or
        # whatnot of same dtype)
        x = tensor.vector('x', dtype=self.dtype)
        y = tensor.vector('y', dtype=self.dtype)
        c = tensor.iscalar('c')
        f = theano.function([c, x, y], ifelse(c, x, y), mode=self.mode)
        self.assertFunctionContains1(f, self.get_ifelse(1))
        rng = numpy.random.RandomState(utt.fetch_seed())

        xlen = rng.randint(200)
        ylen = rng.randint(200)

        vx = numpy.asarray(rng.uniform(size=(xlen,)), self.dtype)
        vy = numpy.asarray(rng.uniform(size=(ylen,)), self.dtype)

        assert numpy.allclose(vx, f(1, vx, vy))
        assert numpy.allclose(vy, f(0, vx, vy)) 

def __init__(self, rng, input, n_in, n_out, is_train,
                 activation, dropout_rate, mask=None, W=None, b=None):
        super(DropoutHiddenLayer, self).__init__(
                rng=rng, input=input, n_in=n_in, n_out=n_out, W=W, b=b,
                activation=activation)

        self.dropout_rate = dropout_rate
        self.srng = T.shared_randomstreams.RandomStreams(rng.randint(999999))
        self.mask = mask
        self.layer = self.output

        # Computes outputs for train and test phase applying dropout when needed.
        train_output = self.layer * T.cast(self.mask, theano.config.floatX)
        test_output = self.output * (1 - dropout_rate)
        self.output = ifelse(T.eq(is_train, 1), train_output, test_output)
        return 

def test_c_thunks():
    a = tensor.scalars('a')
    b, c = tensor.vectors('bc')
    cases = [False]
    if theano.config.cxx:
        cases.append(True)
    for c_thunks in cases:
        f = function([a, b, c], ifelse(a, a * b, b * c),
                     mode=Mode(
                         optimizer=None,
                         linker=vm.VM_Linker(c_thunks=c_thunks,
                                             use_cloop=False)))
        f(1, [2], [3, 2])
        from nose.tools import assert_raises
        assert_raises(ValueError, f, 0, [2], [3, 4])
        assert any([hasattr(t, 'cthunk') for t in f.fn.thunks]) == c_thunks 

def gate_layer(tparams, X_word, X_char, options, prefix, pretrain_mode, activ='lambda x: x', **kwargs):
    """ 
    compute the forward pass for a gate layer

    Parameters
    ----------
    tparams        : OrderedDict of theano shared variables, {parameter name: value}
    X_word         : theano 3d tensor, word input, dimensions: (num of time steps, batch size, dim of vector)
    X_char         : theano 3d tensor, char input, dimensions: (num of time steps, batch size, dim of vector)
    options        : dictionary, {hyperparameter: value}
    prefix         : string, layer name
    pretrain_mode  : theano shared scalar, 0. = word only, 1. = char only, 2. = word & char
    activ          : string, activation function: 'liner', 'tanh', or 'rectifier'

    Returns
    -------
    X              : theano 3d tensor, final vector, dimensions: (num of time steps, batch size, dim of vector)

    """      
    # compute gating values, Eq.(3)
    G = tensor.nnet.sigmoid(tensor.dot(X_word, tparams[p_name(prefix, 'v')]) + tparams[p_name(prefix, 'b')][0])
    X = ifelse(tensor.le(pretrain_mode, numpy.float32(1.)),  
               ifelse(tensor.eq(pretrain_mode, numpy.float32(0.)), X_word, X_char),
               G[:, :, None] * X_char + (1. - G)[:, :, None] * X_word)   
    return eval(activ)(X) 

def concat_layer(tparams, X_word, X_char, options, prefix, pretrain_mode, activ='lambda x: x', **kwargs):
    """ 
    compute the forward pass for a concat layer

    Parameters
    ----------
    tparams        : OrderedDict of theano shared variables, {parameter name: value}
    X_word         : theano 3d tensor, word input, dimensions: (num of time steps, batch size, dim of vector)
    X_char         : theano 3d tensor, char input, dimensions: (num of time steps, batch size, dim of vector)
    options        : dictionary, {hyperparameter: value}
    prefix         : string,  layer name
    pretrain_mode  : theano shared scalar, 0. = word only, 1. = char only, 2. = word & char
    activ          : string, activation function: 'liner', 'tanh', or 'rectifier'

    Returns
    -------
    X              : theano 3d tensor, final vector, dimensions: (num of time steps, batch size, dim of vector)

    """
    X = ifelse(tensor.le(pretrain_mode, numpy.float32(1.)),
               ifelse(tensor.eq(pretrain_mode, numpy.float32(0.)), X_word, X_char),
               tensor.dot(tensor.concatenate([X_word, X_char], axis=2), tparams[p_name(prefix, 'W')]) + tparams[p_name(prefix, 'b')]) 
    return eval(activ)(X) 

def connect(self, inputs, is_train):
    """ Trick to speed up model compiling at decoding time.
        (Avoids building a complicated CG.)
    """
    if not self.fix_mask:
      self.generate_mask(inputs.shape, is_train)
     
    if self.fast_predict:
      return inputs * (1 - self.dropout_prob)

    return ifelse(is_train,
            inputs * self.dropout_mask,
            inputs * (1 - self.dropout_prob)) 

def lazy_or(name='none', *args):
    """
    .. todo::

        WRITEME
    """
    def apply_me(args):
        if len(args) == 1:
            return args[0]
        else:
            rval = ifelse(args[0], true, apply_me(args[1:]),
                          name=name + str(len(args)))
            return rval
    return apply_me(args) 

def lazy_and(name='node', *args):
    """
    .. todo::

        WRITEME
    """
    def apply_me(args):
        if len(args) == 1:
            return args[0]
        else:
            rval = ifelse(TT.eq(args[0], zero), false, apply_me(args[1:]),
                         name=name + str(len(args)))
            return rval
    return apply_me(args) 

def learning_updates(self):
        batch_counter = theano.shared(np.array(0, dtype="int32"), "batch_counter")
        batch_size = self.batch_size
        to_update = batch_counter >= batch_size

        for param in self.network.parameters:
            # delta = self.learning_rate * T.grad(self.J, param)
            gsum = theano.shared(np.zeros(param.get_value().shape, dtype=FLOATX), "batch_gsum_%s" % param.name)
            yield gsum, ifelse(to_update, T.zeros_like(gsum), gsum + T.grad(self.cost, param))
            delta = self.learning_rate * gsum / batch_size
            yield param, ifelse(to_update, param - delta, param)

        yield batch_counter, ifelse(to_update, T.constant(0, dtype="int32"), batch_counter + 1) 

def ifelse(condition, then_branch, else_branch):
    return theano_ifelse(condition, then_branch, else_branch) 

def iftrain(self, then_branch, else_branch):
        """
        Execute `then_branch` when training.
        """
        return ifelse(self._training_flag, then_branch, else_branch, name="iftrain") 

def test_ndim_mismatch(self):
        rng = numpy.random.RandomState(utt.fetch_seed())
        data = rng.rand(5).astype(self.dtype)
        x = self.shared(data)
        y = tensor.col('y', self.dtype)
        cond = theano.tensor.iscalar('cond')

        self.assertRaises(TypeError, ifelse, cond, x, y)
        self.assertRaises(TypeError, ifelse, cond, y, x) 

def __init__(self, input, sigma=20.0, window_radius=60):
        self.input = input
        self.sigma = theano.shared(value=np.array(sigma, dtype=theano.config.floatX), name='sigma')
        apply_blur = T.gt(self.sigma, 0.0)
        no_blur = T.le(self.sigma, 0.0)
        self.output = ifelse(no_blur, input, gaussian_filter(input.dimshuffle('x', 0, 1), self.sigma, window_radius)[0, :, :])
        self.params = [self.sigma] 

def get_aggregator(self):
        initialized = shared_like(0.)
        numerator_acc = shared_like(self.numerator)
        denominator_acc = shared_like(self.denominator)

        # Dummy default expression to use as the previously-accumulated
        # value, that has the same shape as the new result
        numerator_zeros = tensor.as_tensor(self.numerator).zeros_like()
        denominator_zeros = tensor.as_tensor(self.denominator).zeros_like()

        conditional_update_num = self.numerator + ifelse(initialized,
                                                         numerator_acc,
                                                         numerator_zeros)
        conditional_update_den = self.denominator + ifelse(initialized,
                                                           denominator_acc,
                                                           denominator_zeros)

        initialization_updates = [(numerator_acc,
                                   tensor.zeros_like(numerator_acc)),
                                  (denominator_acc,
                                   tensor.zeros_like(denominator_acc)),
                                  (initialized, 0.)]
        accumulation_updates = [(numerator_acc,
                                 conditional_update_num),
                                (denominator_acc,
                                 conditional_update_den),
                                (initialized, 1.)]
        aggregator = Aggregator(aggregation_scheme=self,
                                initialization_updates=initialization_updates,
                                accumulation_updates=accumulation_updates,
                                readout_variable=(numerator_acc /
                                                  denominator_acc))
        return aggregator 

def test_grad_test_values(self):
        """
        Regression test for test values of `ifelse` gradient.
        """
        backup = theano.config.compute_test_value
        theano.config.compute_test_value = 'raise'
        try:
            x = tensor.scalar('x')
            x.tag.test_value = 1
            # Used to crash due to undefined test value.
            tensor.grad(ifelse(0, x, x), x)
        finally:
            theano.config.compute_test_value = backup 

def test_merge_ifs_true_false(self):
        raise SkipTest("Optimization temporarily disabled")
        x1 = tensor.scalar('x1')
        x2 = tensor.scalar('x2')
        y1 = tensor.scalar('y1')
        y2 = tensor.scalar('y2')
        w1 = tensor.scalar('w1')
        w2 = tensor.scalar('w2')
        c = tensor.iscalar('c')

        out = ifelse(c,
                     ifelse(c, x1, x2) + ifelse(c, y1, y2) + w1,
                     ifelse(c, x1, x2) + ifelse(c, y1, y2) + w2)
        f = theano.function([x1, x2, y1, y2, w1, w2, c], out,
                            allow_input_downcast=True)
        assert len([x for x in f.maker.fgraph.toposort()
                    if isinstance(x.op, IfElse)]) == 1

        rng = numpy.random.RandomState(utt.fetch_seed())
        vx1 = rng.uniform()
        vx2 = rng.uniform()
        vy1 = rng.uniform()
        vy2 = rng.uniform()
        vw1 = rng.uniform()
        vw2 = rng.uniform()
        assert numpy.allclose(f(vx1, vx2, vy1, vy2, vw1, vw2, 1),
                              vx1 + vy1 + vw1)
        assert numpy.allclose(f(vx1, vx2, vy1, vy2, vw1, vw2, 0),
                              vx2 + vy2 + vw2) 

def test_pushout2(self):
        raise SkipTest("Optimization temporarily disabled")
        x1 = tensor.scalar('x1')
        x2 = tensor.scalar('x2')
        y1 = tensor.scalar('y1')
        y2 = tensor.scalar('y2')
        w1 = tensor.scalar('w1')
        w2 = tensor.scalar('w2')
        c = tensor.iscalar('c')
        x, y = ifelse(c, (x1, y1), (x2, y2), name='f1')
        z = ifelse(x > y, w1, w2, name='f2')
        out = x * z * y

        f = theano.function([x1, x2, y1, y2, w1, w2, c], out,
                            allow_input_downcast=True)
        assert isinstance(f.maker.fgraph.toposort()[-1].op, IfElse)
        rng = numpy.random.RandomState(utt.fetch_seed())
        vx1 = rng.uniform()
        vx2 = rng.uniform()
        vy1 = rng.uniform()
        vy2 = rng.uniform()
        vw1 = rng.uniform()
        vw2 = rng.uniform()
        if vx1 > vy1:
            vw = vw1
        else:
            vw = vw2
        assert numpy.allclose(f(vx1, vx2, vy1, vy2, vw1, vw2, 1),
                              vx1 * vy1 * vw)

        if vx2 > vy2:
            vw = vw1
        else:
            vw = vw2
        assert numpy.allclose(f(vx1, vx2, vy1, vy2, vw1, vw2, 0),
                              vx2 * vy2 * vw) 

def test_pushout1(self):
        raise SkipTest("Optimization temporarily disabled")
        x1 = tensor.scalar('x1')
        x2 = tensor.scalar('x2')
        y1 = tensor.scalar('y1')
        y2 = tensor.scalar('y2')
        w1 = tensor.scalar('w1')
        w2 = tensor.scalar('w2')
        c = tensor.iscalar('c')
        x, y = ifelse(c, (x1, y1), (x2, y2), name='f1')
        z = ifelse(c, w1, w2, name='f2')
        out = x * z * y

        f = theano.function([x1, x2, y1, y2, w1, w2, c], out,
                            allow_input_downcast=True)
        assert isinstance(f.maker.fgraph.toposort()[-1].op, IfElse)
        rng = numpy.random.RandomState(utt.fetch_seed())
        vx1 = rng.uniform()
        vx2 = rng.uniform()
        vy1 = rng.uniform()
        vy2 = rng.uniform()
        vw1 = rng.uniform()
        vw2 = rng.uniform()

        assert numpy.allclose(f(vx1, vx2, vy1, vy2, vw1, vw2, 1),
                              vx1 * vy1 * vw1)
        assert numpy.allclose(f(vx1, vx2, vy1, vy2, vw1, vw2, 0),
                              vx2 * vy2 * vw2) 

def test_remove_useless_inputs2(self):
        raise SkipTest("Optimization temporarily disabled")
        x1 = tensor.vector('x1')
        x2 = tensor.vector('x2')
        y1 = tensor.vector('y1')
        y2 = tensor.vector('y2')
        c = tensor.iscalar('c')
        z = ifelse(c, (x1, x1, x1, x2, x2), (y1, y1, y2, y2, y2))
        f = theano.function([c, x1, x2, y1, y2], z)

        ifnode = [x for x in f.maker.fgraph.toposort()
                  if isinstance(x.op, IfElse)][0]
        assert len(ifnode.outputs) == 3 

def test_remove_useless_inputs1(self):
        raise SkipTest("Optimization temporarily disabled")
        x = tensor.vector('x')
        y = tensor.vector('y')
        c = tensor.iscalar('c')
        z = ifelse(c, (x, x), (y, y))
        f = theano.function([c, x, y], z)

        ifnode = [n for n in f.maker.fgraph.toposort()
                  if isinstance(n.op, IfElse)][0]
        assert len(ifnode.inputs) == 3 

def test_merge(self):
        raise SkipTest("Optimization temporarily disabled")
        x = tensor.vector('x')
        y = tensor.vector('y')
        c = tensor.iscalar('c')
        z1 = ifelse(c, x + 1, y + 1)
        z2 = ifelse(c, x + 2, y + 2)
        z = z1 + z2
        f = theano.function([c, x, y], z)
        assert len([n for n in f.maker.fgraph.toposort()
                    if isinstance(n.op, IfElse)]) == 1 

def test_multiple_out_grad(self):
        # Tests that we can compute the gradients through lazy if
        x1 = tensor.vector('x1')
        x2 = tensor.vector('x2')
        y1 = tensor.vector('y1')
        y2 = tensor.vector('y2')
        c = tensor.iscalar('c')
        z = ifelse(c, (x1, x2), (y1, y2))
        grads = tensor.grad(z[0].sum() + z[1].sum(),
                            [x1, x2, y1, y2])

        f = theano.function([c, x1, x2, y1, y2], grads)
        rng = numpy.random.RandomState(utt.fetch_seed())

        lens = [rng.randint(200) for i in range(4)]
        values = [numpy.asarray(rng.uniform(size=(l,)), theano.config.floatX)
                  for l in lens]
        outs_1 = f(1, *values)
        assert all([x.shape[0] == y for x, y in zip(outs_1, lens)])
        assert numpy.all(outs_1[0] == 1.)
        assert numpy.all(outs_1[1] == 1.)
        assert numpy.all(outs_1[2] == 0.)
        assert numpy.all(outs_1[3] == 0.)

        outs_0 = f(0, *values)
        assert all([x.shape[0] == y for x, y in zip(outs_1, lens)])
        assert numpy.all(outs_0[0] == 0.)
        assert numpy.all(outs_0[1] == 0.)
        assert numpy.all(outs_0[2] == 1.)
        assert numpy.all(outs_0[3] == 1.)