Python cntk.constant() Examples

def _padding(x, pattern, axis):
    base_shape = x.shape
    if b_any([dim < 0 for dim in base_shape]):
        raise ValueError('CNTK Backend: padding input tensor with '
                         'shape `%s` contains non-specified dimension, '
                         'which is not supported. Please give fixed '
                         'dimension to enable padding.' % base_shape)
    if pattern[0] > 0:
        prefix_shape = list(base_shape)
        prefix_shape[axis] = pattern[0]
        prefix_shape = tuple(prefix_shape)
        x = C.splice(C.constant(value=0, shape=prefix_shape), x, axis=axis)
        base_shape = x.shape

    if pattern[1] > 0:
        postfix_shape = list(base_shape)
        postfix_shape[axis] = pattern[1]
        postfix_shape = tuple(postfix_shape)
        x = C.splice(x, C.constant(value=0, shape=postfix_shape), axis=axis)

    return x 

def _padding(x, pattern, axis):
    base_shape = x.shape
    if b_any([dim < 0 for dim in base_shape]):
        raise ValueError('CNTK Backend: padding input tensor with '
                         'shape `%s` contains non-specified dimension, '
                         'which is not supported. Please give fixed '
                         'dimension to enable padding.' % base_shape)
    if pattern[0] > 0:
        prefix_shape = list(base_shape)
        prefix_shape[axis] = pattern[0]
        prefix_shape = tuple(prefix_shape)
        x = C.splice(C.constant(value=0, shape=prefix_shape), x, axis=axis)
        base_shape = x.shape
    if pattern[1] > 0:
        postfix_shape = list(base_shape)
        postfix_shape[axis] = pattern[1]
        postfix_shape = tuple(postfix_shape)
        x = C.splice(x, C.constant(value=0, shape=postfix_shape), axis=axis)
    return x 

def _padding(x, pattern, axis):  # pragma: no cover
    base_shape = x.shape
    if b_any([dim < 0 for dim in base_shape]):
        raise ValueError('CNTK Backend: padding input tensor with '
                         'shape `%s` contains non-specified dimension, '
                         'which is not supported. Please give fixed '
                         'dimension to enable padding.' % base_shape)
    if pattern[0] > 0:
        prefix_shape = list(base_shape)
        prefix_shape[axis] = pattern[0]
        prefix_shape = tuple(prefix_shape)
        x = C.splice(C.constant(value=0, shape=prefix_shape), x, axis=axis)
        base_shape = x.shape
    if pattern[1] > 0:
        postfix_shape = list(base_shape)
        postfix_shape[axis] = pattern[1]
        postfix_shape = tuple(postfix_shape)
        x = C.splice(x, C.constant(value=0, shape=postfix_shape), axis=axis)
    return x 

def _padding(x, pattern, axis):
    base_shape = x.shape
    if b_any([dim < 0 for dim in base_shape]):
        raise ValueError('CNTK Backend: padding input tensor with '
                         'shape `%s` contains non-specified dimension, '
                         'which is not supported. Please give fixed '
                         'dimension to enable padding.' % base_shape)
    if pattern[0] > 0:
        prefix_shape = list(base_shape)
        prefix_shape[axis] = pattern[0]
        prefix_shape = tuple(prefix_shape)
        x = C.splice(C.constant(value=0, shape=prefix_shape), x, axis=axis)
        base_shape = x.shape
    if pattern[1] > 0:
        postfix_shape = list(base_shape)
        postfix_shape[axis] = pattern[1]
        postfix_shape = tuple(postfix_shape)
        x = C.splice(x, C.constant(value=0, shape=postfix_shape), axis=axis)
    return x 

def _padding(x, pattern, axis):
    base_shape = x.shape
    if b_any([dim < 0 for dim in base_shape]):
        raise ValueError('CNTK Backend: padding input tensor with '
                         'shape `%s` contains non-specified dimension, '
                         'which is not supported. Please give fixed '
                         'dimension to enable padding.' % base_shape)
    if pattern[0] > 0:
        prefix_shape = list(base_shape)
        prefix_shape[axis] = pattern[0]
        prefix_shape = tuple(prefix_shape)
        x = C.splice(C.constant(value=0, shape=prefix_shape), x, axis=axis)
        base_shape = x.shape
    if pattern[1] > 0:
        postfix_shape = list(base_shape)
        postfix_shape[axis] = pattern[1]
        postfix_shape = tuple(postfix_shape)
        x = C.splice(x, C.constant(value=0, shape=postfix_shape), axis=axis)
    return x 

def _padding(x, pattern, axis):
    base_shape = x.shape
    if b_any([dim < 0 for dim in base_shape]):
        raise ValueError('CNTK Backend: padding input tensor with '
                         'shape `%s` contains non-specified dimension, '
                         'which is not supported. Please give fixed '
                         'dimension to enable padding.' % base_shape)
    if pattern[0] > 0:
        prefix_shape = list(base_shape)
        prefix_shape[axis] = pattern[0]
        prefix_shape = tuple(prefix_shape)
        x = C.splice(C.constant(value=0, shape=prefix_shape), x, axis=axis)
        base_shape = x.shape
    if pattern[1] > 0:
        postfix_shape = list(base_shape)
        postfix_shape[axis] = pattern[1]
        postfix_shape = tuple(postfix_shape)
        x = C.splice(x, C.constant(value=0, shape=postfix_shape), axis=axis)
    return x 

def _padding(x, pattern, axis):
    base_shape = x.shape
    if b_any([dim < 0 for dim in base_shape]):
        raise ValueError('CNTK Backend: padding input tensor with '
                         'shape `%s` contains non-specified dimension, '
                         'which is not supported. Please give fixed '
                         'dimension to enable padding.' % base_shape)
    if pattern[0] > 0:
        prefix_shape = list(base_shape)
        prefix_shape[axis] = pattern[0]
        prefix_shape = tuple(prefix_shape)
        x = C.splice(C.constant(value=0, shape=prefix_shape), x, axis=axis)
        base_shape = x.shape
    if pattern[1] > 0:
        postfix_shape = list(base_shape)
        postfix_shape[axis] = pattern[1]
        postfix_shape = tuple(postfix_shape)
        x = C.splice(x, C.constant(value=0, shape=postfix_shape), axis=axis)
    return x 

def gradients(loss, variables):
    # cntk does not support gradients as symbolic op,
    # to hook up with keras model
    # we will return a constant as place holder, the cntk learner will apply
    # the gradient during training.
    global grad_parameter_dict
    if isinstance(variables, list) is False:
        variables = [variables]
    grads = []
    for v in variables:
        g = C.constant(0, shape=v.shape, name='keras_grad_placeholder')
        grads.append(g)
        grad_parameter_dict[g] = v
    return grads 

def gradients(loss, variables):
    # cntk does not support gradients as symbolic op,
    # to hook up with keras model
    # we will return a constant as place holder, the cntk learner will apply
    # the gradient during training.
    global grad_parameter_dict
    if isinstance(variables, list) is False:
        variables = [variables]
    grads = []
    for v in variables:
        g = C.constant(0, shape=v.shape, name='keras_grad_placeholder')
        grads.append(g)
        grad_parameter_dict[g] = v
    return grads 

def constant(value, dtype=None, shape=None, name=None):
    if dtype is None:
        dtype = floatx()
    if shape is None:
        shape = ()
    np_value = value * np.ones(shape)
    const = C.constant(np_value,
                       dtype=dtype,
                       name=_prepare_name(name, 'constant'))
    const._keras_shape = const.shape
    const._uses_learning_phase = False
    return const 

def gradients(loss, variables):
    # cntk does not support gradients as symbolic op,
    # to hook up with keras model
    # we will return a constant as place holder, the cntk learner will apply
    # the gradient during training.
    global grad_parameter_dict
    if isinstance(variables, list) is False:
        variables = [variables]
    grads = []
    for v in variables:
        g = C.constant(0, shape=v.shape, name='keras_grad_placeholder')
        grads.append(g)
        grad_parameter_dict[g] = v
    return grads 

def constant(value, dtype=None, shape=None, name=None):
    if dtype is None:
        dtype = floatx()
    if shape is None:
        shape = ()
    np_value = value * np.ones(shape)
    const = C.constant(np_value,
                       dtype=dtype,
                       name=_prepare_name(name, 'constant'))
    const._keras_shape = const.shape
    const._uses_learning_phase = False
    return const 

def gradients(loss, variables):
    # cntk does not support gradients as symbolic op,
    # to hook up with keras model
    # we will return a constant as place holder, the cntk learner will apply
    # the gradient during training.
    global grad_parameter_dict
    if isinstance(variables, list) is False:
        variables = [variables]
    grads = []
    for v in variables:
        g = C.constant(0, shape=v.shape, name='keras_grad_placeholder')
        grads.append(g)
        grad_parameter_dict[g] = v
    return grads 

def constant(value, dtype=None, shape=None, name=None):
    if dtype is None:
        dtype = floatx()
    if shape is None:
        shape = ()
    np_value = value * np.ones(shape)
    const = C.constant(np_value,
                       dtype=dtype,
                       name=_prepare_name(name, 'constant'))
    const._keras_shape = const.shape
    const._uses_learning_phase = False
    return const 

def gradients(loss, variables):
    # cntk does not support gradients as symbolic op,
    # to hook up with keras model
    # we will return a constant as place holder, the cntk learner will apply
    # the gradient during training.
    global grad_parameter_dict
    if isinstance(variables, list) is False:
        variables = [variables]
    grads = []
    for v in variables:
        g = C.constant(0, shape=v.shape, name='keras_grad_placeholder')
        grads.append(g)
        grad_parameter_dict[g] = v
    return grads 

def constant(value, dtype=None, shape=None, name=None):
    if dtype is None:
        dtype = floatx()
    if shape is None:
        shape = ()
    np_value = value * np.ones(shape)
    const = C.constant(np_value,
                       dtype=dtype,
                       name=_prepare_name(name, 'constant'))
    const._keras_shape = const.shape
    const._uses_learning_phase = False
    return const 

def gradients(loss, variables):
    # cntk does not support gradients as symbolic op,
    # to hook up with keras model
    # we will return a constant as place holder, the cntk learner will apply
    # the gradient during training.
    global grad_parameter_dict
    if isinstance(variables, list) is False:
        variables = [variables]
    grads = []
    for v in variables:
        g = C.constant(0, shape=v.shape, name='keras_grad_placeholder')
        grads.append(g)
        grad_parameter_dict[g] = v
    return grads 

def constant(value, dtype=None, shape=None, name=None):
    if dtype is None:
        dtype = _FLOATX
    if shape is None:
        shape = ()
    np_value = value * np.ones(shape)
    const = C.constant(np_value,
                       dtype=dtype,
                       name=_prepare_name(name, 'constant'))
    const._keras_shape = shape
    const._uses_learning_phase = False
    return const 

def gradients(loss, variables):
    # cntk does not support gradients as symbolic op,
    # to hook up with keras model
    # we will return a constant as place holder, the cntk learner will apply
    # the gradient during training.
    global grad_parameter_dict
    if isinstance(variables, list) is False:
        variables = [variables]
    grads = []
    for v in variables:
        g = C.constant(0, shape=v.shape, name='keras_grad_placeholder')
        grads.append(g)
        grad_parameter_dict[g] = v
    return grads 

def constant(value, dtype=None, shape=None, name=None):
    if dtype is None:
        dtype = floatx()
    if shape is None:
        shape = ()
    np_value = value * np.ones(shape)
    const = C.constant(np_value,
                       dtype=dtype,
                       name=_prepare_name(name, 'constant'))
    const._keras_shape = const.shape
    const._uses_learning_phase = False
    return const 

def create_model(base_model_file, input_features, num_classes,  dropout_rate = 0.5, freeze_weights = False):
    # Load the pretrained classification net and find nodes
    base_model   = load_model(base_model_file)
    feature_node = find_by_name(base_model, 'features')
    beforePooling_node = find_by_name(base_model, "z.x.x.r")
    #graph.plot(base_model, filename="base_model.pdf") # Write graph visualization

    # Clone model until right before the pooling layer, ie. until including z.x.x.r
    modelCloned = combine([beforePooling_node.owner]).clone(
        CloneMethod.freeze if freeze_weights else CloneMethod.clone,
        {feature_node: placeholder(name='features')})

    # Center the input around zero and set model input.
    # Do this early, to avoid CNTK bug with wrongly estimated layer shapes
    feat_norm = input_features - constant(114)
    model = modelCloned(feat_norm)

    # Pool over all spatial dimensions and add dropout layer
    avgPool = GlobalAveragePooling(name = "poolingLayer")(model)
    if dropout_rate > 0:
        avgPoolDrop = Dropout(dropout_rate)(avgPool)
    else:
        avgPoolDrop = avgPool

    # Add new dense layer for class prediction
    finalModel = Dense(num_classes, activation=None, name="prediction") (avgPoolDrop)
    return finalModel


# Trains a transfer learning model 

def constant(value, dtype=None, shape=None, name=None):
    if dtype is None:
        dtype = floatx()
    if shape is None:
        shape = ()
    np_value = value * np.ones(shape)
    const = C.constant(np_value,
                       dtype=dtype,
                       name=_prepare_name(name, 'constant'))
    const._keras_shape = const.shape
    const._uses_learning_phase = False
    return const 

def gradients(loss, variables):
    # cntk does not support gradients as symbolic op,
    # to hook up with keras model
    # we will return a constant as place holder, the cntk learner will apply
    # the gradient during training.
    global grad_parameter_dict
    if isinstance(variables, list) is False:
        variables = [variables]
    grads = []
    for v in variables:
        g = C.constant(0, shape=v.shape, name='keras_grad_placeholder')
        grads.append(g)
        grad_parameter_dict[g] = v
    return grads 

def std_normalized_l2_loss(output, target):
    std_inv = np.array([6.6864805402, 5.2904440280, 3.7165409939, 4.1421640454, 8.1537399389, 7.0312877415, 2.6712380967,
                        2.6372177876, 8.4253649884, 6.7482162880, 9.0849960354, 10.2624412692, 3.1325531319, 3.1091179819,
                        2.7337937590, 2.7336441031, 4.3542467871, 5.4896293687, 6.2003761588, 3.1290341469, 5.7677042738,
                        11.5460919611, 9.9926451700, 5.4259818848, 20.5060642486, 4.7692101480, 3.1681517575, 3.8582905289,
                        3.4222250436, 4.6828286809, 3.0070785113, 2.8936539301, 4.0649030157, 25.3068458731, 6.0030623160,
                        3.1151977458, 7.7773542649, 6.2057372469, 9.9494258692, 4.6865422850, 5.3300697628, 2.7722027974,
                        4.0658663003, 18.1101618617, 3.5390113731, 2.7794520068], dtype=np.float32)
    weights = C.constant(value=std_inv) #.reshape((1, label_dim)))
    dif = output - target
    ret = C.reduce_mean(C.square(C.element_times(dif, weights)))
    return ret 

def lrelu(input, leak=0.2, name=""):
    return C.param_relu(C.constant((np.ones(input.shape)*leak).astype(np.float32)), input, name=name) 

def broadcast_xy(input_vec, h, w):
    """ broadcast input vector of length d to tensor (d x h x w) """
    assert(h > 0 and w > 0)
    d = input_vec.shape[0]
    # reshape vector to d x 1 x 1
    x = C.reshape(input_vec, (d, 1, 1))
    # create a zeros-like tensor of size (d x h x w)
    t = np.zeros((d, h, w), dtype=np.float32)
    y = C.constant(t)
    z = C.reconcile_dynamic_axes(y, x)
    z = z + x
    return z 

def create_model(base_model_file, input_features, num_classes,  dropout_rate = 0.5, freeze_weights = False):
    # Load the pretrained classification net and find nodes
    base_model   = load_model(base_model_file)
    feature_node = find_by_name(base_model, 'features')
    beforePooling_node = find_by_name(base_model, "z.x.x.r")
    #graph.plot(base_model, filename="base_model.pdf") # Write graph visualization

    # Clone model until right before the pooling layer, ie. until including z.x.x.r
    modelCloned = combine([beforePooling_node.owner]).clone(
        CloneMethod.freeze if freeze_weights else CloneMethod.clone,
        {feature_node: placeholder(name='features')})

    # Center the input around zero and set model input.
    # Do this early, to avoid CNTK bug with wrongly estimated layer shapes
    feat_norm = input_features - constant(114)
    model = modelCloned(feat_norm)

    # Pool over all spatial dimensions and add dropout layer
    avgPool = GlobalAveragePooling(name = "poolingLayer")(model)
    if dropout_rate > 0:
        avgPoolDrop = Dropout(dropout_rate)(avgPool)
    else:
        avgPoolDrop = avgPool

    # Add new dense layer for class prediction
    finalModel = Dense(num_classes, activation=None, name="prediction") (avgPoolDrop)
    return finalModel


# Trains a transfer learning model 

def constant(value, dtype=None, shape=None, name=None):
    if dtype is None:
        dtype = floatx()
    if shape is None:
        shape = ()
    np_value = value * np.ones(shape)
    const = C.constant(np_value,
                       dtype=dtype,
                       name=_prepare_name(name, 'constant'))
    const._keras_shape = const.shape
    const._uses_learning_phase = False
    return const 

def gradients(loss, variables):
    # cntk does not support gradients as symbolic op,
    # to hook up with keras model
    # we will return a constant as place holder, the cntk learner will apply
    # the gradient during training.
    global grad_parameter_dict
    if isinstance(variables, list) is False:
        variables = [variables]
    grads = []
    for v in variables:
        g = C.constant(0, shape=v.shape, name='keras_grad_placeholder')
        grads.append(g)
        grad_parameter_dict[g] = v
    return grads 

def constant(value, dtype=None, shape=None, name=None):
    if dtype is None:
        dtype = floatx()
    if shape is None:
        shape = ()
    np_value = value * np.ones(shape)
    const = C.constant(np_value,
                       dtype=dtype,
                       name=_prepare_name(name, 'constant'))
    const._keras_shape = const.shape
    const._uses_learning_phase = False
    return const