Xc@sddlZddlZddlZddlmZmZddlmZ ddl m Z m Z ddlm Z ddlmZddlmZdd lmZd efd YZd ddeeeed ZdefdYZdefdYZdZdZdS(iNi(tLayert InputSpec(tbackend(t func_dumpt func_load(t regularizers(t constraints(t activations(t initializerstMergec BsweZdZd dddd d d d d d d ZdZd dZdZd dZdZ e d Z RS( sA `Merge` layer can be used to merge a list of tensors into a single tensor, following some merge `mode`. # Example ```python model1 = Sequential() model1.add(Dense(32, input_dim=32)) model2 = Sequential() model2.add(Dense(32, input_dim=32)) merged_model = Sequential() merged_model.add(Merge([model1, model2], mode='concat', concat_axis=1)) ``` # Arguments layers: Can be a list of Keras tensors or a list of layer instances. Must be more than one layer/tensor. mode: String or lambda/function. If string, must be one of: 'sum', 'mul', 'concat', 'ave', 'cos', 'dot', 'max'. If lambda/function, it should take as input a list of tensors and return a single tensor. concat_axis: Integer, axis to use in mode `concat`. dot_axes: Integer or tuple of integers, axes to use in mode `dot` or `cos`. output_shape: Either a shape tuple (tuple of integers), or a lambda/function to compute `output_shape` (only if merge mode is a lambda/function). If the argument is a tuple, it should be expected output shape, *not* including the batch size (same convention as the `input_shape` argument in layers). If the argument is callable, it should take as input a list of shape tuples (1:1 mapping to input tensors) and return a single shape tuple, including the batch size (same convention as the `compute_output_shape` method of layers). node_indices: Optional list of integers containing the output node index for each input layer (in case some input layers have multiple output nodes). will default to an array of 0s if not provided. tensor_indices: Optional list of indices of output tensors to consider for merging (in case some input layer node returns multiple tensors). output_mask: Mask or lambda/function to compute the output mask (only if merge mode is a lambda/function). If the latter case, it should take as input a list of masks and return a single mask. tsumic Cstjddd||_||_||_||_||_||_||_|r^|ni|_ d|_ g|_ g|_ i|_g|_g|_t|_t|_d|_| s|jjj} | dttj| } n| |_|r|s-gtt|D] } d^q}n| s[gtt|D] } d^qF} n|j|||||| t|_ g} g}xct!|D]U\}}||}| |}|j |}| j"|j#||j"|j$|qW|| d|n t|_ dS(NsThe `Merge` layer is deprecated and will be removed after 08/2017. Use instead layers from `keras.layers.merge`, e.g. `add`, `concatenate`, etc.t stacklevelit_itmask(%twarningstwarntlayerstmodet concat_axistdot_axest _output_shapet node_indicest _output_maskt argumentstNonet_initial_weightst inbound_nodestoutbound_nodesRt_trainable_weightst_non_trainable_weightstTruetsupports_maskingtFalsetuses_learning_phaset input_spect __class__t__name__tlowertstrtKtget_uidtnametrangetlent_arguments_validationtbuiltt enumeratetappendtoutput_tensorst output_masks(tselfRRRRt output_shapet output_maskRRttensor_indicesR)tprefixR t input_tensorst input_maskstitlayert node_indext tensor_indext inbound_node((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyt__init__>sT                   ((      cCst|sI|dddddddhkrItdt|qInt|ttf sqt|d krtd t|n|dkrgt t|D] }d^q}ng}xZt |D]L\} } | j || } t| tr| || } n|j | qW|dddddhkr|t |} t| d kr|td |d d|q|n|ddhkrt|d krt|dn|d} |d }t| }t|}t|tr'|dkr||||g|_q'|gd |_nt|jttfsNtdnt|jd krrtdnt|jdt st|jd t rtdn| |jd||jd krtd| |jd||jd fd| |fqn|dkrg|D]}t|^q$}t }x?t t|D]+} || |j=|jt|| qXWt|d krtd|qndS(sVValidates user-passed arguments and raises exceptions as appropriate. R tmultconcattavetcostdottmaxsInvalid merge mode: isTA Merge should only be applied to a list of layers with at least 2 elements. Found: is5Only layers of same output shape can be merged using s mode. sLayer shapes: %ss merge takes exactly 2 layersis-Invalid type for dot_axes - should be a list.s:Invalid format for dot_axes - should contain two elements.s<Invalid format for dot_axes - list elements should be "int".s4Dimension incompatibility using dot mode: %s != %s. sLayer shapes: %s, %ssl"concat" mode can only merge layers with matching output shapes except for the concat axis. Layer shapes: %sN(tcallablet ValueErrorR&t isinstancetlistttupleR+t TypeErrorRR*R.tget_output_shape_atR/tsettintRRtadd(R2RRRRRR5R t input_shapesR9R:tlayer_output_shapetinput_shapes_settshape1tshape2tn1tn2tshapetreduced_inputs_shapest shape_set((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyR,ysd !( (       ."   c Cst|t s"t|dkr;tdt|nt|jr|j}tj |j}d|j kr||dR,RjRnR~Rt classmethodR(((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyR s.    8 D 3 + % *R cCs:tjdddt}x'|D]} t| ds t}Pq q W|rg} g} g} xG|D]?} | j\} }}| j| | j|| j|qbWt| d|d|d|d|d |d |d | d | d | }|jdj dStd|d|d|d|d |d |d |}||SdS(sFunctional merge, to apply to Keras tensors (NOT layers). Returns a Keras tensor. # Example ```python tensor_a = Input(shape=(32,)) tensor_b = Input(shape=(32,)) merged_tensor = merge([tensor_a, tensor_b], mode='concat', concat_axis=1) ``` # Arguments mode: String or lambda/function. If string, must be one of: 'sum', 'mul', 'concat', 'ave', 'cos', 'dot', 'max'. If lambda/function, it should take as input a list of tensors and return a single tensor. concat_axis: Integer, axis to use in mode `concat`. dot_axes: Integer or tuple of integers, axes to use in mode `dot` or `cos`. output_shape: Shape tuple (tuple of integers), or lambda/function to compute output_shape (only if merge mode is a lambda/function). If the latter case, it should take as input a list of shape tuples (1:1 mapping to input tensors) and return a single shape tuple, including the batch size (same convention as the `compute_output_shape` method of layers). node_indices: Optional list of integers containing the output node index for each input layer (in case some input layers have multiple output nodes). will default to an array of 0s if not provided. tensor_indices: Optional list of indices of output tensors to consider for merging (in case some input layer node returns multiple tensors). sThe `merge` function is deprecated and will be removed after 08/2017. Use instead layers from `keras.layers.merge`, e.g. `add`, `concatenate`, etc.R it_keras_historyRRRR3R4RRR5R)iN( RRRRsR RR/R RR0(RcRRRR3R4RR)tall_keras_tensorstxt input_layersRR5t input_layerR;R<t merge_layer((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pytmergesD!        t MaxoutDensec BsYeZdZddddddddedd ZdZdZdZdZ RS( sA dense maxout layer. A `MaxoutDense` layer takes the element-wise maximum of `nb_feature` `Dense(input_dim, output_dim)` linear layers. This allows the layer to learn a convex, piecewise linear activation function over the inputs. Note that this is a *linear* layer; if you wish to apply activation function (you shouldn't need to --they are universal function approximators), an `Activation` layer must be added after. # Arguments output_dim: int > 0. nb_feature: number of Dense layers to use internally. init: name of initialization function for the weights of the layer (see [initializations](../initializations.md)), or alternatively, Theano function to use for weights initialization. This parameter is only relevant if you don't pass a `weights` argument. weights: list of Numpy arrays to set as initial weights. The list should have 2 elements, of shape `(input_dim, output_dim)` and (output_dim,) for weights and biases respectively. W_regularizer: instance of [WeightRegularizer](../regularizers.md) (eg. L1 or L2 regularization), applied to the main weights matrix. b_regularizer: instance of [WeightRegularizer](../regularizers.md), applied to the bias. activity_regularizer: instance of [ActivityRegularizer](../regularizers.md), applied to the network output. W_constraint: instance of the [constraints](../constraints.md) module (eg. maxnorm, nonneg), applied to the main weights matrix. b_constraint: instance of the [constraints](../constraints.md) module, applied to the bias. bias: whether to include a bias (i.e. make the layer affine rather than linear). input_dim: dimensionality of the input (integer). This argument (or alternatively, the keyword argument `input_shape`) is required when using this layer as the first layer in a model. # Input shape 2D tensor with shape: `(nb_samples, input_dim)`. # Output shape 2D tensor with shape: `(nb_samples, output_dim)`. # References - [Maxout Networks](http://arxiv.org/abs/1302.4389) itglorot_uniformc Kstjd||_||_tj||_tj||_tj||_ tj||_ t j||_ t j| |_ | |_||_tdd|_| |_|jr|jf| d( R2RRRtweightsRRRRRRRtkwargs((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyR>s       c Cs|d}tdtjdd|f|_|j|j||jfd|jddd|j d|j |_ |j r|j|j|jfdd dd d|j d|j|_n d|_|jdk r|j|j|`nt|_dS( NitdtypeRVt initializerR)tWt regularizert constrainttzerotb(RR'tfloatxRR"t add_weightRRRRRRRRRRRt set_weightsRR-(R2RmR((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pytbuild#s&       cCs/|rt|dkst|d|jfS(Nii(R+RkR(R2Rm((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyRn;scCsGtj||j}|jr.||j7}ntj|dd}|S(NRYi(R'RCRRRRD(R2RRh((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyRj?s  cCsi |jd6tj|jd6|jd6tj|jd6tj|jd6tj|jd6t j|j d6t j|j d6|j d 6|j d 6}tt|j}tt|jt|jS( NRRRRRRRRRR(RRt serializeRRRRRRRRRRRRRRtdictRHtitems(R2Rt base_config((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyRGs    N( R$RRRRR>RRnRjR(((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyRs*    tHighwayc BsPeZdZddddddddedd ZdZdZdZRS(s\Densely connected highway network. Highway layers are a natural extension of LSTMs to feedforward networks. # Arguments init: name of initialization function for the weights of the layer (see [initializations](../initializations.md)), or alternatively, Theano function to use for weights initialization. This parameter is only relevant if you don't pass a `weights` argument. activation: name of activation function to use (see [activations](../activations.md)), or alternatively, elementwise Theano function. If you don't specify anything, no activation is applied (ie. "linear" activation: a(x) = x). weights: list of Numpy arrays to set as initial weights. The list should have 2 elements, of shape `(input_dim, output_dim)` and (output_dim,) for weights and biases respectively. W_regularizer: instance of [WeightRegularizer](../regularizers.md) (eg. L1 or L2 regularization), applied to the main weights matrix. b_regularizer: instance of [WeightRegularizer](../regularizers.md), applied to the bias. activity_regularizer: instance of [ActivityRegularizer](../regularizers.md), applied to the network output. W_constraint: instance of the [constraints](../constraints.md) module (eg. maxnorm, nonneg), applied to the main weights matrix. b_constraint: instance of the [constraints](../constraints.md) module, applied to the bias. bias: whether to include a bias (i.e. make the layer affine rather than linear). input_dim: dimensionality of the input (integer). This argument (or alternatively, the keyword argument `input_shape`) is required when using this layer as the first layer in a model. # Input shape 2D tensor with shape: `(nb_samples, input_dim)`. # Output shape 2D tensor with shape: `(nb_samples, input_dim)`. # References - [Highway Networks](http://arxiv.org/abs/1505.00387v2) Rc Kstjdd| kr6| jdtjdntj||_tj||_tj||_ tj||_ tj||_ t j||_ t j||_| |_||_tdd|_| |_|jr|jf| d( R2RRRRRRRRRRR((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyR>~s$       c Cs,|d}tdtjdd|f|_|j||fd|jddd|jd|j|_ |j||fd|jdd |_ |j r|j|fdd dd d|j d|j |_|j|fdd dd |_n d|_|jdk r|j|j|`nt|_dS(NiRRVRR)RRRtW_carryRRtonetb_carry(RR'RRR"RRRRRRRRRRRRRRR-(R2RmR((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyRs2         cCstj||j}|jr.||j7}ntj|}tj||j}|jrk||j7}n|j |}||9}|d||}|S(Ni( R'RCRRRRtsigmoidRRR(R2Rtyttransform_weighttactRh((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyRjs   cCsi tj|jd6tj|jd6tj|jd6tj|jd6tj|jd6t j|j d6t j|j d6|j d6|j d 6}tt|j}tt|jt|jS( NRRRRRRRRR(RRRRRRRRRRRRRRRRRRRHR(R2RR((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyRs  N( R$RRRRR>RRjR(((s2/tmp/pip-build-isqEY4/keras/keras/legacy/layers.pyRVs&   cOsXddlm}d|kr.|jd}nd}||ds$    F