ó ¾÷Xc@spddlmZddlmZddlmZddlZddd„Z dd„Z d „Z d d „Z dS( iÿÿÿÿ(tprint_functioni(tconvert_kerneli(tbackendNc s›|jjdkrt}n;t}x2|jjƒD]!}t|ƒdkr1t}q1q1W|rÅ|ped}ˆpzdddg‰ˆddkr³gˆD]}t||ƒ^q”‰ndd d g}n–|pÎd }ˆpæd d ddg‰ˆddkrgˆD]}t||ƒ^q‰ndd d dg}g‰x!|jjƒD]}ˆ|7‰qGWd„‰td|ƒˆ|ˆƒtd|ƒ‡‡fd†}‡‡‡fd†}|j } xtt t| ƒƒD]`} |rí|| | ƒn|| | ƒ| t| ƒdkr"td|ƒqÐtd|ƒqÐWt | ddƒ\} } tdj | | ƒƒtdj | ƒƒtdj | ƒƒtd|ƒdS(sPrints a summary of a model. # Arguments model: Keras model instance. line_length: total length of printed lines positions: relative or absolute positions of log elements in each line. If not provided, defaults to `[.33, .55, .67, 1.]`. t SequentialiiAgÍÌÌÌÌÌÜ?g333333ë?gð?iÿÿÿÿs Layer (type)s Output ShapesParam #idg…ëQ¸Õ?gš™™™™™á?gq= ×£på?s Connected tocSsŒd}xutt|ƒƒD]a}|dkr<|d d}n|t||ƒ7}||| }|d||t|ƒ7}qWt|ƒdS(Ntiiÿÿÿÿt (trangetlentstrtprint(tfieldst positionstlineti((s6/tmp/pip-build-isqEY4/keras/keras/utils/layer_utils.pyt print_row+s  t_t=csny |j}Wntk r&d}nX|j}|jj}|d|d||jƒg}ˆ|ˆƒdS(Ntmultiples (t)(t output_shapetAttributeErrortnamet __class__t__name__t count_params(tlayerRRtcls_nameR (R R(s6/tmp/pip-build-isqEY4/keras/keras/utils/layer_utils.pytprint_layer_summary9s     !cs±y |j}Wntk r&d}nXg}xÆt|jƒD]µ\}}ˆr{|jdt|ƒ}|ˆkr{q=q{nxttt|jƒƒD]]}|j|j}|j |}|j |} |j |dt|ƒdt| ƒdƒq‘Wq=W|j} |j j } |sd} n |d} | d| d ||jƒ| g} ˆ| ˆƒt|ƒd kr­xCtd t|ƒƒD])}ddd||g} ˆ| ˆƒq}Wnd S( scPrints a summary for a single layer. # Arguments layer: target layer. Rs_ib-t[s][t]Ris (RiN(RRt enumeratet inbound_nodesRRRRtinbound_layerst node_indicesttensor_indicestappendRRR(RRt connectionst node_indextnodetnode_keyR t inbound_layertinbound_node_indextinbound_tensor_indexRRtfirst_connectionR (R Rtrelevant_nodes(s6/tmp/pip-build-isqEY4/keras/keras/utils/layer_utils.pyt$print_layer_summary_with_connectionsCs4       5    $ t layer_setsTotal params: {:,}sTrainable params: {:,}sNon-trainable params: {:,}N(RRtTruetnodes_by_depthtvaluesRtFalsetintR tlayersRtcount_total_paramstNonetformat( tmodelt line_lengthR tsequential_liketvtpt to_displayRR-R4R ttrainable_counttnon_trainable_count((R RR,s6/tmp/pip-build-isqEY4/keras/keras/utils/layer_utils.pyt print_summarysN    & &   $ cCs|dkrtƒ}nd}d}xÏ|D]Ç}||krCq+n|j|ƒt|dƒrŽt|j|ƒ\}}||7}||7}q+|tjg|jD]}t j |ƒ^q¡ƒ7}|tjg|j D]}t j |ƒ^qÓƒ7}q+W||fS(s/Counts the number of parameters in a list of layers. # Arguments layers: list of layers. layer_set: set of layers already seen (so that we don't count their weights twice). # Returns A tuple (count of trainable weights, count of non-trainable weights.) iR4N( R6tsettaddthasattrR5R4tnptsumttrainable_weightstKRtnon_trainable_weights(R4R.R>R?RtttntR<((s6/tmp/pip-build-isqEY4/keras/keras/utils/layer_utils.pyR5zs       26cCs†ddddh}g}xZ|jD]O}|jj|kr"tj|jƒ}t|ƒ}|j|j|fƒq"q"Wtj|ƒdS(s·Converts all convolution kernels in a model from Theano to TensorFlow. Also works from TensorFlow to Theano. # Arguments model: target model for the conversion. tConv1DtConv2DtConv3DtConv2DTransposeN( R4RRRGt get_valuetkernelRR#tbatch_set_value(R8t conv_classest to_assignRtoriginal_kerneltconverted_kernel((s6/tmp/pip-build-isqEY4/keras/keras/utils/layer_utils.pytconvert_all_kernels_in_model—s   tchannels_firstc Cs:|ddhkst‚|jƒ\}}xöt|jdƒD]á}|dkr¢|\}}}|||f} |dd…|fj| ƒ} tj| dƒ} nO|\}}}|||f} |dd…|fj| ƒ} tj| dƒ} tj| tj|ƒfƒ|dd…|fW|j||gƒdS( sbUtility useful when changing a convnet's `data_format`. When porting the weights of a convnet from one data format to the other, if the convnet includes a `Flatten` layer (applied to the last convolutional feature map) followed by a `Dense` layer, the weights of that `Dense` layer should be updated to reflect the new dimension ordering. # Arguments dense: The target `Dense` layer. previous_feature_map_shape: A shape tuple of 3 integers, e.g. `(512, 7, 7)`. The shape of the convolutional feature map right before the `Flatten` layer that came before the target `Dense` layer. target_data_format: One of "channels_last", "channels_first". Set it "channels_last" if converting a "chnnels_first" model to "channels_last", or reciprocally. t channels_lastRWiNii(iii(iii( tAssertionErrort get_weightsRtshapetreshapeRDt transposetprodt set_weights( tdensetprevious_feature_map_shapettarget_data_formatRPtbiasR tcthtwtoriginal_fm_shapetki((s6/tmp/pip-build-isqEY4/keras/keras/utils/layer_utils.pyt!convert_dense_weights_data_format°s 2(t __future__Rt conv_utilsRRRRGtnumpyRDR6R@R5RVRi(((s6/tmp/pip-build-isqEY4/keras/keras/utils/layer_utils.pyts r