o ?e0[@s&dZddlZddlZddlmZddlmZddlmZddlm Z ddl m Z ddl m Z dd l mZdd l mZdd lmZdd lmZdd lmZddlmZddlmZddlmZddlmZddlmZddlmZ ddl!m"Z"dZ#Gddde j$Z%ddZ&ddZ'ddZ(ddZ)dS) zHome of the `Sequential` model.N)tf2)ops) tensor_util)layers) base_layer) functional) input_layer)training_utils)model_serialization) generic_utils) layer_utils) tf_inspect)tf_utils)module) np_arrays) tf_logging)base)nestzuAll layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.cseZdZdZejd(fdd ZefddZejddZ ejd d Z ej d)d d Z e j d)fd d Zd(fdd ZddZddZd*ddZd*ddZfddZed)ddZedd Zejd!d Zed"d#Zd$d%Zfd&d'ZZS)+ Sequentiala`Sequential` groups a linear stack of layers into a `tf.keras.Model`. `Sequential` provides training and inference features on this model. Examples: >>> # Optionally, the first layer can receive an `input_shape` argument: >>> model = tf.keras.Sequential() >>> model.add(tf.keras.layers.Dense(8, input_shape=(16,))) >>> # Afterwards, we do automatic shape inference: >>> model.add(tf.keras.layers.Dense(4)) >>> # This is identical to the following: >>> model = tf.keras.Sequential() >>> model.add(tf.keras.Input(shape=(16,))) >>> model.add(tf.keras.layers.Dense(8)) >>> # Note that you can also omit the `input_shape` argument. >>> # In that case the model doesn't have any weights until the first call >>> # to a training/evaluation method (since it isn't yet built): >>> model = tf.keras.Sequential() >>> model.add(tf.keras.layers.Dense(8)) >>> model.add(tf.keras.layers.Dense(4)) >>> # model.weights not created yet >>> # Whereas if you specify the input shape, the model gets built >>> # continuously as you are adding layers: >>> model = tf.keras.Sequential() >>> model.add(tf.keras.layers.Dense(8, input_shape=(16,))) >>> model.add(tf.keras.layers.Dense(4)) >>> len(model.weights) 4 >>> # When using the delayed-build pattern (no input shape specified), you can >>> # choose to manually build your model by calling >>> # `build(batch_input_shape)`: >>> model = tf.keras.Sequential() >>> model.add(tf.keras.layers.Dense(8)) >>> model.add(tf.keras.layers.Dense(4)) >>> model.build((None, 16)) >>> len(model.weights) 4 ```python # Note that when using the delayed-build pattern (no input shape specified), # the model gets built the first time you call `fit`, `eval`, or `predict`, # or the first time you call the model on some input data. model = tf.keras.Sequential() model.add(tf.keras.layers.Dense(8)) model.add(tf.keras.layers.Dense(1)) model.compile(optimizer='sgd', loss='mse') # This builds the model for the first time: model.fit(x, y, batch_size=32, epochs=10) ``` Ncsttj|j|ddd|_d|_d|_d|_d|_d|_ i|_ t |_ d|_ d|_|r@t|ttfs6|g}|D] }||q8dSdS)zCreates a `Sequential` model instance. Args: layers: Optional list of layers to add to the model. name: Optional name for the model. F)nameZautocastTN)superr Functional__init__Zsupports_maskingZ _compute_output_and_mask_jointlyZ_auto_track_sub_layers_inferred_input_shape_has_explicit_input_shapeZ _input_dtype_layer_call_argspecsset_created_nodes_graph_initialized_use_legacy_deferred_behavior isinstancelisttupleadd)selfrrlayer __class__j/home/www/facesmatcher.com/pyenv/lib/python3.10/site-packages/tensorflow/python/keras/engine/sequential.pyrgs(  zSequential.__init__cs8tt|j}|rt|dtjr|ddS|ddS)Nr)rrrr r InputLayer)r$rr&r(r)rs   zSequential.layersc Cst|dr|jd}t|tjr|}tdt|tjr)t|t j s(t |}nt dt|t|g||sDtd|jfd|_d}|dg|jst|tjr[d}nt|\}}|rutj|||jd d }||d}|rt|jd j}t|d krtt||_t !|jd|_"d|_d|_#n|jr||jd}tt|d krtt|g|_d|_|s|j$r|%|j"|jd|_$n |j&||'|gt()|j*|j+|<d S)aAdds a layer instance on top of the layer stack. Args: layer: layer instance. Raises: TypeError: If `layer` is not a layer instance. ValueError: In case the `layer` argument does not know its input shape. ValueError: In case the `layer` argument has multiple output tensors, or is already connected somewhere else (forbidden in `Sequential` models). _keras_historyrzPlease add `keras.layers.InputLayer` instead of `keras.Input` to Sequential model. `keras.Input` is intended to be used by Functional model.z;The added layer must be an instance of class Layer. Found: zAll layers added to a Sequential model should have unique names. Name "%s" is already the name of a layer in this model. Update the `name` argument to pass a unique name.F_self_tracked_trackablesT_input batch_shapedtyperr*N),hasattrr,r rr+loggingwarningrModulerZLayerrZ ModuleWrapper TypeErrorstrrZassert_no_legacy_layers_is_layer_name_unique ValueErrorrbuiltZ_maybe_create_attributer-r Zget_input_shape_and_dtypeInputrflatten_inbound_nodesoutputslenSINGLE_LAYER_OUTPUT_ERROR_MSGr Zget_source_inputsinputsrr_init_graph_networkappendZ#_handle_deferred_layer_dependenciesr getfullargspeccallr) r$r%Z origin_layerZ set_inputsr0r1xr?Z output_tensorr(r(r)r#sj               zSequential.addcCs|jstd|j}|j||js)d|_d|_d|_d|_d|_ d|_ dS|j rGg|jd_ |jdj g|_| |j|jd|_dSdS)znRemoves the last layer in the model. Raises: TypeError: if there are no layers in the model. z!There are no layers in the model.NFr2T)rr7r-poprr?rBr;rrr_outbound_nodesoutputrC)r$r%r(r(r)rHs"     zSequential.popc Csn|dus|js dStrtsdS|js|jst|}|jdur%|}nt |j|}|dur||jkrt jt j |||jdj dd}|}t}|jD]5}t||jz||}Wnd|_YWddStt|dkr{ttt|||}|} qO||_z ||| d|_Wnd|_YWdn1swY||_dSdSdSdSdS)Nrr.r/Tr*)rrenabledrZ#executing_eagerly_outside_functionsrrr"rrelax_input_shapeZ init_scoperr<rrclear_previously_created_nodesrr@rr=r:rA track_nodes_created_by_last_callrCr) r$ input_shapeZ input_dtypeZ new_shaperBZ layer_input created_nodesr%Z layer_outputr?r(r(r)'_build_graph_network_for_inferred_shape sX       #    6 z2Sequential._build_graph_network_for_inferred_shapecs`|jr ||j|jn|durtd|||js+t|}||_t t | |d|_dS)Nz+You must provide an `input_shape` argument.T) rrCrBr?r:rQr;r"_build_input_shaperrbuild)r$rOr&r(r)rSVs  zSequential.buildcs|js0t|s(t|tjs(d|_tt ||_ t r't dt||fn||j|j|jrI|js>||j|jtt|j|||dS|}|jD]7}i}|j|j}d|vr`||d<d|vrh||d<||fi|}tt|dkr}t t!|}t"|dd}qN|S)NTzLayers in a Sequential model should only have a single input tensor, but we receive a %s input: %s Consider rewriting this model with the Functional API.)trainingmaskrUrTr* _keras_mask)#rrZ is_tf_typer rZndarrayrrZ map_structure_get_shape_tuplerRrrKr4r5typerQshaper1rr;rCrBr?rrrFrrargsr@r=r:rAgetattr)r$rBrTrUr?r%kwargsZargspecr&r(r)rFds<    zSequential.callcCs|}|jD]}||}q|SN)rcompute_output_shape)r$rOrYr%r(r(r)r^s  zSequential.compute_output_shapecCs|j||d}t|ddS)N)rUrV)rFr[)r$rBrUr?r(r(r) compute_masks zSequential.compute_mask rcCs>td||||}|dks|dkrtd|S)ayGenerates class probability predictions for the input samples. The input samples are processed batch by batch. Args: x: input data, as a Numpy array or list of Numpy arrays (if the model has multiple inputs). batch_size: integer. verbose: verbosity mode, 0 or 1. Returns: A Numpy array of probability predictions. zq`model.predict_proba()` is deprecated and will be removed after 2021-01-01. Please use `model.predict()` instead.gg?zNetwork returning invalid probability values. The last layer might not normalize predictions into probabilities (like softmax or sigmoid would).)warningswarnpredictminmaxr4r5)r$rG batch_sizeverbosepredsr(r(r) predict_probas  zSequential.predict_probacCsBtd|j|||d}|jddkr|jddS|dkdS)afGenerate class predictions for the input samples. The input samples are processed batch by batch. Args: x: input data, as a Numpy array or list of Numpy arrays (if the model has multiple inputs). batch_size: integer. verbose: verbosity mode, 0 or 1. Returns: A numpy array of class predictions. a`model.predict_classes()` is deprecated and will be removed after 2021-01-01. Please use instead:* `np.argmax(model.predict(x), axis=-1)`, if your model does multi-class classification (e.g. if it uses a `softmax` last-layer activation).* `(model.predict(x) > 0.5).astype("int32")`, if your model does binary classification (e.g. if it uses a `sigmoid` last-layer activation).)rfrgr2r*)Zaxisg?Zint32)rarbrcrYZargmaxZastype)r$rGrfrgZprobar(r(r)predict_classess   zSequential.predict_classescsVg}tt|jD] }|t|q|jt|d}|j s)|j dur)|j |d<|S)N)rrbuild_input_shape) rrrrDr Zserialize_keras_objectrcopydeepcopyZ_is_graph_networkrR)r$ layer_configsr%configr&r(r) get_configs zSequential.get_configc Csd|vr|d}|d}|d}nd}d}|}||d}|D]}tj||d}||q|js?|r?t|ttfr?|||S)Nrrkr)r)custom_objects) get layer_moduleZ deserializer#rBr r"r!rS) clsrorqrrkrnmodelZ layer_configr%r(r(r) from_configs$       zSequential.from_configcCs6t|dr|jS|jrt|jddr|jdjSdS)N_manual_input_specr input_spec)r3rwrrxr$r(r(r)rxs  zSequential.input_speccCs ||_dSr])rw)r$valuer(r(r)rx cCs t|Sr])r ZSequentialSavedModelSaverryr(r(r)_trackable_saved_model_saverr{z'Sequential._trackable_saved_model_savercCs*|jD]}|j|jkr||urdSqdS)NFT)rr)r$r%Z ref_layerr(r(r)r9s z Sequential._is_layer_name_uniquecs|jrdSttj|dSr])rrrr_assert_weights_createdryr&r(r)r} sz"Sequential._assert_weights_created)NNr])r`r)__name__ __module__ __qualname____doc__ trackableZ no_automatic_dependency_trackingrpropertyrr#rHrQr defaultrSrFr^r_rirjrp classmethodrvrxsetterr|r9r} __classcell__r(r(r&r)r.s>8$ W  L ,       rcCs<t|dr|j}t|tr|S|jdurt|SdSdS)NrY)r3rYr r"Zrankas_list)trYr(r(r)rWs    rWcCs@|dus|dur dSt|t|krdStddt||DS)Ncss$|] \}}||kr dn|VqdSr]r().0Zd1Zd2r(r(r) $s"z$relax_input_shape..)r@r"zip)Zshape_1Zshape_2r(r(r)rLs rLcsR|jD]}|j}t|D] }fdd|jD|_q qfdd|jD|_dS)zARemove nodes from `created_nodes` from the layer's inbound_nodes.cg|]}|vr|qSr(r(rnrPr(r) ,s z2clear_previously_created_nodes..crr(r(rrr(r)r/s N)r>inbound_layersrr=rI)r%rPnode prev_layers prev_layerr(rr)rM's     rMcCsP|jsdS||jd|jdj}t|D] }|jr%||jdqdS)zFAdds to `created_nodes` the nodes created by the last call to `layer`.Nr2)r>r#rrr=rI)r%rPrrr(r(r)rN3s rN)*rrlraZtensorflow.pythonrZtensorflow.python.frameworkrrZtensorflow.python.kerasrrsZtensorflow.python.keras.enginerrrr Z*tensorflow.python.keras.saving.saved_modelr Ztensorflow.python.keras.utilsr r r rZtensorflow.python.modulerZtensorflow.python.ops.numpy_opsrZtensorflow.python.platformrr4Ztensorflow.python.trackablerrZtensorflow.python.utilrrArrrWrLrMrNr(r(r(r)s<                  i