o #j!@s|ddlZddlmZddlmZddlmZddlmZgZddd Z d d Z d d Z GdddZ dddZ dddZdS)N)_C_ops)check_variable_and_dtype) LayerHelper)in_dynamic_mode-q=cCst|jdkr d}tr$t||durdn||d\}}tj||gdSt|dddtd it }|j |j d }|j |j d }|j dd|i||d |durQdn||d d tj||gdS)NrFaxisX)Zfloat32Zfloat64norm l2_normalizedtype)ZOutZNorm)r epsilon)typeinputsZoutputsattrs)r ) lenshaperrr paddleZsqueezerrlocalsZ"create_variable_for_type_inferencerZ append_op)xr rnameoutr helperra/var/www/html/Deteccion_Ine/venv/lib/python3.10/site-packages/paddle/nn/utils/weight_norm_hook.pyl2_norms$  rcCs|j}t|}|dkrttt|dS|dkr,t||ddf}t|ddS||dkrBt|d|df}t|ddStt |}||d<d||<t ||}t |dSNrrrr ) rrrsqrtsumsquarereshaperlistrange transposenorm_except_dim)pdimrndimsp_matrixperm p_transposedrrrr(1s      r(c CsD|j}t|}|dkr|ttt|d}nu|dkr9t||ddf}tjjj |dd}t||}nW||dkrYt|d|df}tjjj |dd}t||}n7t t |}||d<d||<t ||}|j} t||jddf}tjjj |dd}t|| }t ||}tj jj|||dur|ndd} | Sr)rrrr!r"r#r$nnZ functional normalizer%r&r'ZtensormathZ_multiply_with_axis) vgr*rr+Z v_normalizedr,r-r.Ztransposed_shapeweightrrr _weight_normDs2      r5c@s8eZdZddZddZeddZddZd d Zd S) WeightNormcCs|durd}||_||_dS)Nr )rr*)selfrr*rrr__init__cs zWeightNorm.__init__cCs.t||jd}t||jd}t|||jS)N_g_v)getattrrr5r*)r7layerr3r2rrrcompute_weightiszWeightNorm.compute_weightc CsL|jD]\}}t|tr|j|krtd|q|dur!d}t|j|j}||kr3|d|ks7Jd|dkrA|||}t||}t ||}|j|=t ||}|j |j|j d} | |d| |j |j|j d} | |d| tt|| t|| Wdn1swYt|||||||S)Nzdim must set between [-R, R), R means the dimension of weight.rr:r9)_forward_pre_hooksitems isinstancer6r RuntimeErrorr _parametersrr;r(create_parameterr add_parameterrno_gradassignsetattrr=Zregister_forward_pre_hook) r<rr*khookZ weight_dimfnwZg_varr2r3rrrapplyns>       zWeightNorm.applycCs||}t||j|j|jd=|j|jd=|j|j|jd}||j|t t ||WddS1s?wYdS)Nr9r:r) r=delattrrrBrCrrrDrrErF)r7r<Zw_varrKrrrremoves   "zWeightNorm.removecCst||j||dS)N)rGrr=)r7r<rrrr__call__szWeightNorm.__call__N) __name__ __module__ __qualname__r8r= staticmethodrLrNrOrrrrr6bs % r6r4cCst||||S)a Applies weight normalization to a parameter according to the following formula: .. math:: \mathbf{w} = g \dfrac{v}{\|v\|} Weight normalization is a reparameterization of the weight vectors in a neural network that decouples the magnitude of those weight vectors from their direction. Weight normalization replaces the parameter specified by ``name`` (eg: 'weight') with two parameters: one parameter specifying the magnitude (eg: 'weight_g') and one parameter specifying the direction (eg: 'weight_v'). Weight normalization has been implemented as discussed in this paper: `Weight Normalization: A Simple Reparameterization to Accelerate Training of Deep Neural Networks `_. Parameters: layer(Layer): Layer of paddle, which has weight. name(str, optional): Name of the weight parameter. Default: 'weight'. dim(int, optional): Dimension over which to compute the norm. Dim is a non-negative number which is less than the rank of weight Tensor. For Example, dim can be chosen from 0, 1, 2, 3 for convolution whose weight shape is [cout, cin, kh, kw] and rank is 4. If dim is set to None, meaning that all elements will be normalized. Default: 0. Returns: Origin layer with weight norm hook. Examples: .. code-block:: python >>> from paddle.nn import Conv2D >>> from paddle.nn.utils import weight_norm >>> conv = Conv2D(3, 5, 3) >>> wn = weight_norm(conv) >>> print(conv.weight_g.shape) [5] >>> print(conv.weight_v.shape) [5, 3, 3, 3] )r6rL)r<rr*rrr weight_norms*rTcCsV|jD]\}}t|tr |j|kr |||j|=|Sqtd|d|)a remove weight normalization from layer. Parameters: layer(Layer): Layer of paddle, which has weight. name(str, optional): Name of the weight parameter. Default: 'weight'. Returns: Layer, the origin layer without weight norm Examples: .. code-block:: python >>> import paddle >>> from paddle.nn import Conv2D >>> from paddle.nn.utils import weight_norm, remove_weight_norm >>> paddle.seed(2023) >>> conv = Conv2D(3, 5, 3) >>> wn = weight_norm(conv) >>> print(conv.weight_g) Parameter containing: Tensor(shape=[5], dtype=float32, place=Place(cpu), stop_gradient=False, [1.35883713, 1.32126212, 1.56303072, 1.20874095, 1.22893476]) >>> remove_weight_norm(conv) >>> # The following is the effect after removing the weight norm: >>> # print(conv.weight_g) >>> # AttributeError: 'Conv2D' object has no attribute 'weight_g' zweight_norm of 'z' not found in )r>r?r@r6rrN ValueError)r<rrHrIrrrremove_weight_norms rV)rN)r4r)r4)rrZbase.data_feederrZbase.layer_helperrZ frameworkr__all__rr(r5r6rTrVrrrrs      @.