˲j!adZddlZddlZddlmZddlZddlmZd/dZd/dZ d/dZ dZ d Z d/d Z d0d edededeejdef dZ d0d edededeejdef dZ d1d edededededeejdefdZd ededefdZd edefdZd edefdZdZd2dZd Z d3d ed!edeejdefd"Z d3d ed!edeejdefd#Zd$Z d4d eded'ed(edeejf d)Z d4d eded'ed(edeejf d*Z d5deejfd+Z d6deejfd-Z d.Z!e!eZ"e!eZ#e!eZ$e!eZ%e!eZ&e!eZ'e!eZ(e!eZ)e!eZ*e!eZ+e!e Z,dS)7zHThis file contains utilities for initializing neural network parameters.N)Optional)Tensorctj5||||cdddS#1swxYwYdSN generator)torchno_graduniform_tensorabrs >C:\Projects\Historical\venv\Lib\site-packages\torch/nn/init.py_no_grad_uniform_rs ::q!y99:::::::::::::::::: 9==ctj5||||cdddS#1swxYwYdSr)r r normal_r meanstdrs r_no_grad_normal_rs >>~~dC9~==>>>>>>>>>>>>>>>>>>rc"d}||d|zz ks ||d|zzkrtjddtj5|||z |z }|||z |z }|d|zdz d|zdz ||||tjdz| || |||cdddS#1swxYwYdS) Nc`dtj|tjdz zdz S)N?@)matherfsqrt)xs rnorm_cdfz(_no_grad_trunc_normal_..norm_cdfs)dhq49S>>1222c99zjmean is more than 2 std from [a, b] in nn.init.trunc_normal_. The distribution of values may be incorrect. stacklevelrr)minmax) warningswarnr r r erfinv_mul_rradd_clamp_) r rrrrrr!lus r_no_grad_trunc_normal_r1s::: q1s7{q1s7{ 2 2  ;     Ha$h#% & & Ha$h#% & & A 1q519 BBB   C$)C..())) D  ! ###+sB2DD Dctj5||cdddS#1swxYwYdSN)r r fill_r vals r_no_grad_fill_r7>s !!||C  !!!!!!!!!!!!!!!!!!s 6::ctj5|cdddS#1swxYwYdSr3)r r zero_r s r_no_grad_zero_r;Cs} ||~~s 599cgd}||vs|dkrdS|dkrdS|dkrtjdS|dkrw|d }nUt|tst|tst|t r|}nt d |d tjdd|d zzz S|dkr dSt d|)aReturn the recommended gain value for the given nonlinearity function. The values are as follows: ================= ==================================================== nonlinearity gain ================= ==================================================== Linear / Identity :math:`1` Conv{1,2,3}D :math:`1` Sigmoid :math:`1` Tanh :math:`\frac{5}{3}` ReLU :math:`\sqrt{2}` Leaky Relu :math:`\sqrt{\frac{2}{1 + \text{negative\_slope}^2}}` SELU :math:`\frac{3}{4}` ================= ==================================================== .. warning:: In order to implement `Self-Normalizing Neural Networks`_ , you should use ``nonlinearity='linear'`` instead of ``nonlinearity='selu'``. This gives the initial weights a variance of ``1 / N``, which is necessary to induce a stable fixed point in the forward pass. In contrast, the default gain for ``SELU`` sacrifices the normalization effect for more stable gradient flow in rectangular layers. Args: nonlinearity: the non-linear function (`nn.functional` name) param: optional parameter for the non-linear function Examples: >>> gain = nn.init.calculate_gain('leaky_relu', 0.2) # leaky_relu with negative_slope=0.2 .. _Self-Normalizing Neural Networks: https://papers.nips.cc/paper/2017/hash/5d44ee6f2c3f71b73125876103c8f6c4-Abstract.html )linearconv1dconv2dconv3dconv_transpose1dconv_transpose2dconv_transpose3dsigmoidr&tanhg?relur leaky_reluN{Gz?znegative_slope z not a valid numberr#selug?zUnsupported nonlinearity )rr isinstanceboolintfloat ValueError) nonlinearityparam linear_fnsnegative_slopes rcalculate_gainrSHsDJz!!\Y%>%>q   w   y~~  % % =!NN5$'' K5#&& K%'' K #NNIuIIIJJ JyNA$5 56777      C\CCDDDr"rr rrrreturnctj|r+tjt|f||||St ||||S)aFill the input Tensor with values drawn from the uniform distribution. :math:`\mathcal{U}(a, b)`. Args: tensor: an n-dimensional `torch.Tensor` a: the lower bound of the uniform distribution b: the upper bound of the uniform distribution generator: the torch Generator to sample from (default: None) Examples: >>> w = torch.empty(3, 5) >>> nn.init.uniform_(w) r )r overrideshas_torch_function_variadichandle_torch_functionr rr s rr r s`( 226:: 44 vi!qI5    VQ9 5 55r"rrctj|r+tjt|f||||St ||||S)aFill the input Tensor with values drawn from the normal distribution. :math:`\mathcal{N}(\text{mean}, \text{std}^2)`. Args: tensor: an n-dimensional `torch.Tensor` mean: the mean of the normal distribution std: the standard deviation of the normal distribution generator: the torch Generator to sample from (default: None) Examples: >>> w = torch.empty(3, 5) >>> nn.init.normal_(w) r)r rWrXrYrrrs rrrs`( 226:: 44 fYvDcY5    FD#y 9 99r"rc,t||||||S)aFill the input Tensor with values drawn from a truncated normal distribution. The values are effectively drawn from the normal distribution :math:`\mathcal{N}(\text{mean}, \text{std}^2)` with values outside :math:`[a, b]` redrawn until they are within the bounds. The method used for generating the random values works best when :math:`a \leq \text{mean} \leq b`. Args: tensor: an n-dimensional `torch.Tensor` mean: the mean of the normal distribution std: the standard deviation of the normal distribution a: the minimum cutoff value b: the maximum cutoff value generator: the torch Generator to sample from (default: None) Examples: >>> w = torch.empty(3, 5) >>> nn.init.trunc_normal_(w) r)r1)r rrrrrs r trunc_normal_r]s8 "&$QY O O OOr"r6ctj|r)tjt|f||St ||S)zFill the input Tensor with the value :math:`\text{val}`. Args: tensor: an n-dimensional `torch.Tensor` val: the value to fill the tensor with Examples: >>> w = torch.empty(3, 5) >>> nn.init.constant_(w, 0.3) r5)r rWrXrY constant_r7r5s rr_r_sX 226:: 44 yS5    &# & &&r"c"t|dS)zFill the input Tensor with the scalar value `1`. Args: tensor: an n-dimensional `torch.Tensor` Examples: >>> w = torch.empty(3, 5) >>> nn.init.ones_(w) r)r7r:s rones_ras &# & &&r"c t|S)zFill the input Tensor with the scalar value `0`. Args: tensor: an n-dimensional `torch.Tensor` Examples: >>> w = torch.empty(3, 5) >>> nn.init.zeros_(w) )r;r:s rzeros_rcs & ! !!r"c|dkrtdtj5tj|j||jddddn #1swxYwY|S)a=Fill the 2-dimensional input `Tensor` with the identity matrix. Preserves the identity of the inputs in `Linear` layers, where as many inputs are preserved as possible. Args: tensor: a 2-dimensional `torch.Tensor` Examples: >>> w = torch.empty(3, 5) >>> nn.init.eye_(w) r#,Only tensors with 2 dimensions are supported)out requires_gradN) ndimensionrNr r eyeshapergr:s reye_rksaGHHH QQ 61, each group of channels preserves identity Args: tensor: a {3, 4, 5}-dimensional `torch.Tensor` groups (int, optional): number of groups in the conv layer (default: 1) Examples: >>> w = torch.empty(3, 16, 5, 5) >>> nn.init.dirac_(w) >>> w = torch.empty(3, 24, 5, 5) >>> nn.init.dirac_(w, 3) )z5Only tensors with 3, 4, or 5 dimensions are supportedrz!dim 0 must be divisible by groupsr&rmr#rnN)rhrNsizer'r r r9range)r groups dimensionssizesout_chans_per_grpmin_dimgds rdirac_ry$s ""$$J""PQQQ KKMME Qx&A<===aF*#U1X..G  v  A7^^  ??PQF10014aQ19LLMM1__  --1 A!+ A!+- --1 A!+ A!+ A!+ -  , MsC8FFFc&|}|dkrtd|d}|d}d}|dkr|jddD]}||z}||z}||z}||fS)Nr#zNFan in and fan out can not be computed for tensor with fewer than 2 dimensionsr&r)dimrNrprj)r rsnum_input_fmapsnum_output_fmapsreceptive_field_sizesfan_infan_outs r_calculate_fan_in_and_fan_outrYsJA~~ \   kk!nnO{{1~~ zz||aabb! & &A A % 3 3F!55G 7?r"gainct|\}}|tjdt||zz z}tjd|z}t || ||S)aFill the input `Tensor` with values using a Xavier uniform distribution. The method is described in `Understanding the difficulty of training deep feedforward neural networks` - Glorot, X. & Bengio, Y. (2010). The resulting tensor will have values sampled from :math:`\mathcal{U}(-a, a)` where .. math:: a = \text{gain} \times \sqrt{\frac{6}{\text{fan\_in} + \text{fan\_out}}} Also known as Glorot initialization. Args: tensor: an n-dimensional `torch.Tensor` gain: an optional scaling factor generator: the torch Generator to sample from (default: None) Examples: >>> w = torch.empty(3, 5) >>> nn.init.xavier_uniform_(w, gain=nn.init.calculate_gain('relu')) Note: Be aware that ``fan_in`` and ``fan_out`` are calculated assuming that the weight matrix is used in a transposed manner, (i.e., ``x @ w.T`` in ``Linear`` layers, where ``w.shape = [fan_out, fan_in]``). This is important for correct initialization. If you plan to use ``x @ w``, where ``w.shape = [fan_in, fan_out]``, pass in a transposed weight matrix, i.e. ``nn.init.xavier_uniform_(w.T, ...)``. r@)rrrrMr)r rrrrrrs rxavier_uniform_rnsdD4F;;OFG 3v'7!8!8899 9C #A VaRI 6 66r"ct|\}}|tjdt||zz z}t |d||S)aFill the input `Tensor` with values using a Xavier normal distribution. The method is described in `Understanding the difficulty of training deep feedforward neural networks` - Glorot, X. & Bengio, Y. (2010). The resulting tensor will have values sampled from :math:`\mathcal{N}(0, \text{std}^2)` where .. math:: \text{std} = \text{gain} \times \sqrt{\frac{2}{\text{fan\_in} + \text{fan\_out}}} Also known as Glorot initialization. Args: tensor: an n-dimensional `torch.Tensor` gain: an optional scaling factor generator: the torch Generator to sample from (default: None) Examples: >>> w = torch.empty(3, 5) >>> nn.init.xavier_normal_(w) Note: Be aware that ``fan_in`` and ``fan_out`` are calculated assuming that the weight matrix is used in a transposed manner, (i.e., ``x @ w.T`` in ``Linear`` layers, where ``w.shape = [fan_out, fan_in]``). This is important for correct initialization. If you plan to use ``x @ w``, where ``w.shape = [fan_in, fan_out]``, pass in a transposed weight matrix, i.e. ``nn.init.xavier_normal_(w.T, ...)``. rrT)rrrrMr)r rrrrrs rxavier_normal_rsPB4F;;OFG 3v'7!8!8899 9C FCi 8 88r"c|}ddg}||vrtd|d|t|\}}|dkr|n|S)NrrzMode z" not supported, please use one of )lowerrNr)r mode valid_modesrrs r_calculate_correct_fanrsh ::<>> w = torch.empty(3, 5) >>> nn.init.kaiming_uniform_(w, mode='fan_in', nonlinearity='relu') Note: Be aware that ``fan_in`` and ``fan_out`` are calculated assuming that the weight matrix is used in a transposed manner, (i.e., ``x @ w.T`` in ``Linear`` layers, where ``w.shape = [fan_out, fan_in]``). This is important for correct initialization. If you plan to use ``x @ w``, where ``w.shape = [fan_in, fan_out]``, pass in a transposed weight matrix, i.e. ``nn.init.kaiming_uniform_(w.T, ...)``. )r rrrOrr,Initializing zero-element tensors is a no-oprrN)r rWrXrYkaiming_uniform_rjr)r*rrSrrr r ) r rrrOrfanrrbounds rrrsUV 226::  44  I%5    FL DEEE  . .C , * *D 3 C IcNNS E CCvu BBCCCCCCCCCCCCCCCCCCs C22C69C6c8d|jvrtjd|St||}t ||}|t j|z }tj5| d||cdddS#1swxYwYdS)aFill the input `Tensor` with values using a Kaiming normal distribution. The method is described in `Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification` - He, K. et al. (2015). The resulting tensor will have values sampled from :math:`\mathcal{N}(0, \text{std}^2)` where .. math:: \text{std} = \frac{\text{gain}}{\sqrt{\text{fan\_mode}}} Also known as He initialization. Args: tensor: an n-dimensional `torch.Tensor` a: the negative slope of the rectifier used after this layer (only used with ``'leaky_relu'``) mode: either ``'fan_in'`` (default) or ``'fan_out'``. Choosing ``'fan_in'`` preserves the magnitude of the variance of the weights in the forward pass. Choosing ``'fan_out'`` preserves the magnitudes in the backwards pass. nonlinearity: the non-linear function (`nn.functional` name), recommended to use only with ``'relu'`` or ``'leaky_relu'`` (default). generator: the torch Generator to sample from (default: None) Examples: >>> w = torch.empty(3, 5) >>> nn.init.kaiming_normal_(w, mode='fan_out', nonlinearity='relu') Note: Be aware that ``fan_in`` and ``fan_out`` are calculated assuming that the weight matrix is used in a transposed manner, (i.e., ``x @ w.T`` in ``Linear`` layers, where ``w.shape = [fan_out, fan_in]``). This is important for correct initialization. If you plan to use ``x @ w``, where ``w.shape = [fan_in, fan_out]``, pass in a transposed weight matrix, i.e. ``nn.init.kaiming_normal_(w.T, ...)``. rrrN) rjr)r*rrSrrr r r)r rrrOrrrrs rkaiming_normal_r sV FL DEEE  . .C , * *D 3 C ;;~~a ~::;;;;;;;;;;;;;;;;;;s*BBBc|dkrtd|dkr|S|d}||z}|||fdd|}||kr|tj |\}}tj |d}| } || z}||kr|tj 5| ||||dddn #1swxYwY|S)aFill the input `Tensor` with a (semi) orthogonal matrix. Described in `Exact solutions to the nonlinear dynamics of learning in deep linear neural networks` - Saxe, A. et al. (2013). The input tensor must have at least 2 dimensions, and for tensors with more than 2 dimensions the trailing dimensions are flattened. Args: tensor: an n-dimensional `torch.Tensor`, where :math:`n \geq 2` gain: optional scaling factor generator: the torch Generator to sample from (default: None) Examples: >>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_LAPACK) >>> w = torch.empty(3, 5) >>> nn.init.orthogonal_(w) r#z4Only tensors with 2 or more dimensions are supportedrr&rN)rhrNnumelrp new_emptyrt_r linalgqrdiagsignr view_ascopy_r,) r rrrowscols flattenedqrrxphs r orthogonal_r>s,QOPPP ||~~ ;;q>>D <<>>T !D  $..66q!y6QQI d{{  E<<FFrHc|dkrtd|j\}}tt j||z}t j5|d||t|D]'}t j |}|d|} d|| |f<( dddn #1swxYwY|S)aFill the 2D input `Tensor` as a sparse matrix. The non-zero elements will be drawn from the normal distribution :math:`\mathcal{N}(0, 0.01)`, as described in `Deep learning via Hessian-free optimization` - Martens, J. (2010). Args: tensor: an n-dimensional `torch.Tensor` sparsity: The fraction of elements in each column to be set to zero std: the standard deviation of the normal distribution used to generate the non-zero values generator: the torch Generator to sample from (default: None) Examples: >>> w = torch.empty(3, 5) >>> nn.init.sparse_(w, sparsity=0.1) r#rerrN) rhrNrjrLrceilr r rrqrandperm) r sparsityrrrr num_zeroscol_idx row_indices zero_indicess rsparse_rqs.aGHHHJD$DIho..//I ..q#333T{{ . .G...K&z z2L,-F<( ) ) ................ Ms)ACC  C cljddfd}dddd|_|_|S)NcZtjdddtd|i|S)Nz `nn.init.z)` is now deprecated in favor of `nn.init.z`.r#r$)r)r* FutureWarning)argskwargsmethnew_nameold_names rdeprecated_initz(_make_deprecate..deprecated_initsO W W W8 W W W     tT$V$$$r"z z_(...) .. warning:: This method is now deprecated in favor of :func:`torch.nn.init.z"`. See :func:`~torch.nn.init.z` for details.)__name____doc__)rrrrs` @@r_make_deprecaters}H}H%%%%%%%: ::IQ :: ( :::O (O r"r3)rTrN)rTrr[rN)r&)rN)rrrGN)r&N)rHN)-rrr)typingr _Optionalr rrrr1r7r;rSrM Generatorr rr]r_rarcrkryrrrrstrrrrrruniformnormalconstantridiracxavier_uniform xavier_normalkaiming_uniformkaiming_normal orthogonalsparser"rrs!NN (((((( :::: >>>> """"J!!!  CECECECEP,0 66 6 6 6) 6  6666:,0 :: : : :) :  :::::,0 PP P P P P  P ) P PPPP>'f'5'V''''$ '& 'V ' ' ' ' "6 "f " " " "*2222j.,0&7&7 &7 &7)&7 &7&7&7&7V,0$9$9 $9 $9)$9 $9$9$9$9N333$,0 >C>C >C >C >C >C ) >C>C>C>CF$,0 2;2; 2; 2; 2; 2; ) 2;2;2;2;n ,000)0000l ,0 ##) ####N. /( # #  ! ! ?9 % %od 11// !/"233 11 _[ ) )  ! !r"