ó àÆ÷Xc@`sÓddlmZmZmZddlZddlZddlmZmZddlm Z ddl m Z defd„ƒYZ defd „ƒYZ e eƒZe eƒZe eƒZe eƒZd „ZdS( i(tabsolute_importtprint_functiontdivisionN(tOptApply(tdiscrete_dtypes(tgrad_undefinedtSparseBlockGemvcB`sJeZdZdZgZed„Zd„Zd„Zd„Z d„Z RS(sÔ This op computes the dot product of specified pieces of vectors and matrices, returning pieces of vectors:: for b in range(batch_size): for j in range(o.shape[1]): for i in range(h.shape[1]): o[b, j, :] += numpy.dot(h[b, i], W[iIdx[b, i], oIdx[b, j]]) where b, h, W, o iIdx, oIdx are defined in the docstring of make_node. .. image:: ../../../images/blocksparse.png :scale: 50 % tinplacecC`s,||_|jr(idgd6|_ndS(Ni(Rt destroy_map(tselfR((s>/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pyt__init__s  cC`sHtjj|ƒ}tjj|ƒ}tjj|ƒ}tjj|ƒ}tjj|ƒ}|jdkrxtdƒ‚n|jdkr–tdƒ‚n|jdkr´tdƒ‚n|jdkrÒtdƒ‚n|jdkrðtdƒ‚n|jjtkst‚|jjtks t‚t ||||||g|jƒgƒS( sB Compute the dot product of the specified pieces of vectors and matrices. The parameter types are actually their expected shapes relative to each other. Parameters ---------- o : batch, oWin, oSize output vector W : iBlocks, oBlocks, iSize, oSize weight matrix h : batch, iWin, iSize input from lower layer (sparse) inputIdx : batch, iWin indexes of the input blocks outputIdx : batch, oWin indexes of the output blocks Returns ------- (batch, oWin, oSize) dot(W[i, j], h[i]) + o[j] Notes ----- - `batch` is the number of examples in a minibatch (batch size). - `iBlocks` is the total number of blocks in the input (from lower layer). - `iSize` is the size of each of these input blocks. - `iWin` is the number of blocks that will be used as inputs. Which blocks will be used is specified in `inputIdx`. - `oBlocks` is the number or possible output blocks. - `oSize` is the size of each of these output blocks. - `oWin` is the number of output blocks that will actually be computed. Which blocks will be computed is specified in `outputIdx`. is The output o must be a 2D tensoris'The weight matrix W must be a 4D tensorsThe input h must be a 3D tensoris.The input indices inputIdx must be a 2D tensors0The output indices outputIdx must be a 2D tensor( ttheanottensortas_tensor_variabletndimt TypeErrorttypetdtypeRtAssertionErrorR(R totWthtinputIdxt outputIdx((s>/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pyt make_node#s$(c C`s|d \}}}}}|js1|jƒ}nx¼t|jdƒD]§} xžt|jdƒD]‰} || | f} xpt|jdƒD][} || | f} || | f}|| | dd…fctj|| | f|ƒ7/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pytperformas AcC`s |dgS(Ni((R Rt input_shapes((s>/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pyt infer_shapepsc C`s³|\}}}}}|d}t|jƒ} t|jƒ} | |jƒ||||ƒ} | |jƒ|jdƒ|||ƒ} || | t|d|dƒt|d|dƒgS( Niiiisgrad of inputIdx makes no senseis grad of outputIdx makes no sense(iiii(tSparseBlockOuterRRt zeros_liket dimshuffleR( R tinputstgradsRRRRRtgot outer_funtgemv_funtWgradthgrad((s>/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pytgradss    (sinplace( t__name__t __module__t__doc__t __props__tregistered_optstFalseR RR(R*R5(((s>/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pyR s  >  R+cB`sDeZdZdZgZed„Zdd„Zd„Z d„Z RS(s& This computes the outer product of two sets of pieces of vectors updating a full matrix with the results:: for b in range(batch_size): o[xIdx[b, i], yIdx[b, j]] += (alpha * outer(x[b, i], y[b, j])) This op is involved in the gradient of SparseBlockGemv. RcC`s,||_|jr(idgd6|_ndS(Ni(RR (R R((s>/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pyR “s  cC`s—tjjtjdddƒƒ}tjj|ƒ}tjj|ƒ}tjj|ƒ}|dkrl|}nt|||||||g|jƒgƒS(s“ Compute the dot product of the specified pieces of vectors and matrices. The parameter types are actually their expected shapes relative to each other. Parameters ---------- o : xBlocks, yBlocks, xSize, ySize x : batch, xWin, xSize y : batch, yWin, ySize xIdx : batch, iWin indexes of the x blocks yIdx : batch, oWin indexes of the y blocks Returns ------- (xBlocks, yBlocks, xSize, ySize) outer(x[i], y[j]) + o[i, j] Notes ----- - `batch` is the number of examples in a minibatch (batch size). - `xBlocks` is the total number of blocks in x. - `xSize` is the size of each of these x blocks. - `xWin` is the number of blocks that will be used as x. Which blocks will be used is specified in `xIdx`. - `yBlocks` is the number or possible y blocks. - `ySize` is the size of each of these y blocks. - `yWin` is the number of y blocks that will actually be computed. Which blocks will be computed is specified in `yIdx`. gð?Rtfloat32N( R R tconstantRtasarrayRtNoneRR(R RtxtytxIdxtyIdxtalphatone((s>/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pyR˜s$!  cC`s |dgS(Ni((R RR)((s>/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pyR*Çsc C`só|d \}}}}}} |js4|jƒ}nxªt|jdƒD]•} xŒt|jdƒD]w} xnt|jdƒD]Y} ||| | f|| | ffctj|| | f|| | dd…fƒ7/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pyR(Ês 7((sinplaceN( R6R7R8R9R:R;R R?RR*R((((s>/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pyR+„s   / cC`s§|j|jdkst‚|j|jks4t‚|jdkr|jdddƒ}|jddƒ}|jddƒ}ntƒ|j|ddƒ||||ƒS(s Compute the dot product (plus bias) of the specified pieces of vectors and matrices. See SparseBlockGemv to get more information. The parameter types are actually their expected shapes relative to each other. Parameters ---------- W : iBlocks, oBlocks, iSize, oSize weight matrix h : batch, iWin, iSize input from lower layer (sparse) inputIdx : batch, iWin indexes of the input blocks b : oBlocks, oSize bias vector outputIdx : batch, oWin indexes of the output blocks Returns ------- (batch, oWin, oSize) dot(W[i, j], h[i]) + b[j] but b[j] is only added once Notes ----- - `batch` is the number of examples in a minibatch (batch size). - `iBlocks` is the total number of blocks in the input (from lower layer). - `iSize` is the size of each of these input blocks. - `iWin` is the number of blocks that will be used as inputs. Which blocks will be used is specified in `inputIdx`. - `oBlocks` is the number or possible output blocks. - `oSize` is the size of each of these output blocks. - `oWin` is the number of output blocks that will actually be computed. Which blocks will be computed is specified in `outputIdx`. iiR@itaxis(RRR-Rttake(RRRR$R((s>/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pytsparse_block_dotÞs'(t __future__RRRRR RRt theano.tensorRttheano.gradientRRR+R;tsparse_block_gemvtTruetsparse_block_gemv_inplacetsparse_block_outertsparse_block_outer_inplaceRI(((s>/tmp/pip-build-X4mzal/theano/theano/tensor/nnet/blocksparse.pyts  zT