| Input data is a N * D matrix. Apply box-cox
| transform for each column. lambda1 and lambda2
| is of size D that defines the hyper-parameters for
| the transform of each column x of the input
| data:
|
| ln(x + lambda2), if lambda1 == 0
| ((x + lambda2)^lambda1 - 1)/lambda1, if lambda1 != 0
|
| Bucketize the float_features into
| sparse features.
|
| The float_features is a N * D tensor where
| N is the batch_size, and D is the feature_dim.
|
| The indices is a 1D tensor containing
| the indices of the features that need
| to be bucketized.
|
| The lengths is a 1D tensor that splits
| the following ‘boundaries’ argument.
|
| The boundaries is a 1D tensor containing
| the border list for each feature.
|
| With in each batch, indices should
| not have duplicate number, and the number
| of elements in indices should be less
| than or equal to D.
|
| Each element in lengths vector (lengths[i])
| represents the number of boundaries
| in the sub border list.
|
| The sum of all elements in lengths
| must be equal to the size of boundaries.
|
| If lengths[0] = 2, the first sub border
| list is [0.5, 1.0], which separate the
| value to (-inf, 0.5], (0,5, 1.0], (1.0,
| inf). The bucketized feature will have
| three possible values (i.e. 0, 1, 2).
|
| This Op is a inverse of BatchSparseToDenseOp.
|
| Basically, given a lengths vector, a indices
| vector, and a dense matrix dense, output value
| vector so that, along with lengths vector and
| indices vector, forms a sparse representation of
| the dense matrix.
|
| A sparse matrix is represented by lengths
| vector, indices vector, and values vector.
|
| Each element in lengths vector (lengths[i])
| represents the number of indices in this batch
| (batch i).
|
| With in each batch, indices should not have
| duplicate number.
|
| Batch gather operation, first dimension in DATA is
| the batch size.
|
| Given DATA tensor of rank r >= 2, and INDICES
| tensor of rank q >= 1, gather entries of the
| second outer dimension (axis == 1) of DATA indexed
| by INDICES, and concatenate them in an output
| tensor of rank q + (r - 1).
| Batch Matrix multiplication Yi = Ai * Bi, where
| A has shape (dim0, dim1, … M, K), B has shape
| (dim0, dim1, … K, N), Y has shape (dim0, dim1,
| … M, N) and i ranges from 0 to (dim0 * dim1 …)
| - 1. rank(A) == rank(B) >= 2.
|
| In case of A and B being two dimensional, it
| behaves like normal matrix multiplication.
| Batch permutation of an input tensor
| X given input indices.
|
| First dimension of X equals batch size
| N.
|
| The indices stores a be permutation
| of N.
|
| The output Y is a tensor of same shape
| as X, with data re-ordered according
| to the indices within the batch size.
|
| Convert sparse matrix representation
| into dense matrix.
|
| A sparse matrix is represented by lengths
| vector, indices vector, and values
| vector.
|
| Each element in lengths vector (lengths[i])
| represents the number of indices in
| this batch (batch i).
|
| With in each batch, indices should
| not have duplicate number.
|
| Buffers used by the MKL version are cached
| across calls.
|