[GitHub] [tvm] mbrookhart commented on a change in pull request #7126: Sparse fill empty rows op

[GitBox <gi...@apache.org>]

mbrookhart commented on a change in pull request #7126:
URL: https://github.com/apache/tvm/pull/7126#discussion_r546916764



##########
File path: python/tvm/relay/op/transform.py
##########
@@ -1320,3 +1320,84 @@ def adv_index(inputs):
         Output tensor.
     """
     return _make.adv_index(Tuple(inputs))
+
+
+def sparsefillemptyrows(sparse_indices, sparse_values, dense_shape, default_value):
+    """
+    Fill first column of the empty rows with default values for a sparse array.
+
+    Parameters
+    ----------
+    sparse_indices : relay.Expr
+        A 2-D tensor[N, n_dim] of integers containing location of sparse values, where N is the
+        number of sparse values and n_dim is the number of dimensions of the dense_shape
+
+    sparse_values : relay.Expr
+        A 1-D tensor[N] containing the sparse values for the sparse indices.
+
+    dense_shape : relay.Expr
+        A list of integers. Shape of the dense output tensor.
+
+    default_value : relay.Expr
+        A 0-D tensor containing the default value for the remaining locations.
+        Defaults to 0.
+
+    Returns
+    -------
+    TupleWrapper with the following four outputs
+
+    new_sparse_indices : relay.Expr
+        A 2-D tensor[N + dense_shape[0], n_dim] of integers containing location of new sparse
+        indices where N is the number of sparse values. It is filled with -1 at to_be_discarded
+        indices.
+
+    empty_row_indicator : relay.Expr
+        A 1-D Boolean tensor[dense_shape[0]] indicating whether the particular row is empty
+
+    new_sparse_values : relay.Expr
+        A 1-D tensor[dense_shape[0]] containing the sparse values for the sparse indices. It is
+        filled with -1 at to_be_discarded indices.
+
+    slice_element_index : relay.Expr
+        A 1-D tensor containing the amount of elements in the sparse_indices and new_sparse_values
+        expression to be sliced in a future op discarding non-useful elements in new_sparse_indices
+        and new_sparse_values
+
+    Examples
+    -------
+
+    .. code-block:: python
+
+    sparse_indices = [[0, 1],
+                      [0, 3],
+                      [2, 0],
+                      [3, 1]]
+    sparse_values = [1, 2, 3, 4]
+    default_value = [10]
+    dense_shape = [5, 6]
+    new_sparse_indices, empty_row_indicator, new_sparse_values, slice_element_index =
+                        relay.sparsereshape(
+                        sparse_indices,
+                        sparse_values,
+                        prev_shape,
+                        new_shape)
+    new_sparse_indices =       [[0, 1],
+                               [0, 3],
+                               [2, 0],
+                               [3, 1],
+                               [1, 0],
+                               [4, 0],
+                               [-1, -1],

Review comment:
       We have it set up that way because that's how MXNet does it (TOPI's original opset was mostly cloned from MXNet), and MXNet is fairly conservative about static shapes. From a performance perspective, if the expectation is that we'll always use this with a dynamic strided slice after, I'd think it would be better to just emit a dynamic shape. If we have another usecase, however, I think this is fine.
   
   Do you have a framework level usecase in mind?




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org