Clarify why COO is not efficient with common TF sparsities

Author: penpornk

rfcs/20200519-csr-sparse-matrix.md

@@ -24,7 +24,7 @@ Support the [compressed sparse row (CSR)](https://en.wikipedia.org/wiki/Sparse_m
 
 ## Motivation
 
-Sparse tensor representation has a significant impact on performance. Modern architectures are most suited for non-random, bulk memory accesses. Sparse formats that optimize for data locality and reuse can achieve large performance gains. TensorFlow currently stores sparse tensors in [coordinate (COO)](https://www.tensorflow.org/api_docs/python/tf/sparse/SparseTensor) format, which is inefficient for sparsities observed in most deep learning and sparse linear algebra applications. The compressed sparse row (CSR) format is one of the most commonly used formats. It generally requires less storage and is faster than COO, sometimes by up to orders of magnitudes.
+Sparse tensor representation has a significant impact on performance. Modern architectures are most suited for non-random, bulk memory accesses. Sparse formats that optimize for data locality and reuse can achieve large performance gains. TensorFlow currently stores sparse tensors in [coordinate (COO)](https://www.tensorflow.org/api_docs/python/tf/sparse/SparseTensor) format, which works well for tensors with very few nonzeroes, and is inefficient otherwise. Deep learning and sparse linear algebra applications typically do not have sufficient sparsities to benefit from the COO format. The compressed sparse row (CSR) format is one of the most commonly used formats. It generally requires less storage and is faster than COO, sometimes by up to orders of magnitudes.
 
 We propose supporting the CSR format in TensorFlow to accelerate sparse linear algebra and applicable deep learning applications in TensorFlow.