A quick and dirty TensorFlow Reader Op for Matlab .MAT file parsing support.

• Requires the MatIO library (https://sourceforge.net/projects/matio/), and
• (optionally) its dependencies for Matlab 7.3 and compressed .MAT file support.

1. Build the op with TensorFlow binary or source.
2. Load and run the op in Python

import tensorflow as tf
parse_mat_module = tf.load_op_library('parse_mat.so')
mat_tensor = parse_mat_module.parse_mat([filename], [matrixname], dtype=[datatype])

3. To use the parser in the input pipeline (cf. https://www.tensorflow.org/how_tos/reading_data/index.html), use an IdentityReader to produce file name strings from a file name queue, and then feed the resulting tensor to parse_mat.

• uint32 and uint64 types are not supported because of missing types in TensorFlow source as of r0.9: #1894

TODO:

• Add uint32 and uint64 support once TensorFlow supported types are updated
• Probably redo the op as a Reader to get rid of the IdentityReader thing