support read xvector egs from python (#3850)

* support read xvector egs from python

* add some document for NnetExample

Author: fanlu

src/pybind/kaldi/table.py

@@ -33,6 +33,9 @@
 from kaldi_pybind.nnet3 import _RandomAccessNnetChainExampleReader
 from kaldi_pybind.nnet3 import _NnetChainExampleWriter
 
+from kaldi_pybind.nnet3 import _SequentialNnetExampleReader
+from kaldi_pybind.nnet3 import _RandomAccessNnetExampleReader
+
 from kaldi_pybind.feat import _SequentialWaveReader
 from kaldi_pybind.feat import _RandomAccessWaveReader
 from kaldi_pybind.feat import _SequentialWaveInfoReader
@@ -194,6 +197,11 @@ class SequentialNnetChainExampleReader(_SequentialReaderBase,
     '''Sequential table reader for nnet chain examples.'''
     pass
 
+class SequentialNnetExampleReader(_SequentialReaderBase,
+                                  _SequentialNnetExampleReader):
+    '''Sequential table reader for nnet examples.'''
+    pass
+
 
 class SequentialWaveReader(_SequentialReaderBase, _SequentialWaveReader):
     '''Sequential table reader for wave files.'''
@@ -350,6 +358,10 @@ class RandomAccessNnetChainExampleReader(_RandomAccessReaderBase,
     '''Random access table reader for nnet chain examples.'''
     pass
 
+class RandomAccessNnetExampleReader(_RandomAccessReaderBase,
+                                    _RandomAccessNnetExampleReader):
+    '''Random access table reader for nnet examples.'''
+    pass
 
 class RandomAccessWaveReader(_RandomAccessReaderBase, _RandomAccessWaveReader):
     '''Random access table reader for wave files.'''
@@ -572,11 +584,6 @@ class SequentialKwsIndexFstReader(
         '''Sequential table reader for FSTs over the KWS index semiring.'''
         pass
 
-    class SequentialNnetExampleReader(_SequentialReaderBase,
-                                      _kaldi_table.SequentialNnetExampleReader):
-        '''Sequential table reader for nnet examples.'''
-        pass
-
     class SequentialRnnlmExampleReader(_SequentialReaderBase,
                                        _kaldi_table.SequentialRnnlmExampleReader
                                       ):
@@ -658,12 +665,6 @@ class RandomAccessKwsIndexFstReader(
         '''Random access table reader for FSTs over the KWS index semiring.'''
         pass
 
-    class RandomAccessNnetExampleReader(
-            _RandomAccessReaderBase,
-            _kaldi_table.RandomAccessNnetExampleReader):
-        '''Random access table reader for nnet examples.'''
-        pass
-
     class RandomAccessIntReader(_RandomAccessReaderBase,
                                 _kaldi_table.RandomAccessIntReader):
         '''Random access table reader for integers.'''

src/pybind/nnet3/nnet_example_pybind.cc

@@ -2,6 +2,8 @@
 
 // Copyright 2019   Mobvoi AI Lab, Beijing, China
 //                  (author: Fangjun Kuang, Yaguang Hu, Jian Wang)
+//           2020   JD AI, Beijing, China
+//                  (author: Lu Fan)
 
 // See ../../../COPYING for clarification regarding multiple authors
 //
@@ -21,6 +23,7 @@
 #include "nnet3/nnet_example_pybind.h"
 
 #include "nnet3/nnet-example.h"
+#include "util/kaldi_table_pybind.h"
 
 using namespace kaldi;
 using namespace kaldi::nnet3;
@@ -40,4 +43,25 @@ void pybind_nnet_example(py::module& m) {
             "SparseMatrix would be the natural format for posteriors).");
     // TODO(fangjun): other constructors, fields and methods can be wrapped when
   }
+  {
+    using PyClass = NnetExample;
+    py::class_<PyClass>(m, "NnetExample",
+    "NnetExample is the input data and corresponding label (or labels) for one or "
+    "more frames of input, used for standard cross-entropy training of neural "
+    "nets (and possibly for other objective functions). ")
+        .def(py::init<>())
+        .def_readwrite("io", &PyClass::io,
+        "\"io\" contains the input and output.  In principle there can be multiple "
+        "types of both input and output, with different names.  The order is "
+        "irrelevant.")
+        .def("Compress", &PyClass::Compress,
+             "Compresses any (input) features that are not sparse.")
+        .def("Read", &PyClass::Read, py::arg("is"), py::arg("binary"));
+
+    pybind_sequential_table_reader<KaldiObjectHolder<PyClass>>(
+      m, "_SequentialNnetExampleReader");
+
+    pybind_random_access_table_reader<KaldiObjectHolder<PyClass>>(
+      m, "_RandomAccessNnetExampleReader");
+  }
 }