Making RecordBatchToExamplesEncoder consistent with tf.io.parse_example for RaggedFeatures.

Previously, if RT had uniform dimensions, the encoding would be incorrect.

PiperOrigin-RevId: 514783895

Author: iindyk

tensorflow_transform/beam/impl_test.py

@@ -3139,29 +3139,29 @@ def analyzer_fn(inputs):
               'scale_by_min_max$ragged_values': [
                   0., 0.285714, 0.428571, 0.142857, 0.285714, 1.
               ],
-              'scale_by_min_max$row_lengths_1': [0, 4, 0, 2],
-              'scale_to_z_score$row_lengths_1': [0, 4, 0, 2],
+              'scale_by_min_max$row_lengths_1': [0, 2, 0, 1],
+              'scale_to_z_score$row_lengths_1': [0, 2, 0, 1],
               'scale_to_z_score$ragged_values': [
                   -1.3333334, -0.22222228, 0.33333328, -0.77777785, -0.22222228,
                   2.5555556
               ],
-              'scale_to_0_1$row_lengths_1': [0, 4, 0, 2],
+              'scale_to_0_1$row_lengths_1': [0, 2, 0, 1],
               'scale_to_0_1$ragged_values': [
                   0., 0.2857143, 0.42857143, 0.14285715, 0.2857143, 1.
               ],
           }, {
               'scale_to_0_1$ragged_values': [
                   0.42857143, 0.42857143, 0.14285715, 0.2857143
               ],
-              'scale_to_0_1$row_lengths_1': [4],
+              'scale_to_0_1$row_lengths_1': [2],
               'scale_by_min_max$ragged_values': [
                   0.42857143, 0.42857143, 0.14285715, 0.2857143
               ],
-              'scale_by_min_max$row_lengths_1': [4],
+              'scale_by_min_max$row_lengths_1': [2],
               'scale_to_z_score$ragged_values': [
                   0.33333328, 0.33333328, -0.77777785, -0.22222228
               ],
-              'scale_to_z_score$row_lengths_1': [4],
+              'scale_to_z_score$row_lengths_1': [2],
           }],
       ),
   )
@@ -4544,13 +4544,13 @@ def preprocessing_fn(inputs):
             'x_ones$ragged_values': [1, 1, 1],
             'x_ones$row_lengths_1': [1, 0, 2],
             'y_ones$ragged_values': [1, 1],
-            'y_ones$row_lengths_1': [2],
+            'y_ones$row_lengths_1': [1],
         },
         {
             'x_ones$ragged_values': [],
             'x_ones$row_lengths_1': [0],
             'y_ones$ragged_values': [1, 1, 1, 1],
-            'y_ones$row_lengths_1': [2, 0, 2],
+            'y_ones$row_lengths_1': [1, 0, 1],
         },
     ]
     expected_metadata = tft.DatasetMetadata.from_feature_spec({

tensorflow_transform/impl_helper_test.py

@@ -207,7 +207,7 @@ def _sparse_value_name(tensor_name='sparse'):
             'j_row_lengths2': [2, 1, 1],
             'k_val': [0, 1, 2, 3, 4, 5, 6, 7, 8],
             'l_val': [0, 1, 2, 3, 4, 5],
-            'l_row_lengths1': [4, 2],
+            'l_row_lengths1': [2, 1],
         }, {
             'a': 100,
             'b': 2.0,
@@ -227,7 +227,7 @@ def _sparse_value_name(tensor_name='sparse'):
             'j_row_lengths2': [1],
             'k_val': [9, 10, 11],
             'l_val': [6, 7],
-            'l_row_lengths1': [2],
+            'l_row_lengths1': [1],
         }],
         record_batch=_MULTIPLE_FEATURES_CASE_RECORD_BATCH,
         feed_dict=_FEED_DICT),
@@ -271,7 +271,7 @@ def _sparse_value_name(tensor_name='sparse'):
             'l_val':
                 np.array([0, 1, 2, 3, 4, 5]),
             'l_row_lengths1':
-                np.array([4, 2]),
+                np.array([2, 1]),
         }, {
             'a': np.int64(100),
             'b': np.array(2.0, np.float32),
@@ -291,7 +291,7 @@ def _sparse_value_name(tensor_name='sparse'):
             'j_row_lengths2': np.array([1]),
             'k_val': np.array([9, 10, 11]),
             'l_val': np.array([6, 7]),
-            'l_row_lengths1': np.array([2]),
+            'l_row_lengths1': np.array([1]),
         }],
         record_batch=_MULTIPLE_FEATURES_CASE_RECORD_BATCH,
         feed_dict=_FEED_DICT),