tft.get_num_buckets_for_transformed_feature  |  TFX  |  TensorFlow (original) (raw)

tft.get_num_buckets_for_transformed_feature

Stay organized with collections Save and categorize content based on your preferences.

Provides the number of buckets for a transformed feature if annotated.

tft.get_num_buckets_for_transformed_feature(
    transformed_feature: common_types.TensorType
) -> tf.Tensor

This for example can be used for the direct output of tft.bucketize,tft.apply_buckets, tft.compute_and_apply_vocabulary,tft.apply_vocabulary. These methods annotate the transformed feature with additional information. If the given transformed_feature isn't annotated, this method will fail.

Example:

def preprocessing_fn(inputs): bucketized = tft.bucketize(inputs['x'], num_buckets=3) integerized = tft.compute_and_apply_vocabulary(inputs['x']) zeros = tf.zeros_like(inputs['x'], tf.int64) return { 'bucketized': bucketized, 'bucketized_num_buckets': ( zeros + tft.get_num_buckets_for_transformed_feature(bucketized)), 'integerized': integerized, 'integerized_num_buckets': ( zeros + tft.get_num_buckets_for_transformed_feature(integerized)), } raw_data = [dict(x=3),dict(x=23)] feature_spec = dict(x=tf.io.FixedLenFeature([], tf.int64)) raw_data_metadata = tft.DatasetMetadata.from_feature_spec(feature_spec) with tft_beam.Context(temp_dir=tempfile.mkdtemp()): transformed_dataset, transform_fn = ( (raw_data, raw_data_metadata) | tft_beam.AnalyzeAndTransformDataset(preprocessing_fn)) transformed_data, transformed_metadata = transformed_dataset transformed_data [{'bucketized': 1, 'bucketized_num_buckets': 3, 'integerized': 0, 'integerized_num_buckets': 2}, {'bucketized': 2, 'bucketized_num_buckets': 3, 'integerized': 1, 'integerized_num_buckets': 2}]