TensorFlow Transform #

- ExampleGen splits dataset
- StatisticsGen calculates statistics per feature
- SchemaGen infers the types of features
- ExampleValidator checks for data anomalies
- Transform performs feature engineering
Transform Inputs
- Data
- Schema
Transform Outputs
- Transformed Data
- Transform Graph
The result transform graph (tf.Graph) holds all necessary constants and transformations. The same chain of transformations can be easily applied at serving time.
Analizers Framework #
Analizers:
- Scaling
- scale_to_z_score
- scale_to_0_1
- Bucketizing
- quantiles
- apply_buckets
- bucketize
- Vocabulary
- bag_of_words
- tfidf
- ngrams
- Dimensionality Reduciton
- pca