TensorFlow Transform is a library for preprocessing data with TensorFlow. tf.Transform is useful for data that requires a full-pass, such as:

• Normalize an input value by mean and standard deviation.
• Convert strings to integers by generating a vocabulary over all input values.
• Convert floats to integers by assigning them to buckets based on the observed data distribution.

TensorFlow has built-in support for manipulations on a single example or a batch of examples. tf.Transform extends these capabilities to support full-passes over the example data.

The output of tf.Transform is exported as a TensorFlow graph to use for training and serving. Using the same graph for both training and serving can prevent skew since the same transformations are applied in both stages.

For an introduction to tf.Transform, see the tf.Transform section of the TFX Dev Summit talk on TFX (link).

Caution: tf.Transform may be backwards incompatible before version 1.0.

The tensorflow-transform PyPI package is the recommended way to install tf.Transform:

pip install tensorflow-transform

TensorFlow is required.

Apache Beam is required; it's the way that efficient distributed computation is supported. By default, Apache Beam runs in local mode but can also run in distributed mode using Google Cloud Dataflow and other Apache Beam runners.

Apache Arrow is also required. TFT uses Arrow to represent data internally in order to make use of vectorized numpy functions.

The following table is the tf.Transform package versions that are compatible with each other. This is determined by our testing framework, but other untested combinations may also work.

Please direct any questions about working with tf.Transform to Stack Overflow using the tensorflow-transform tag.