Get in touch
or send us a question?

TFX Python function component tutorial

bookmark_borderNote: We recommend running this tutorial in a Colab notebook, with no setup required! Just click “Run in Google Colab”.

Run in Google ColabView source on GitHubDownload notebook

This notebook contains an examples on how to author and run Python function components within the TFX InteractiveContext and in a locally-orchestrated TFX pipeline.

For more context and information, see the Custom Python function components page on the TFX documentation site.


We will first install TFX and import necessary modules. TFX requires Python 3.

Check the system Python version

import sys
'3.9.16 (main, Dec  7 2022, 01:11:51) \n[GCC 9.4.0]'

Upgrade Pip

To avoid upgrading Pip in a system when running locally, check to make sure that we’re running in Colab. Local systems can of course be upgraded separately.

  import colab
  !pip install --upgrade pip

Install TFX

Note: In Google Colab, because of package updates, the first time you run this cell you must restart the runtime (Runtime > Restart runtime …).**

pip install tfx

Uninstall shapely

TODO(b/263441833) This is a temporal solution to avoid an ImportError. Ultimately, it should be handled by supporting a recent version of Bigquery, instead of uninstalling other extra dependencies.

pip uninstall shapely -y

Did you restart the runtime?

If you are using Google Colab, the first time that you run the cell above, you must restart the runtime (Runtime > Restart runtime …). This is because of the way that Colab loads packages.

Import packages

We import TFX and check its version.

# Check version
from tfx import v1 as tfx
2023-03-15 09:26:45.733506: W tensorflow/compiler/xla/stream_executor/platform/default/] Could not load dynamic library ''; dlerror: cannot open shared object file: No such file or directory
2023-03-15 09:26:45.733614: W tensorflow/compiler/xla/stream_executor/platform/default/] Could not load dynamic library ''; dlerror: cannot open shared object file: No such file or directory
2023-03-15 09:26:45.733624: W tensorflow/compiler/tf2tensorrt/utils/] TF-TRT Warning: Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.

Custom Python function components

In this section, we will create components from Python functions. We will notbe doing any real ML problem — these simple functions are just used to illustrate the Python function component development process.

See Python function based component guide for more documentation.

Create Python custom components

We begin by writing a function that generate some dummy data. This is written to its own Python module file.


import os
import tensorflow as tf  # Used for writing files.

from tfx import v1 as tfx

# Non-public APIs, just for showcase.
from tfx.types.experimental.simple_artifacts import Dataset

def MyGenerator(data: tfx.dsl.components.OutputArtifact[Dataset]):
  """Create a file with dummy data in the output artifact."""
  with, 'data_file.txt'), 'w') as f:
    f.write('Dummy data')

  # Set metadata and ensure that it gets passed to downstream components.
  data.set_string_custom_property('my_custom_field', 'my_custom_value')

Next, we write a second component that uses the dummy data produced. We will just calculate hash of the data and return it.


import hashlib
import os
import tensorflow as tf

from tfx import v1 as tfx

# Non-public APIs, just for showcase.
from tfx.types.experimental.simple_artifacts import Dataset
from tfx.types.standard_artifacts import String

def MyConsumer(data: tfx.dsl.components.InputArtifact[Dataset],
               hash: tfx.dsl.components.OutputArtifact[String],
               algorithm: tfx.dsl.components.Parameter[str] = 'sha256'):
  """Reads the contents of data and calculate."""
      os.path.join(data.uri, 'data_file.txt'), 'r') as f:
    contents =
  h =
  hash.value = h.hexdigest()

  # Read a custom property from the input artifact and set to the output.
  custom_value = data.get_string_custom_property('my_custom_field')
  hash.set_string_custom_property('input_custom_field', custom_value)

Run in-notebook with the InteractiveContext

Now, we will demonstrate usage of our new components in the TFX InteractiveContext.

For more information on what you can do with the TFX notebook InteractiveContext, see the in-notebook TFX Keras Component Tutorial.

from my_generator import MyGenerator
from my_consumer import MyConsumer

Construct the InteractiveContext

# Here, we create an InteractiveContext using default parameters. This will
# use a temporary directory with an ephemeral ML Metadata database instance.
# To use your own pipeline root or database, the optional properties
# `pipeline_root` and `metadata_connection_config` may be passed to
# InteractiveContext. Calls to InteractiveContext are no-ops outside of the
# notebook.
from tfx.orchestration.experimental.interactive.interactive_context import InteractiveContext
context = InteractiveContext()
WARNING:absl:InteractiveContext pipeline_root argument not provided: using temporary directory /tmpfs/tmp/tfx-interactive-2023-03-15T09_26_49.253300-6cj5wrmz as root for pipeline outputs.
WARNING:absl:InteractiveContext metadata_connection_config not provided: using SQLite ML Metadata database at /tmpfs/tmp/tfx-interactive-2023-03-15T09_26_49.253300-6cj5wrmz/metadata.sqlite.

Run your component interactively with

Next, we run our components interactively within the notebook with Our consumer component uses the outputs of the generator component.

generator = MyGenerator()
consumer = MyConsumer(

After execution, we can inspect the contents of the “hash” output artifact of the consumer component on disk.

tail -v {consumer.outputs['hash'].get()[0].uri}
==> /tmpfs/tmp/tfx-interactive-2023-03-15T09_26_49.253300-6cj5wrmz/MyConsumer/hash/2/value <==

That’s it, and you’ve now written and executed your own custom components!

Write a pipeline definition

Next, we will author a pipeline using these same components. While using the InteractiveContext within a notebook works well for experimentation, defining a pipeline lets you deploy your pipeline on local or remote runners for production usage.

Here, we will demonstrate usage of the LocalDagRunner running locally on your machine. For production execution, the Airflow or Kubeflow runners may be more suitable.