ivan-rivera/tfx

Playing around with Tensorflow Extended

Tensorflow Extended Demo

In this repository I'm taking TFX for a spin.

Requirements

Aside from requirements.txt, this project requires Python 3.8.12. Lastly, the original content of the fraud directory was populated with the following command:

 tfx template copy \
 --model=taxi \
 --pipeline_name="fraud" \
 --destination_path="fraud"

Data

Since the dataset is fairly large, it is not committed to git, but it can be downloaded from here and saved in data/. We are using a credit card fraud detection dataset.

TFX resources

Here is a list of relevant TFX resources that were used for this exercise:

• Building TFX Pipeline Locally
• Create a TFX pipeline using templates with local orchestrator - Colab Notebook
• TFX in interactive context
• TFX in a notebook
• Creating a custom TFX executor
• Creating a custom TFX component
• Example of a custom component using the Slack API

How to get it working

• clone
• download application_data.csv from Kaggle to the top-level data/ directory
• create a new folder pipeline_outputs in the project folder
• run make sample_data
• run make create_pipeline
• run make update_and_run
• run make tensorboard to check out the training logs
• Check out the notebooks (work in progress)

Note that with every run you are accumulating output data.