PyTorch TFX proposal

proposals/20220802-project_pytorch_example.md

@@ -0,0 +1,39 @@
+**Your name:** Hannes Hapke
+
+**Your email:** hannes--at--digits--dot--com
+
+**Your company/organization:** Digits Financial Inc
+
+**Project name:** TFX PyTorch Example
+
+## Project Description
+Adding a TFX pipeline example for PyTorch models to the TFX Addons repository.
+
+## Project Category
+Example
+
+## Project Use-Case(s)
+While there are a few non-TF model-based examples for TFX (e.g. JAX or Scikit), there isn't a maintained example for PyTorch models.
+
+## Project Implementation
+The pipeline example includes the following components:
+- Load a known dataset, e.g. MNIST,  via the CSVExampleGen component
+- Run the standard statistics and schema steps via StatisticsGen and SchemaGen
+- Performs a pseudo transformation (passthrough of the values) with the new PandasTransform component from tfx-addons
+- Add a custom run_fn function for PyTorch for the Trainer component
+- Add a TFMA example how to analysis PyTorch models to obtain a model blessing
+- Push the models to a local path
+
+## Project Dependencies
+The example will on TFX (1.9.1), TFX addons (0.2), Apache Beam, and PyTorch (1.0.2)
+
+## Project Team
+Hannes Hapke (gh: hanneshapke, email: hannes--at--digits--dot--com)
+More contributors more than welcome
+
+# Note
+Please be aware of the processes and requirements which are outlined here:
+
+* [SIG-TFX-Addons](https://github.com/tensorflow/tfx-addons)
+* [Contributing Guidelines](https://github.com/tensorflow/tfx-addons/blob/main/CONTRIBUTING.md)
+* [TensorFlow Code of Conduct](https://github.com/tensorflow/tfx-addons/blob/main/CODE_OF_CONDUCT.md)