GenSynth Documentation

Preparing your Data

GenSynth needs your data (training, validation, and test) in the design generation process.


At this time, GenSynth only supports the NHWC data format. If you are using NCHW-style data format for tensors, contact DarwinAI customer support for assistance.

For this task, you ultimately need to write code in a Dataset entity to present your data in a format that can be used by GenSynth and your model.

Identify where the training and validation data is located and what format it is in. Identify what transformations will need to be done to the data in order to present it to the model.

To use your data with GenSynth you will need to copy it to a folder within the GenSynth workspace. You may wish to transform the data at the time you copy it. Converting data into the TFRecord file format allows for efficient reading at run-time.

To provide the data to GenSynth you will need to write (or obtain) a Python 3 plugin to adapt the data for GenSynth. The plugin provides data via the TensorFlow Dataset interface. We recommend starting from a template provided by DarwinAI, and working within the Data Entity page.

You will develop your data plugin for GenSynth on the Entities tab (Dataset Manager).


You may use Python's print() command to debug the dataset; output will be visible when viewing the validation results.

Develop your module in GenSynth's Dataset editor. Start from an existing template or example and modify it for your case. This is where you write the code to read, transform, and augment the data. Validate it and get it working within GenSynth's Dataset Entity screen.