Creating a pipeline is the first step in making your custom pipeline. Using this command will generate files where you can customize data processors, models, and user logic.

Sub-commands

• –pipeline — Give an initial name to the pipeline you are developing. This won’t be the pipeline name, but the initial name. Once you finish one stage or multiple, you can build the pipeline and give it an official name.

• –directory — Add the directory address for your custom pipeline. This is where the folder with all the files will be added.

• –modify — Write the name of the built-in pipeline you want to modify. Instead of writing the whole pipeline from zero, you can choose to modify one of our built-in models.

Files created

Once you run this command, a folder will be generated in the directory you specified. This folder includes the following files:

• dataFormatter.py — Decide whether you will allow the users of your pipeline to use data of a different format than the one you expect. For example, if your pipeline expects a CSV file, but you want to allow your users to use a TXT file, you can enter the logic of turning the TXT file into a CSV file. This is optional and you can leave this untouched.

• processor.py — Insert your code for processing your data. You must code some specific functions for the pipeline to work. All instructions are included in the file.

Once you finish coding the processor, you can build the model and run the prepare stage to test it.

• model.py — Insert the code for your model in this file. There are absolutely no restrictions on the model. The model can be of any type under the earth. More details are included in the file.

• trainer.py — This is where you insert the code for the training loop and evaluation logic. The only requirements are to code a train and evaluate function. The logic inside these functions can be anything. More details are included in the file.

Once you code the model and the trainer, you can build and run the train stage to test it.

• user.py — This is where you enter the code for using the model. The only requirement here is to code the function use_model. More details are included in the file.

Once you code the model, the trainer, and the user, you can build and run the deploy stage to test it.