Plugin Develop

The first, and most important, is to understand what a plugin is, and what its structure is.

After reviewing this entire guide, you can check out this page that teaches you how to upload your plugin to PyPI.

How to upload my plugin or package to PyPI?

What do I need to create a plugin?

To create a plugin you need to have the python classes of the software available. These can be available in 2 ways:

DashAI is installed as a library in the python virtual environment (not recommended option, doesn’t allow test the plugin).
Have the repository cloned on the computer. In this case, to develop a plugin, the best option is to create a folder called “plugins,” in the main DashAI folder.
```
DashAI
|— DashAI
|…
|— images
|— plugins <--- Create this folder
|— tests
| …
```
In this folder, the files associated with the plugin you want to create must be created.

Creating new components

The first thing when creating a plugin is to identify what types of components you want to create. These can be Machine Learning Models, Dataloaders, Configurable Objects, among others.

To ensure that this works, you must extend the classes provided by DashAI.

For example, if we want to create a tabular classification model, we would have something like this:

from DashAI.back.models.tabular_classification_model import TabularClassificationModel

class MyTabularClassificationModel(TabularClassificationModel):
    ...

Recommendations when creating plugins

Review other plugin examples

Look at examples of other plugins in the software. Both in the python files and the configuration file.

Plugin example: Image Classification Package
Test the plugin

The best way to test the plugin during its development is to clone the repository, and start by creating only the python and json files. Then move these files to where they belong within the repository so that they are available when starting the software. With this, you can start the software and test if your implementations really work.
Review requirements for a plugin
Review conflicts between libraries

When installing new libraries with pip, make sure that installing it does not give a conflict message indicating that some library was uninstalled.
```
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
fastapi 0.106.0 requires starlette<0.28.0,>=0.27.0, but you have starlette 0.20.0 which is incompatible.
```
To ensure this, after installing a new package, you can run the command
```
pip check
```
If there are conflicts, it will display a message like this
```
fastapi 0.106.0 has requirement starlette<0.28.0,>=0.27.0, but you have starlette 0.20.0.
```
Otherwise, the message will be
```
No broken requirements found.
```
and everything is fine.

It is important that when installing a new library for a plugin or package, it must be included among the dependencies of the plugin or package.
Use prints in the flow of the added components

Including prints in the flow of the component you created can be very useful to verify the proper functioning of the component you have created in the software.