frontends/react/src/usage.md

### Manual dataset

1. Go to **Dataset** tab and choose **Manual**

2. Click on the plot to add your data points

3. Change the point label to change classes

4. Go to the **Model** tab, and select a classifier. Click **Get decision boundary** to generate decision regions

5. Optionally add sklearn model arguments with comma-separated key-value pairs, e.g. for Logistic Regression  
`C=10, penalty=l1, solver=liblinear`  
To see available arguments, refer to the scikit-learn documentation (search for "sklearn <classifier name>").


### Upload or select a dataset

1. Go to **Dataset** tab and select **Upload** or a **Preset** dataset

2. If upload, click button to upload a CSV file

3. Input which columns you would like to use for the model inputs, e.g. to use the first 3 colums:  
`1,2,3`

4. Input which column is the model target, e.g. for the iris dataset:
`5`

5. Normalize data and add input noise if desired

6. Select how to project the dataset onto the 2d plot. There are two options:

    - Coordinate projection - project onto an input coordinate
    - PCA - project onto the top 2 PCA components

7. Go to the **Model** tab, and select a classifier. Click **Get decision boundary** to generate decision regions.

8. Optionally add sklearn model arguments with comma-separated key-value pairs, e.g. for Logistic Regression  
`C=10, penalty=l1, solver=liblinear`  
To see available arguments, refer to the scikit-learn documentation (search for "sklearn <classifier name>").


###  Which data does the model see?
For **Manual** datasets, the model is trained on the datapoints on the plot and decision regions are evaluated using the regions visible in the plot.

For **uploaded and preset** datasets, the model is trained on the dataset after normalization and noise, but before the 2d projection. When plotting decision regions, the visible points on the plot are mapped to the original input space via an inverse projection. For coordinate projection, non-projected coordinates are filled using the dataset mean.