My First Project
In this tutorial, we show you how to create a Project in EXPAI from scratch using your models and datasets. At the end, you will find a video tutorial.
On the initial page "My Projects", click on "New Project" and give it a name and description. You can modify this information later.
If you want further information on how to create a ML model using Pipelines to achieve the best integration with EXPAI, take a look to our tutorial here.
Click on "New model" and fill out the form. Each of the parameters is defined as follows:
- Name: unique string that allows you to identify the model
- Description (optional): text you can use to provide further information about the model
- File: you must include here the binary file for your model. For example, you can use Pickle to store your ML models. More information here.
- Objective: select whether your model is a regressor (predicts a continues value) or a classifier among different labels.
- Prediction: this field is variable depending on your "Objective". For regression, you can only choose "Regression". However, for classifiers, you must select whether you are working with a binary classifier (2 classes) or a multi-class model (>2 labels).
- Library: "Pickle", "Tensorflow" or "Joblib". Select the framework you used to export your model.
There are other optional fields that appear depending on your previous selections.
For binary classifiiers:
- Cutoff: This value allows you to tune the decision boundary for binary classification. Samples with output probability above this number will be considered as class 1. By default 0.5 is used (rounding).
- Output classes: This field lets you introduce names to your output classes separated by commas. By default, an ML model outputs numerical values for each of the labels (e.g. 0,1,2,3). However, to ease the explanations generation, you can provide a name to each of them. For example, if 0: "red"; 1: "blue"; 2: "green"; 3: "yellow", you will introduce red,blue,green,yellow in this field.
For classifiers considering >2 labels:
- Number of classes: select how many output classes does your model have.
Click on "New dataset" and fill out the form. Each of the parameters is defined as follows:
- Name: unique string that allows you to identify the dataset
- Description (optional): text you can use to provide further information about the dataset
- File: upload a CSV file containing the data used as input to the model. It is really important that this dataset contains the columns expected by the
model.predict()function. It may contain additional columns for filtering, as we will see in a moment. - Encoding: encoding used to store your csv file.
- Separator: separator string for your csv file.
- Target column: select the groundtruth column that your model is trying to predict.
- Protected columns (optional): you may want to select some columns that contain sensitive attributes that may lead to discrimination such as gender or race.
- Columns to remove (optional): if your CSV contains additional columns that should not be used as input to the model, you can select them here and EXPAI will remove them before performing the forward pass. However, you will be able to use them to filter your data and generate fine-grained explanations for different subgroups.
- Index column (optional): if your CSV file was stored with an index column, you must select it here.
Once you have uploaded your model and datasets to the platform, you can start generating explanations. See our tutorial here on how to generate explanations.