> ## Documentation Index
> Fetch the complete documentation index at: https://docs.permutive.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Validating Model Weights

> Learn how to interpret model weights to ensure your lookalike model is accurate.

Model weights provide transparency into the machine learning process by showing which behavioral signals (cohorts) the model found most significant when identifying lookalikes.

## How to View Weights

Once a model has finished training, you can view the weights in the Permutive dashboard:

1. Navigate to **Modeled Cohorts > Models**.
2. Click on your trained model.
3. Scroll down to the **Model Weights** section.

## Interpreting Weights

The model assigns a numerical weight to each cohort in your project's feature space.

* **Positive Weights**: Users in these cohorts are *more likely* to be similar to your seed segment.
* **Negative Weights**: Users in these cohorts are *less likely* to be similar to your seed segment.
* **Magnitude**: The larger the absolute value of the weight, the stronger the signal.

### Example

If you are building a lookalike model for a "Sports Lover" seed segment:

* A cohort like "Rugby Fans" should ideally have a **high positive weight**.
* A cohort like "Vegetarian Cooking" might have a **negative or neutral weight**.

## Using Weights for Validation

Reviewing weights is a powerful way to "sanity check" your model:

* **Unexpected Signals**: If a completely unrelated cohort has a very high weight, it may indicate noise in your data or a coincidental correlation.
* **Missing Signals**: If a cohort you expect to be highly related has a low or negative weight, the model may not be capturing the behavioral patterns you intended.

<Tip>
  If the weights don't align with your expectations, consider expanding your feature space with more diverse cohorts or adjusting your seed segment definition.
</Tip>