Models

Learn more about Models within Vectice.

Models overview

Models reflect the model metadata logged to Vectice during model development. All models are versioned and tagged with their deployment environment within Vectice.

Model artifacts in Vectice

Below is a list of artifacts you can find in the Vectice app after data scientists log the models artifacts from their iterative development.

ArtifactsDescription

Model snapshot

Highlights the number of model versions grouped by model status.

Model Status

The model status states whether the model is in Production, Staging, or Experimentation mode.

Model creation details

Creation details show when the model was created and who it was created by.

Model type

Classifies the model by model type. For example, clustering, classification, or regression.

Technique

The technique highlights the exact modeling algorithm used for the model. (i.e., Linear Regression)

Model lineage

The model lineage can showcase the dataset used for the model, the code captured from git and the model version output.

Model metrics

The model metrics captures metrics used to evaluate model performance. (i.e., MAE and RSME values)

Versions

Models with the same name as another model you already logged in Vectice. As you log models with the same name, the versions are incremented, maintaining the model's history.

Attachments

Attachments showcase files that were logged with the model. For example, this could be notebooks, images, or excel sheets of data analysis, model results, etc. All file types are supported for model attachments and can be downloaded from the Vectice app.

However, the supported previews types in the Vectice app are PNG, JPEG, CSV, Notebook, and TXT files.

Models best practices

  • Project phases that aim to train, test, or validate models should have iteration sections to log the models accordingly. This is typically completed in the Modeling phase of a CRISP-DM project.

  • Mark your most valuable iterations and assets in the Vectice app by selecting the star next to the corresponding iteration and assets before beginning the phase review. This will make it easier for stakeholders and subject matter experts to identify the iterations and assets in review.

Select the star next to the valuable iterations before completing a phase, even without review. This allows you to identify the most valuable iterations for that phase.

How to log models to Vectice?

View our How to log models guide for more information on logging models during iterative development.