Introduction

3LC is a tool for understanding and improving your machine learning models and datasets. Powerful visualization tools, collection and analysis of custom metrics, and seamless editing of your dataset can all be unlocked with just 3 Lines of Code.

1. Import the tlc Python package:

import tlc

2. Create a Table object from your dataset:

...
dataset = tlc.Table.from_torch_dataset(dataset, ...)
...

3. Collect metrics to analyze in the 3LC Dashboard:

...
tlc.collect_metrics(dataset, ...)
...

3LC Workflow

The 3LC workflow loop. Data is used to train an ML model, then analyzed and modified using 3LC. New models can be trained on the modified data, further identifying issues in your dataset. The end result is an improved model and dataset.

The 3LC workflow loop. Data is used to train an ML model, then analyzed and modified using 3LC. New models can be trained on the modified data, further identifying issues in your dataset. The end result is an improved model and dataset.

Using 3LC begins in your own notebook or Python script. With only minor modifications to your code using the tlc package, 3LC lets you analyze your data and custom metrics, and modify your dataset as desired. Upon running your code, all of the collected data will appear in the 3LC Dashboard as a Run.

In the 3LC Dashboard, you can plot, filter, sort and search through the collected data. You can also use this data to compute new, derived Virtual Columns, giving you deeper insight into your model and data. If you find any issues with your data, or you just want to experiment, you can edit or re-weight samples in your dataset directly in the Dashboard.

Once you have made changes to your dataset, there is no need to manually download, export, or re-factor your code. Simply re-run your code, and 3LC will automatically use the modified dataset, collect new metrics, and allow you to compare the results against previous runs and iteratively improve your model and dataset.