Product Feature Spotlight: Teach
March 11, 2019 / Business, Data Science, Tutorials
When it comes to machine learning, clients are rarely satisfied with one-size-fits-all, pre-trained machine learning models. It makes perfect sense. Your business challenges and data are unique to you.
At the same time, it’s no small feat to get the necessary training data for developing machine learning solutions in place. Generating massive labeled datasets, designing and building data pipelines, implementing both hardware and software infrastructure for hosting large models, packaging the results into a human-friendly output—none of it is part of your core business and it’s costly in terms of time and money to understand and execute on the various pieces.
That’s why we’ve developed a point and click interface to handle the brunt of the work for you. With Indico’s IPA Product Modules, you can generate AI-powered custom models for analysis, classification, and comparison of your unstructured content.
In the first installment of this product spotlight blog series, we’ll introduce you to Teach, an intuitive interface that allows your subject matter experts to easily gather and label a relatively small subset of data, and quickly evaluate whether it is enough to train a new model.
Labeled training data is where AI success begins and ends. When we talk to customers, we’ve found that they understand this, but have felt frustrated with the challenge of how to actually create quality labeled data. The other big question we get is, “how much training data do we need?” To answer this, we added a predictive capability that provides real-time feedback as to how quickly your model is “learning” the task at hand.
How it works
- Dump the data you want to label into a single column in a CSV. Upload it to your Indico dataset dashboard and select the kind of machine learning task (classification or annotation) for which you’re labeling your dataset.
- Start labeling! Turn on “Predictions” to evaluate the readiness of your dataset for training a machine learning model.
- If you need to remove a label, or correct a prediction, it’s easy to do:
- If all the predictions look good, you can simply accept all predictions for the displayed section of data.
As you label more data, the predictions should improve greatly. Once you feel like the labeled data looks ready, a simple press of a button will present you with a customized machine learning model that is already deployed and ready to access within the Indico API, and which you can evaluate using Indico’s Review functionality (our feature spotlight for next month)!
Want to see our Teach module for yourself? Click here to set up your account.