FUTURESCALE

Machine Learning Playground Released

Over the last few years, "machine learning" has become an increasingly hot topic. And for good reason. Major breakthroughs are being reported on a regular basis, and many believe this could be the path to true artificial intelligence.

Recently several open-source tools for exploring the possibilities unlocked by deep learning have been released, and at the front of the pack is scikit-learn.

Scikit gives you a bunch of powerful tools that you can easily use to do things like make predictions about new, unlabeled data based upon previously observed, labeled data.

You can, for example, take a bunch of "labeled data", such as email subject lines labeled "spam" or "not spam", split it into two piles, one for training and one for testing.

Next you "train a model" with the first pile, and test the accuracy of its predictions by feeding it the data in the second pile.

Then, see if the labels it predicts match the actual labels on the data. If accuracy is poor, what can be done to improve it?

The keys to success appear to be:

• Get a lot of labeled data to train and test with.
• Make sure that each dimension you choose to label has a high value to the model.
• Understand the different types of classifiers, and how to choose the right one for your problem and data set.

We released a playground project for anyone wanting to begin experimenting with scikit-learn. The project uses scikit to train and test a simple model with just a few lines of code. Next, we build the Nearest Neighbor classifier.

That demo loads scikit's built in Iris dataset, which consists of 3 different types of irises’ (Setosa, Versicolour, and Virginica) petal and sepal length.

The goal is to train a model to be able to predict the type of iris given the petal and sepal length.

You could use the project to quickly get insights about your own labeled data.

• Visit the scikit-learn Project Website.
• Fork the project on on GitHub.