Aerosolve: A Machine learning Package Built For humans

Dec. 16, 2017, 3:43 a.m. By: Kirti Bakshi

Aerosolve

The new tool on the list, that made its entry at Airbnb’s 2015 OpenAir developer conference in San Francisco called Aerosolve, that is written mostly in the Java and Scala programming languages is basically a tool by Airbnb and is a machine learning library that has been designed right from the ground up to be human friendly and a few points that make it different from other machine learning libraries are mentioned below:

  • A thrift based feature representation that enables pairwise ranking loss and single context multiple item representations.

  • A feature transform language that in return gives all its users an upper hand in the control over all the features.

  • Human-friendly models that are Debuggable as well.

  • A Separate Java inference code that is lightweight

  • Use of Scala code for the purpose of training

  • The use of a Simple image content analysis code that is suitable for the ordering or ranking of images.

This library is meant to be used with features that are sparse and interpretable such as those that commonly occur in search, that is, in search keywords, filters or pricing may it be the number of rooms, location or even the price. Aerosolve can also more intelligently rank and order things like images. is a secret tool used by Airbnb in order to help people figure out the best price for their Airbnb rooms as well as apartments, it is also now available as a free download for developers who wish to build into their own apps. All this thus only proves that at Airbnb, machine learning works towards a very different end and Aerosolve is, therefore, a tool for what the industry calls as "machine learning," a spot where software and apps actually get smarter over time.

"We have been operating on the belief that enabling humans to partner with a machine in a symbiotic way exceeds the capabilities of humans or machines alone," as written by Airbnb in a blog post.

There are a few reasons as mentioned below that makes us to focus on interpretability:

  • When the corpus is new and not fully defined and one wants more insight into the corpus.

  • Having interpretable models allows iteration more quickly. Figure out where the model disagrees most and have insight into what kind of new features are needed.

  • Debugging of noisy features. By the plotting of the feature weights, one can also discover buggy features or even fit them into splines and then also further discover features that are unexpectedly complex. Something that usually indicates overfitting.

  • You can discover relationships between different variables and your target prediction.

Aerosolve

Aerosolve can also very automatically figure out about details about neighbourhoods are which on a map that is based on Airbnb listing data. And it's getting smarter all the time and is definitely not a constant.

We can, therefore, say that Airbnb rightly refers to Aerosolve as " A Machine Learning Package Built For Humans," and is designed to be easier to use for its users when compared other more complicated solutions.

For More Information: GitHub.