我们是如何以及为什么建立一个定制的梯度提升树包的？

In order to make accurate and fast travel-time predictions, Lyft built a gradient boosted tree (GBT) package from the ground up. It is slower to train than off-the-shelf packages, but can be customized to treat space and time more efficiently and yield less volatile predictions.

为了进行准确和快速的旅行时间预测，Lyft从头开始建立了一个梯度提升树（GBT）包。与现成的软件包相比，它的训练速度较慢，但可以进行定制，以更有效地处理空间和时间，产生更少的不稳定预测。

Image by ejaugsburg from Pixabay

图片来自Pixabay的ejaugsburg

Introduction

简介

Machine learning runs at the core of what we do at Lyft. Examples include predicting travel time between two locations, modeling the probability of a ride being canceled, forecasting supply and demand, and many more. These models enable us to match riders and drivers more efficiently, incentivize drivers to be where they can get more rides, and improve the ride experience. There are many more examples that together enable us to provide better service to our customers at a lower price.

机器学习是我们在Lyft所做工作的核心。这方面的例子包括预测两个地点之间的旅行时间，建立乘车被取消的概率，预测供应和需求，以及更多。这些模型使我们能够更有效地匹配乘客和司机，激励司机在他们能够获得更多乘车的地方，并改善乘车体验。还有很多例子，它们共同使我们能够以更低的价格为客户提供更好的服务。

One important machine learning model we run at Lyft predicts travel time. In this blogpost, I describe how my team developed a custom gradient boosting decision tree package to push the accuracy and efficiency of this model. We’ll also learn a bit about gradient boosting decision trees and how to use location data in your machine learning models.

我们在Lyft运行的一个重要机器学习模型是预测旅行时间。在这篇博文中，我描述了我的团队如何开发了一个定制的梯度提升决策树包来推动这个模型的准确性和效率。我们还将学习一些关于梯度提升决策树的知识，以及如何在你的机器学习模型中使用位置数据。

Location, location, location!

位置，位置，位置!

One of the key features used in many of our models at Lyft is location (for example, passenger location, driver location, or rider destination). In the travel time estimation model, the main features are time, start location and end location of the ride. Location is typically represented by two float values, latitude and longitude, which can be fed directly into a model. However, doing so in models such as logistic regression, gaussian processes, or neu...