Introduction This dataset demonstrates the use of Gaussian processes (GPs) to learn and make inferences about traffic speed distribution over a city-wide road network (of Pittsburgh, Pennsylvania) at 3 different points in time (8 am, 2 pm, and 8 pm) on a typical weekday/weekend. Content The dataset contains 3 separate CSV files: training set, prediction set, and (road) segment data frame. The training set contains the observed (average) traffic speed over select road segments (where traffic sensors are installed) at 3 different times (8 am, 2 pm, and 8 pm) on a typical weekday/weekend over a multi-month period. Each road segment is specified by the longitude and latitude (x and y) coordinates of its two endpoints. The prediction set extends the (spatially inferred) traffic speeds over the entire road network of the city, covering all road segments where no sensors were installed. Similar to the training set, the coverage periods are at those 3 time points on a typical weekday/weekend. The segment data frame only contains the x and y coordinates of the road segments in the city (described by the shapefile downloaded from http://www.wprdc.org/). This data frame is to be merged with the prediction set (given a time point and a day) in order to render the city-wide traffic speed distribution at that particular time. Acknowledgements For detailed description on data collection and curation as well as machine learning methodologies, refer to this paper: https://ieeexplore.ieee.org/abstract/document/7676341