This notebook/dashboard shows USA AIS vessel (typically ship) tracking data collected from marinecadastre, colored by vessel type:

Apart from the actual data shown, which is interesting enough, the notebook also shows:

• How to process a large set of CSV files efficiently, caching result to a fast and efficient Parquet file.
• How to create and persist spatial indexing for SpatialPandas
• How to plot categorical data using Datashader
• How to select and highlight individual data points in a datashaded plot
• How to drill down into a Datashaded plot to reveal metadata about each point not shown in the plot itself

To do items:

• Add a series of static zoomed images like in this issue, to show data at each scale.
• Highlight some interesting bits of the data
• The spatial points selection code isn't currently working; I believe @jlstevens can replace it with some that he has.
• I'm not sure the spatial indexing in general is up and running; it's not fast per zoom even if only zooming slightly more in.
• After selecting a point, it would be useful to update a short but wide table of data below the plot to show that row of data in the original dataset
• It would also be great to be able to use show additional metadata about each selected point:
  • We would first need to build a table with metadata about the vessels covered by this dataset, by reading the data and collecting the latest non-NaN value for ['imo', 'call_sign', 'vessel_name', 'vessel_type','length', 'width'] for each 'mmsi_id'. This data shouldn't vary per ping, and can be stored separately from the ping data.
  • Once a point has been selected, look up this vessel info in a separate table
• It would also be great to show connected tracks:
  • Highlight a specific track to see how things connect, on top of a datashaded points plot
  • Datashade the whole set of pings, which might be slow and also may be confusing when it connects distant points -- may need to filter out gaps.
• Maybe selection can be in one of several modes: select nearest (default; highlights one point with a circle and shows one row in a table), select nearest N (highlights several nearby points and shows a table with multiple rows, perhaps with hovering over one point highlighting that row in the table to disambiguate), select all points with same column value as the selected one (MMSI in this case; highlights all such points and shows a table with multiple rows (up to a max), same but connecting each point into a trajectory by datetime
• Once info is available per point, need to collect interesting examples and talk about them in the notebook
• Test deployed app - responsive like in the notebook?
• Add dataset download info to anaconda-project.yml, putting the data (e.g. vessel metadata and Parquet file of points) on S3?