Creating Heatmap from Scratch in Python (2018)

[alexhutcheson]

I like seaborn's jointplot()[1] function for generating plots like this. I actually prefer the "hexagonal bin" plot, but jointplot() can also do kernel density estimation like this post does.

Hexbin example: https://seaborn.pydata.org/examples/hexbin_marginals.html

KDE example: https://seaborn.pydata.org/examples/joint_kde.html

If you are doing any exploratory analysis using scatterplots, I highly recommend that you also generate hexbin plot or heatmap. Far too often it turns out all of the points are concentrated in a small area of the plot, but that's not always apparent on a scatterplot.

[1] https://seaborn.pydata.org/generated/seaborn.jointplot.html

[anc84]

Please be aware that you MUST NOT use this for geographic coordinates such as "GPS" or "WGS84" latitude & longitude pairs. Depending on the longitude areas are of different sizes. You must use code that knows about that or your results will be utterly wrong.

[mturmon]

The OP makes a related mistake (although for their data, it worked out OK). It uses this grid:

  x_grid=np.arange(x_min-h,x_max+h,grid_size)
  y_grid=np.arange(y_min-h,y_max+h,grid_size)

where h is a small constant, and the range ends are the min/max of the data. In general, this will result in rectangular pixels, which you probably don't want.

In your example, even for a small geographic area, a degree of longitude will typically translate to a smaller distance than a degree of latitude, so as you say you can't treat them the same.

[peteradio]

This would only skew density maps not counts maps.

[kuzehanka]

How does this factor for tools like Datashader?

[yetihehe]

I much prefer kriging, which seems to better model sparse areas. https://en.wikipedia.org/wiki/Kriging too bad there are no image comparisons for those methods. Main difference - instead of sum of distances with cutoff, you make sum of values weighted by distances for each map point.

[plopz]

What you describe sounds more like idw than kriging.

[4thaccount]

Neat...I just wish there was an easy way to give the basemap library a 3-item tuple with (lat, lon, value) and have it automatically do a heat map and not just plot the points. That is an option in some software.

[rakshitadmar]

I have something of the sort. Give it later, lon and a value, it performs cubic interpolation and forms a heatmap. Some of the code is specific to my application but I'm sure you can reuse.

https://github.com/rraks/sigcatch

Open to contributions.

[urschrei]

basemap (like matplotlib) has the contourf method. You can calculate your x, y, and z values using matplotlib.mlab.griddata (altho scipy's are faster), and pass the results to it:

https://stackoverflow.com/a/26885815/416626

[anc84]

Related: basemap is deprecated, use cartopy instead! https://scitools.org.uk/cartopy/docs/latest/

[KyleOS]

I feel like these days we should be creating more user-friendly graphs - both for the creator of the graph and for the reader. I created this graph of the Atmospheric concentrations of CO2 earlier today (https://kyso.io/KyleOS/atmospheric-co2-concentrations#code=h...). Would love to see the OP's heatmap recreated with plotly.

[alexhutcheson]

If you're making heatmaps with GIS data, give some critical thought to the interpretation of the map to make sure you're not just doing this: https://xkcd.com/1138/

[gigatexal]

This is really cool. Thank you for sharing.

[geomatics99]

you're welcome