unexpected output from raster calculator function

Question

Working in ArcMap, I am trying to convert the logit values in raster into their actual probabilities (logistic regression,) using the following function:

1/(1+Exp(-"sn_topo_poly_log"))

Ive used this function before and it should return a raster with values from 0 to 1. Instead I get a raster with all values=0.

the input raster "sn_topo_poly_log," looks correct and has values from close to zero to -871361

• Delonix R.: Have you try to work with 1 as 1.0, at least in the first one (in both woulnd't hurt)?
• FelixIP: Use float(1) and float (1)
• whuber: @Nir That's simply incorrect. Sam: Could you explain how you know all the values are zero? Sometimes ArcMap's automatic legend is screwed up but the values in the dataset are still ok.
• sam: yeah ive tried playing with symbology, but there are no values other than zero to display.

Answer 1

Did you mean to use the equivalent of Python's log1p instead? (So the inverse function)? I think in Arcgis this is the Ln() function

e to the power of 871361.0 is a very large number, and too big to fit in 32bit float, so arc will convert it to nodata. The same thing happens for negative values less than -745, which are too small to represent in a float.

If I do math.log1p(871361.0) I get a value around 13.67 - so e^13.67 is around 871361.

My math is a bit rusty, so apologies if you did mean to use exp() - you'll just need to normalise your data to fit into a smaller range of values.

• whuber: Sam's formula is the correct one for what he wants it to do. See en.wikipedia.org/wiki/Logit.
• sam: yes ive double checked and I'm fairly certain the function is correct . . . and the output should be between 0 and 1 so i dont think they are too large. any ideas as to why raster calculator would return 0s or even where to look to find out what might be going on?