Why does np.arccos(1.0) give nan if fed by np.arange?

Question

Can anyone replicate this? ``` import numpy as np print np.arccos(1.0) print np.arccos(1) for x in np.arange(0.7,1,0.05): print x print np.arccos(x) ``` Output: ``` 0.0 0.0 0.7 0.795398830184 0.75 0.722734247813 0.8 0.643501108793 0.85 0.55481103298 0.9 0.451026811796 0.95 0.317560429292 1.0 nan ``` Note the last value,which should be `np.arccos(1.0)`, is `nan`. When I do `np.arccos(1.0)` in the console it is `0.0` as I would expect. What is causing this behaviour? Note, I am using Python 2.7.6 Spyder IDE Win7

Accepted Answer

This is normal floating point inaccuracy. Adding 0.05 many times to 0.7 does not necessarily add up to 1 exactly.

Changing `print x` to `print repr(x)` outputs 1.0000000000000002 for the last `x`.

```
>>> np.arccos(1.0000000000000002)
__main__:1: RuntimeWarning: invalid value encountered in arccos
nan
```

Why does np.arccos(1.0) give nan if fed by np.arange?

Solution

Problem