I have an angle in degrees and I want it to be encoded between $-1, ..., 1$ to make it easier for using in a neural network.
I thought it might be a good idea to first convert the angle into radians and then take the sin of it. To get the angle back in degrees, first use arcsin and then convert to degrees.
In python (using NumPy math functions) this would look something like this:
np.rad2deg(np.arcsin(np.sin(np.deg2rad(264.0))))
but instead of 264 this produces -84 : why is that?

Let's work this from the inside out:
In short, the way that $\arcsin$ is conventionally defined ensures that your expression will always return a value in the interval $[-90, 90]$. If your original angle is outside that range, you'll get a different angle back.