FFT Javascript Implementation #599
Conversation
june128
added
the
Implementation
| return x.map((_, i) => x[indexes[i]]); | ||
| } | ||
|
|
||
| // Assumes log_2(N) is an integer |
There was a problem hiding this comment.
Why is this comment floating?
The Julia code also doesn't asume log2(n) to be an integer, does this output the same results?
There was a problem hiding this comment.
This is a radix-2 cooley-tukey implementation if I'm not mistaken, and requires that N be a power of two. The bit reversal algorithm also operates with radix-2 so it requires that N be a power of two, and thus that log2(N) be integral.
Here in particular, the displacement array is generated, and will always be of power-of-two length. The recursive nature of the bit_reverse_idxs function requires that the input be a positive integer or it will infinitely recurse as n goes into the negative non-integers.
| let zdiff = 0; | ||
| for (let i = 0; i < X.length; i++) { | ||
| ydiff += (Y[i].sub(T[i])).abs(); | ||
| zdiff += (Z[i].sub(T[i])).abs(); |
There was a problem hiding this comment.
In the Julia code the approximation-check has been made a seperate function, it might be a good idea to do that here as well.
There was a problem hiding this comment.
For consistency or syntactically? I personally think that it's fine where it is and that another function would be too much for this simple piece of code, but I'm not set in that opinion.
There was a problem hiding this comment.
I think it is a bit more clear in a function. At the very least, a comment would be appreciated.
|
I'm happy with the implementation and the line numbers seem correct. If @Trashtalk217 is happy with the javascript as well, I believe this can be merged. |
|
@Trashtalk217 If you have no further comments, I think we are ready to merge this one. |
There was a problem hiding this comment.
If @Trashtalk217 and @benchislett are happy, I am happy to approve!
Made use of Complex.js for dealing with complex numbers.
Implementation is fairly similar to the Python one, a key difference being the bit reversal step.
Instead of dealing with bits, the double-concat algorithm is implemented, described here.