feat: Implements Kaiser Window #27
Conversation
polvalente
left a comment
polvalente
left a comment
There was a problem hiding this comment.
This is looking good! I left a few small comments.
| deftransformp pop_window_size(opts) do | ||
| {n, opts} = Keyword.pop(opts, :n) | ||
|
|
There was a problem hiding this comment.
We should refactor all functions here to be deftransforms where n is the first argument instead. Let's do this in a separate PR though.
| Nx.exp(abs_x) / Nx.sqrt(2 * Nx.Constants.pi() * abs_x) * | ||
| (1.0 + 1.0 / (8.0 * abs_x) + 9.0 / (128.0 * Nx.pow(abs_x, 2))) | ||
|
|
||
| Nx.select(abs_x < 3.75, small_x_result, large_x_result) |
There was a problem hiding this comment.
If you use if abs_x < 3.75 do ... else ... end, Nx will compile it to a cond block which is a bit more efficient in that it'll run one branch only
There was a problem hiding this comment.
I don't think the if would work in this case, abs_x is not a scalar.
lib/nx_signal/windows.ex
Outdated
| end | ||
| end | ||
|
|
||
| defnp i0(x) do |
There was a problem hiding this comment.
should probably rename this function to something more meaningful, or at least kaiser_i0
ycastorium
force-pushed
the
adds_kaiser_window
branch
2 times, most recently
from
7cb46b4 to
32fa8e3
Compare
ycastorium
force-pushed
the
adds_kaiser_window
branch
from
62368c4 to
183f8d3
Compare
|
Thanks! Merging this and I'll apply the refactor I suggested for extracting |
2 participants
Similar solutions usually go with the Chebyshev polynomials, i decided to use the Polynomial/Asymptotic approach instead, the difference should not be that perceptible.
I can change do Chebyshev though.