Empirical CDF enhancements #590
Open
+151
−52
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
use the custom broadcast implementation of ecdf.(v)
precalculates the partial weigths sums
linearly interpolates cdf between adjacent values
Merged
nalimilan
reviewed
Feb 8, 2021
| @@ -121,8 +121,8 @@ aweights(vs::RealArray) = AnalyticWeights(vec(vs)) | |||
| s = w.sum | |||
|
|
|||
| if corrected | |||
| sum_sn = sum(x -> (x / s) ^ 2, w) | |||
| 1 / (s * (1 - sum_sn)) | |||
| sum_w2 = sum(abs2, w) | |||
There was a problem hiding this comment.
Wasn't this here to avoid overflow?
|
|
||
| ### Deprecate August 2020 (v0.33) | ||
| function (ecdf::ECDF)(v::RealArray) | ||
| depwarn("(ecdf::ECDF)(v::RealArray) is deprecated, use `ecdf.(v)` broadcasting instead", "(ECDF::ecdf)(v::RealArray)") |
There was a problem hiding this comment.
More common, and possibly faster:
Suggested change
| depwarn("(ecdf::ECDF)(v::RealArray) is deprecated, use `ecdf.(v)` broadcasting instead", "(ECDF::ecdf)(v::RealArray)") | |
| depwarn("(ecdf::ECDF)(v::RealArray) is deprecated, use `ecdf.(v)` broadcasting instead", :ecdf) |
| @@ -1,68 +1,142 @@ | |||
| # Empirical estimation of CDF and PDF | |||
|
|
|||
| ## Empirical CDF | |||
| """ | |||
| Empirical Cumulative Distribution Function (ECDF). | |||
There was a problem hiding this comment.
Suggested change
| Empirical Cumulative Distribution Function (ECDF). | |
| ECDF{T <: Real, W <: Real, I} | |
| Empirical Cumulative Distribution Function (ECDF). |
| # - `ECDF(x[i])` | ||
| # - `1/(x[i+1] - x[i])` | ||
| # - `ECDF(x[i+1]) - ECDF(x[i])` (the weight of `x[i+1]`) | ||
| sorted_values::Vector{Tuple{T, W, W, W}} |
There was a problem hiding this comment.
Isn't it less efficient to store values as a vector of tuples instead of as several vectors? In terms of memory use AFAIK alignment may require padding between entries.
| r[ord[i]] = weightsum | ||
| i += 1 | ||
| # broadcasts ecdf() over an array | ||
| # caches the last calculated value |
| any(isnan, X) && throw(ArgumentError("ecdf can not include NaN values")) | ||
| isempty(weights) || length(X) == length(weights) || throw(ArgumentError("data and weight vectors must be the same size," * | ||
| "got $(length(X)) and $(length(weights))")) | ||
| evenweights = isnothing(weights) || isempty(weights) |
There was a problem hiding this comment.
Inference works better with ===:
Suggested change
| evenweights = isnothing(weights) || isempty(weights) | |
| evenweights = weights === nothing || isempty(weights) |
| "got $(length(X)) and $(length(weights))")) | ||
| T = eltype(X) | ||
| W0 = evenweights ? Int : eltype(weights) | ||
| W = isnothing(weights) ? Float64 : eltype(one(W0)/sum(weights)) |
There was a problem hiding this comment.
Store sum(weights) to avoid calling it twice?
Comment on lines
+117
to
+123
| push_valprev!() = push!(sorted_vals, (valprev, min(wsumprev/wsum, one(W)), | ||
| inv(val - valprev), valw/wsum)) | ||
|
|
||
| @inbounds for i in ord | ||
| valnew = X[i] | ||
| if (val != valnew) || (i == last(ord)) | ||
| (wsumprev > 0) && push_valprev!() |
There was a problem hiding this comment.
Doesn't this anonymous function trigger boxing of these variables, making the loop quite slow?
Comment on lines
+37
to
+39
| @test_skip fnecdf.weights == fnecdfalt.weights | ||
| @test_skip fnecdf.weights != w1 # check that w wasn't accidently modified in place | ||
| @test_skip fnecdfalt.weights != w2 |
| show(iobuf, obj) | ||
| return String(take!(iobuf)) | ||
| end | ||
|
|
||
| @testset "ECDF" begin | ||
| x = randn(10000000) | ||
| fnecdf = ecdf(x) |
There was a problem hiding this comment.
I'd change all of these to ecdf.(x), and copy this to deprecated.jl.
|
Is there anything here that's still worth it, or should I close? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The PR improves the performance of ECDF and adds interpolation.
Base.Broadcast.broadcasted()for providing enhanced support for vectors. The PR deprecatesecdf(v::AbstractVector)in favor of standard dot notation. I've kept customized broadcasting, but now it just caches the CDF value of the last vector element.ecdf(..., interpolate=true)), which is handy for continuous empirical distributions. (It just linearly interpolates between the CDFs of adjacent values).