Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

handle interp for integers by casting and rounding (#71) #142

Merged
merged 9 commits into from
Apr 19, 2024
Merged

handle interp for integers by casting and rounding (#71) #142

Show file tree
Hide file tree
Changes from 6 commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
ee9856d
handle interp for integers by casting and rounding (#71)
kpa28-git Mar 5, 2024
eb15e82
Add a rounding mode and dispatch when calculating increment.
rofinn Apr 9, 2024
7493e61
Add interp test for exceeding endpoint values.
rofinn Apr 13, 2024
737bda6
Fix tests by clamping the inserted value.
rofinn Apr 13, 2024
2f7fc0a
interpolate integers in floating point using generators to avoid allo…
kpa28-git Apr 17, 2024
ddc8a9c
ignore RoundingMode for non-integers like floats
kpa28-git Apr 17, 2024
26ef7e0
Only dispatch on Unsigned increment calculation and rounding non-inte…
rofinn Apr 17, 2024
a2f8d07
refactor of interpolate integers in floating point using generators...
kpa28-git Apr 17, 2024
486816a
Drop the generator
rofinn Apr 19, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
54 changes: 43 additions & 11 deletions src/imputors/interp.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
"""
Interpolate(; limit=nothing)
Interpolate(; limit=nothing, r=nothing)

Performs linear interpolation between the nearest values in an vector.
The current implementation is univariate, so each variable in a table or matrix will
Expand All @@ -11,6 +11,8 @@ that all missing values will be imputed.

# Keyword Arguments
* `limit::Union{UInt, Nothing}`: Optionally limit the gap sizes that can be interpolated.
* `r::Union{RoundingMode, Nothing}`: Optionally specify a rounding mode.
Avoids `InexactError`s when interpolating over integers.

# Example
```jldoctest
Expand All @@ -34,9 +36,10 @@ julia> impute(M, Interpolate(; limit=2); dims=:rows)
"""
struct Interpolate <: Imputor
limit::Union{UInt, Nothing}
r::Union{RoundingMode, Nothing}
end

Interpolate(; limit=nothing) = Interpolate(limit)
Interpolate(; limit=nothing, r=nothing) = Interpolate(limit, r)

function _impute!(data::AbstractVector{<:Union{T, Missing}}, imp::Interpolate) where T
@assert !all(ismissing, data)
Expand All @@ -51,15 +54,8 @@ function _impute!(data::AbstractVector{<:Union{T, Missing}}, imp::Interpolate) w
gap_sz = (next_idx - prev_idx) - 1

if imp.limit === nothing || gap_sz <= imp.limit
diff = data[next_idx] - data[prev_idx]
incr = diff / T(gap_sz + 1)
val = data[prev_idx] + incr

# Iteratively fill in the values
for j in i:(next_idx - 1)
data[j] = val
val += incr
end
gen = _gen_interp(data[prev_idx], data[next_idx], gap_sz+1, imp.r)
_gen_set!(data, prev_idx, gen)
end

i = next_idx
Expand All @@ -72,3 +68,39 @@ function _impute!(data::AbstractVector{<:Union{T, Missing}}, imp::Interpolate) w

return data
end

"""
Set a vector slice over the values of a generator, starting from `after+1`
"""
function _gen_set!(v::AbstractVector, after::Integer, gen)
rofinn marked this conversation as resolved.
Show resolved Hide resolved
for (i, val) in enumerate(gen)
v[after+i] = val
end
end

"""
Return generator over interpolated values.
"""
function _gen_interp(a, b, n, ::Nothing)
rofinn marked this conversation as resolved.
Show resolved Hide resolved
inc = _calculate_increment(a, b, n)
(a + inc*i for i=1:n)
end

_gen_interp(a, b, n, r::RoundingMode) = _gen_interp(a, b, n, nothing)

function _gen_interp(a::T, b::T, n, ::Nothing) where {T<:Integer}
inc = _calculate_increment(a, b, n)
(convert(T, a + inc*i) for i=1:n)
rofinn marked this conversation as resolved.
Show resolved Hide resolved
end

function _gen_interp(a::T, b::T, n, r::RoundingMode) where {T<:Integer}
inc = _calculate_increment(a, b, n)
(round(T, a + inc*i, r) for i=1:n)
end

_calculate_increment(a, b, n) = (b - a) / n

function _calculate_increment(a::T, b::T, n) where {T<:Integer}
rofinn marked this conversation as resolved.
Show resolved Hide resolved
_calculate_increment(float(a), float(b), n)
end

51 changes: 50 additions & 1 deletion test/imputors/interp.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,10 +90,59 @@
@test ismissing(result[1])
@test ismissing(result[20])

# Test inexact error
# Test with UInt
c = [0x1, missing, 0x3, 0x4]
@test Impute.interp(c) == [0x1, 0x2, 0x3, 0x4]

# Test reverse case where the increment is negative
@test Impute.interp(reverse(c)) == [0x4, 0x3, 0x2, 0x1]

# Test inexact error (no rounding mode provided)
# https://github.com/invenia/Impute.jl/issues/71
c = [1, missing, 2, 3]
@test_throws InexactError Impute.interp(c)

# Test with UInt
c = [0x1, missing, 0x2, 0x3]
@test_throws InexactError Impute.interp(c)

# Test reverse case where the increment is negative
@test_throws InexactError Impute.interp(reverse(c))

# Test inexact cases with a rounding mode
c = [1, missing, 2, 3]
@test Impute.interp(c; r=RoundToZero) == [1, 1, 2, 3]

# Test with UInt
c = [0x1, missing, 0x2, 0x3]
@test Impute.interp(c; r=RoundNearest) == [0x1, 0x2, 0x2, 0x3]

# Test reverse case where the increment is negative
@test Impute.interp(reverse(c); r=RoundUp) == [0x3, 0x2, 0x2, 0x1]

# Test rounding doesn't cause values to exceed endpoint values
@test Impute.interp([1, missing, missing, 2]; r=RoundUp) == [1, 2, 2, 2]
@test Impute.interp([2, missing, missing, 1]; r=RoundUp) == [2, 2, 2, 1]
@test Impute.interp([1, missing, missing, 0]; r=RoundDown) == [1, 0, 0, 0]
@test Impute.interp([0x1, missing, missing, 0x0]; r=RoundDown) == [0x1, 0x0, 0x0, 0x0]

# Test long gaps (above .5 increment)
@test Impute.interp([2, fill(missing, 10)..., 8]; r=RoundNearest) == [2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8]
@test Impute.interp([0x2, fill(missing, 10)..., 0x8]; r=RoundNearest) == [0x2, 0x3, 0x3, 0x4, 0x4, 0x5, 0x5, 0x6, 0x6, 0x7, 0x7, 0x8]
@test Impute.interp([8, fill(missing, 10)..., 2]; r=RoundNearest) == [8, 7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2]
@test Impute.interp([0x8, fill(missing, 10)..., 0x2]; r=RoundNearest) == [0x8, 0x7, 0x7, 0x6, 0x6, 0x5, 0x5, 0x4, 0x4, 0x3, 0x3, 0x2]

# Test long gaps (at .5 increment)
@test Impute.interp([2, fill(missing, 11)..., 8]; r=RoundNearest) == [2, 2, 3, 4, 4, 4, 5, 6, 6, 6, 7, 8, 8]
@test Impute.interp([0x2, fill(missing, 11)..., 0x8]; r=RoundNearest) == [0x2, 0x2, 0x3, 0x4, 0x4, 0x4, 0x5, 0x6, 0x6, 0x6, 0x7, 0x8, 0x8]
@test Impute.interp([8, fill(missing, 11)..., 2]; r=RoundNearest) == [8, 8, 7, 6, 6, 6, 5, 4, 4, 4, 3, 2, 2]
@test Impute.interp([0x8, fill(missing, 11)..., 0x2]; r=RoundNearest) == [0x8, 0x8, 0x7, 0x6, 0x6, 0x6, 0x5, 0x4, 0x4, 0x4, 0x3, 0x2, 0x2]

# Test long gaps (below .5 increment)
@test Impute.interp([2, fill(missing, 12)..., 8]; r=RoundNearest) == [2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8]
@test Impute.interp([0x2, fill(missing, 12)..., 0x8]; r=RoundNearest) == [0x2, 0x2, 0x3, 0x3, 0x4, 0x4, 0x5, 0x5, 0x6, 0x6, 0x7, 0x7, 0x8, 0x8]
@test Impute.interp([8, fill(missing, 12)..., 2]; r=RoundNearest) == [8, 8, 7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2, 2]
@test Impute.interp([0x8, fill(missing, 12)..., 0x2]; r=RoundNearest) == [0x8, 0x8, 0x7, 0x7, 0x6, 0x6, 0x5, 0x5, 0x4, 0x4, 0x3, 0x3, 0x2, 0x2]
end

# TODO Test error cases on non-numeric types
Expand Down
Loading