diff --git a/Project.toml b/Project.toml
index 7002ab52a2c0beba001d16fbfc8e0cf4e2a589c1..ca5795b8266550c0da2ecf8b8ad7b94cfd6e39c7 100644
--- a/Project.toml
+++ b/Project.toml
@@ -1,7 +1,7 @@
name = "PlanarLarvae"
uuid = "c2615984-ef14-4d40-b148-916c85b43307"
authors = ["François Laurent", "Institut Pasteur"]
-version = "0.13"
+version = "0.14"
[deps]
DelimitedFiles = "8bb1440f-4735-579b-a4ab-409b98df4dab"
diff --git a/docs/src/index.md b/docs/src/index.md
index e232ff484d406255e16c2b850c9afc83f2ba19a8..289983d8fea3ad23e1dae54674a7ad15f5c978b2 100644
--- a/docs/src/index.md
+++ b/docs/src/index.md
@@ -253,6 +253,53 @@ larvatrack
derivedtype
```
+# Data loaders
+
+```@docs
+PlanarLarvae.Dataloaders
+```
+
+```@docs
+PlanarLarvae.Dataloaders.Repository
+```
+
+```@docs
+PlanarLarvae.Dataloaders.TimeWindow
+```
+
+```@docs
+PlanarLarvae.Dataloaders.TimeSegment
+```
+
+```@docs
+PlanarLarvae.Dataloaders.LazyIndex
+```
+
+```@docs
+PlanarLarvae.Dataloaders.DataLoader
+```
+
+```@docs
+PlanarLarvae.Dataloaders.ratiobasedsampling
+```
+
+```@docs
+PlanarLarvae.Dataloaders.capacitysampling
+```
+
+```@docs
+PlanarLarvae.Dataloaders.buildindex
+```
+
+```@docs
+PlanarLarvae.Dataloaders.samplesize!
+```
+
+```@docs
+PlanarLarvae.Dataloaders.sample
+```
+
+
# Index
```@index
diff --git a/src/Dataloaders.jl b/src/Dataloaders.jl
index ade162ea5efa3fc009a2742caebfca831813d68a..8edcde7951e1667881d5d289a0900c342f012240 100644
--- a/src/Dataloaders.jl
+++ b/src/Dataloaders.jl
@@ -17,7 +17,7 @@ using ..MWT: interpolate
using Random
export DataLoader, Repository, TimeWindow, ratiobasedsampling, buildindex, sample,
- extendedlength, prioritylabel, capacitysampling
+ extendedlength, prioritylabel, capacitysampling, samplesize!
"""
The default window type defines a time segment as the time of a segment-specific reference
@@ -61,6 +61,16 @@ function extendedlength(w::TimeWindow)
round(Int, extendedduration * w.samplerate) + 1
end
+"""
+ TimeSegment(trackid, anchortime, window, class, timeseries)
+
+Time segment representation that includes:
+* the ID of the originating track (attribute `trackid`),
+* the anchor time (*i.e.* the central timestamp, attribute `anchortime`),
+* the time window used to generate the time segment (attribute `window`),
+* the behavior label or label class (attribute `class`),
+* and the timeseries data (attribute `timeseries`).
+"""
struct TimeSegment
trackid
anchortime
@@ -117,6 +127,23 @@ function segment(file, window, trackid, step, class)
TimeSegment(trackid, anchortime, window, class, segmentdata)
end
+"""
+ Repository(rootpath)
+ Repository(rootpath, pattern; basename_only=false)
+ Repository(rootpath, fileselector)
+
+Data repository. Data file provider.
+
+Files are recursively sought for from root directory `rootpath`.
+
+Regular expression `pattern` can be used to control what files are included based on their
+relative path or filename (if `basename_only=true`).
+
+Instead, as a more general control mechanism, boolean function `fileselector` takes a
+[`PreloadedFile`](@ref Main.PlanarLarvae.Formats.PreloadedFile) as input argument.
+See also [`labelledfiles`](@ref Main.PlanarLarvae.Formats.labelledfiles), argument
+`selection_rule`.
+"""
struct Repository
root::String
files::Vector{Formats.PreloadedFile}
@@ -132,7 +159,7 @@ function Repository(root::String, pattern::Regex; basename_only::Bool=false)
file″= try
preload(file; shallow=true)
catch
- @warn "Cannot read labels from file" file
+ @warn "Cannot preload file" file
continue
end
push!(files, file″)
@@ -172,6 +199,12 @@ root(repo::Repository) = repo.root
files(repo::Repository) = repo.files
filepaths(repo) = [relpath(file.source, root(repo)) for file in files(repo)]
+"""
+ DataLoader(repository, window, index)
+
+Data loader as a simple combination of a datafile repository, a time window and an index of
+time segments.
+"""
struct DataLoader
repository
window
@@ -181,13 +214,13 @@ end
root(loader::DataLoader) = root(loader.repository)
files(loader::DataLoader) = files(loader.repository)
-function countlabels(loader::DataLoader; unload=false)
- countlabels(loader.repository, loader.window; unload=unload)
+function countlabels(loader::DataLoader; kwargs...)
+ countlabels(loader.repository, loader.window; kwargs...)
end
const Count = Dict{Union{String, Vector{String}}, Int}
-function countlabels(repository::Repository, window; unload=false)
+function countlabels(repository::Repository, window; unload=false, skiperrors=false)
counts = Dict{Formats.PreloadedFile, Count}()
ch = Channel() do ch
foreach(files(repository)) do file
@@ -196,7 +229,23 @@ function countlabels(repository::Repository, window; unload=false)
end
c = Threads.Condition()
Threads.foreach(ch) do file
- counts′= countlabels(Formats.getnativerepr(file), window)
+ data = try
+ Formats.getnativerepr(file)
+ catch
+ @error "Failed to load file" file=file.source
+ if skiperrors
+ return
+ else
+ rethrow()
+ end
+ end
+ counts′= try
+ countlabels(data, window)
+ catch
+ # "missing :tags field in NamedTuple" results from `drop_record!` in `sample`
+ @error "Failed to load labels; are you reusing a DataLoader?" file=file.source
+ rethrow()
+ end
unload && unload!(file; gc=true)
lock(c)
try
@@ -248,6 +297,16 @@ end
total(counts::Dict{<:Any, Int}) = counts
+"""
+ LazyIndex(sampler)
+ LazyIndex(maxcounts, targetcounts, sampler)
+
+Index of time segments.
+
+After calling [`buildindex`](@ref), attributes `maxcounts` and `targetcounts` are defined.
+
+See also [`ratiobasedsampling`](@ref) and [`capacitysampling`](@ref).
+"""
mutable struct LazyIndex
maxcounts
targetcounts
@@ -266,35 +325,71 @@ abstract type RatioBasedSampling end
struct ClassRatios <: RatioBasedSampling
selectors
majority_minority_ratio
- seed
+ rng
end
struct IntraClassRatios <: RatioBasedSampling
selectors
majority_minority_ratio
intraclass
- seed
+ rng
end
function withselectors(sampler::T, selectors) where {T}
T((field === :selectors ? selectors : getfield(sampler, field) for field in fieldnames(T))...)
end
-function ratiobasedsampling(selectors, majority_minority_ratio; seed=nothing)
- LazyIndex(ClassRatios(asselectors(selectors), majority_minority_ratio, seed))
+"""
+ ratiobasedsampling(selectors, majority_minority_ratio; seed=nothing, rng=nothing)
+ ratiobasedsampling(selectors, majority_minority_ratio, intraclass; ...)
+
+Important! Returns a [`LazyIndex`](@ref) object.
+
+Sample the labels considering two categories of classes: majority classes and minority
+classes.
+Minority classes share a same sample size (per class), aligned with the least represented
+class.
+Majority classes share another sample size, per default twice as large as that of minority
+classes.
+Argument `majority_minority_ratio` defines the sample size ratio between majority and
+minority classes.
+
+Classes are defined following arguments `selectors`. Per default, each label represents a
+class.
+When some observations come with multiple labels, one of these labels can be specified with
+argument `intraclass` so that the observations with this label are systematically included.
+"""
+function ratiobasedsampling(selectors, majority_minority_ratio; seed=nothing, rng=nothing)
+ if !isnothing(seed)
+ if isnothing(rng)
+ rng = Random.default_rng()
+ end
+ Random.seed!(rng, seed)
+ end
+ LazyIndex(ClassRatios(asselectors(selectors), majority_minority_ratio, rng))
end
-function ratiobasedsampling(selectors, majority_minority_ratio, intraclass; seed=nothing)
+function ratiobasedsampling(selectors, majority_minority_ratio, intraclass;
+ seed=nothing, rng=nothing,
+)
+ if !isnothing(seed)
+ if isnothing(rng)
+ rng = Random.default_rng()
+ end
+ Random.seed!(rng, seed)
+ end
intraclass = if intraclass isa Pair
Dict(asselector(intraclass.first) => intraclass.second)
else
Dict(asselector(selector) => f for (selector, f) in intraclass)
end
LazyIndex(IntraClassRatios(asselectors(selectors), majority_minority_ratio, intraclass,
- seed))
+ rng))
end
-init!(sampler) = isnothing(sampler.seed) || Random.seed!(sampler.seed)
+function init!(_)
+ @warn "init! is deprecated; calls to init! can safely be removed"
+end
classtype(sampler::RatioBasedSampling) = classtype(sampler.selectors)
classtype(::AbstractDict{T, <:Any}) where {T} = T
@@ -408,11 +503,25 @@ function countthresholds(sampler::RatioBasedSampling, counts)
countthresholds(counts, sampler.selectors, sampler.majority_minority_ratio)
end
+"""
+ buildindex(loader; unload=false, verbose=true, skiperrors=false)
+ buildindex(index, repository, window; ...)
+
+Build an index of time segments in a data repository, organized per label class.
+
+`unload=true` frees memory early; timeseries data are unallocated as soon as consumed.
+
+`verbose=true` prints a summary of time segment counts per label class.
+
+`skiperrors=true` skips data files (typically *trx.mat* files) that cannot be loaded.
+"""
function buildindex(loader::DataLoader; kwargs...)
buildindex(loader.index, loader.repository, loader.window; kwargs...)
end
-function buildindex(ix::LazyIndex, repository, window; unload=false, verbose=true)
+function buildindex(ix::LazyIndex, repository, window;
+ unload=false, verbose=true, skiperrors=false,
+)
sampler = ix.sampler
if hasproperty(sampler, :selectors) && isnothing(sampler.selectors)
anyfile = first(files(repository))
@@ -423,7 +532,7 @@ function buildindex(ix::LazyIndex, repository, window; unload=false, verbose=tru
ix.sampler = sampler = withselectors(sampler, asselectors(labels))
end
#
- ix.maxcounts = countlabels(repository, window; unload=unload)
+ ix.maxcounts = countlabels(repository, window; unload=unload, skiperrors=skiperrors)
ix.targetcounts = buildindex(sampler, ix)
#
if verbose
@@ -536,6 +645,21 @@ function prioritylabel(label; verbose=true)
return selector => priority_include
end
+"""
+ sample(f, loader, features=:spine)
+
+Sample time segments and apply function `f` on the segments from each file.
+
+Function `f` takes the following arguments:
+* file index (`Int`),
+* file object ([`PreloadedFile`](@ref Main.PlanarLarvae.Formats.PreloadedFile),
+* couple of:
+ * cumulated count of time segments prior to current file (`Int`),
+ * expected count of time segments for current file (`Int`),
+* list of time segments (`Vector{TimeSegment}`).
+
+See also [`TimeSegment`](@ref).
+"""
function sample(f, loader::DataLoader, features=:spine; kwargs...)
ch = Channel() do ch
state = nothing
@@ -553,7 +677,6 @@ function sample(f, loader::DataLoader, features=:spine; kwargs...)
end
function sample(file::Formats.PreloadedFile, window, ix::LazyIndex, features; kwargs...)
- init!(ix.sampler)
sample(ix.sampler, file, window, ix.targetcounts[file], features; kwargs...)
end
@@ -581,10 +704,12 @@ function sample(sampler, file, window, counts, features; verbose=false)
@info "In file: $(file.source)\nSample sizes (observed => selected):" [(label isa Vector ? Symbol(Symbol.(label)) : Symbol(label)) => (count => get(counts, label, 0)) for (label, count) in pairs(observedcounts)]...
end
+ rng = isnothing(sampler.rng) ? Random.default_rng() : sampler.rng
+
# pick time segments at random to achieve the desired class counts
for (label, count) in pairs(counts)
label in keys(index) || continue
- pointers = shuffle(index[label])
+ pointers = shuffle(rng, index[label])
# counts should not exceed the actual numbers
index[label] = pointers[1:count]
end
@@ -642,15 +767,30 @@ presample(_, cumulatedcount, _, _, counts) = (sum(cumulatedcount), sum(values(co
struct CapacitySampling <: RatioBasedSampling
selectors
maxcount::Integer
- seed
+ rng
end
-function capacitysampling(selectors, maxcount::Integer; seed=nothing)
- LazyIndex(CapacitySampling(asselectors(selectors), maxcount, seed))
+"""
+ capacitysampling(maxcount; seed=nothing, rng=nothing)
+ capacitysampling(selectors, maxcount; ...)
+
+Important! Returns a [`LazyIndex`](@ref) object.
+
+Sample each class up to a maximum count `maxcount`.
+Class definition can be controlled with argument `selectors`.
+"""
+function capacitysampling(selectors, maxcount::Integer; seed=nothing, rng=nothing)
+ if !isnothing(seed)
+ if isnothing(rng)
+ rng = Random.default_rng()
+ end
+ Random.seed!(rng, seed)
+ end
+ LazyIndex(CapacitySampling(asselectors(selectors), maxcount, rng))
end
-function capacitysampling(maxcount::Integer; seed=nothing)
- capacitysampling(nothing, maxcount; seed=seed)
+function capacitysampling(maxcount::Integer; kwargs...)
+ capacitysampling(nothing, maxcount; kwargs...)
end
function ratio(sampler::CapacitySampling, counts)
@@ -659,4 +799,79 @@ function ratio(sampler::CapacitySampling, counts)
return ratios
end
+"""
+ samplesize!(index, n)
+
+Adjust target counts so that the total sample size is `n`.
+This should be performed after [`buildindex`](@ref) and before [`sample`](@ref).
+
+See also [`capacitysampling`](@ref) for per-class control of sample size.
+"""
+function samplesize!(index, sample_size)
+ @assert !isnothing(index.targetcounts) "buildindex must be called before samplesize!"
+
+ total_sample_size = length(index)
+ sample_size < total_sample_size || return index
+
+ ratio = sample_size / total_sample_size
+
+ rng = isnothing(index.sampler.rng) ? Random.default_rng() : index.sampler.rng
+
+ # apply `ratio` to the total counts first
+ maxcounts = Dataloaders.total(index.maxcounts)
+ targetcounts = Dataloaders.total(index.targetcounts)
+ for (label, count) in pairs(targetcounts)
+ targetcounts[label] = round(Int, count * ratio)
+ end
+ totalcount = sum(values(targetcounts))
+ if totalcount < sample_size
+ for label in shuffle(rng, collect(keys(targetcounts)))
+ if targetcounts[label] < maxcounts[label]
+ targetcounts[label] += 1
+ totalcount += 1
+ totalcount == sample_size && break
+ end
+ end
+ elseif sample_size < totalcount
+ for label in shuffle(rng, collect(keys(targetcounts)))
+ if 0 < targetcounts[label]
+ targetcounts[label] -= 1
+ totalcount -= 1
+ totalcount == sample_size && break
+ end
+ end
+ end
+
+ # apply `ratio` at the per-file level
+ targetcountsperfile = copy(index.targetcounts)
+ for counts in values(targetcountsperfile)
+ for (label, count) in pairs(counts)
+ counts[label] = round(Int, count * ratio)
+ end
+ end
+ targetcounts′= Dataloaders.total(targetcountsperfile)
+ for (label, targetcount) in pairs(targetcounts)
+ count = targetcounts′[label]
+ while count < targetcount
+ file = rand(rng, collect(keys(targetcountsperfile)))
+ counts = targetcountsperfile[file]
+ if haskey(counts, label) && counts[label] < index.maxcounts[file][label]
+ counts[label] += 1
+ count += 1
+ end
+ end
+ while targetcount < count
+ file = rand(rng, collect(keys(targetcountsperfile)))
+ counts = targetcountsperfile[file]
+ if 0 < get(counts, label, 0)
+ counts[label] -= 1
+ count -= 1
+ end
+ end
+ end
+
+ index.targetcounts = targetcountsperfile
+ return index
+end
+
end
diff --git a/src/MWT.jl b/src/MWT.jl
index 2fb6a7ba4ff5c57fd65a658ca637e2b785b40de7..39e9c173e364d859a49fea9cc666cc7ad685a06a 100644
--- a/src/MWT.jl
+++ b/src/MWT.jl
@@ -76,7 +76,7 @@ interpolate(x1::T, x2::T, lambda) where {T<:AbstractFloat} = (1 - lambda) * x1 +
interpolate(x1::T,
x2::T,
lambda,
- ) where {T<:AbstractArray} = @. (1 - lambda) * x1 + lambda * x2
+ ) where {T<:AbstractArray{<:AbstractFloat}} = @. (1 - lambda) * x1 + lambda * x2
function interpolate(geometry1::T, geometry2::T, lambda) where {T<:Larva.SpineGeometry}
x1, y1 = collect.(zip(coordinates.(Larva.vertices′(geometry1))...))
x2, y2 = collect.(zip(coordinates.(Larva.vertices′(geometry2))...))
@@ -84,7 +84,8 @@ function interpolate(geometry1::T, geometry2::T, lambda) where {T<:Larva.SpineGe
Larva.PointSeries(interpolate(x1, x2, lambda),
interpolate(y1, y2, lambda)))
end
-function interpolate(tags1::T, tags2::T, lambda) where {T<:Larva.AbstractTags}
+function interpolate(tags1, tags2, lambda)
+ # default nearest-neighbor implementation for tags/labels
lambda <= 0.5 ? tags1 : tags2
end
function interpolate(state1::T, state2::T, lambda) where {T<:NamedTuple}
diff --git a/test/runtests.jl b/test/runtests.jl
index ae75658551b42f80bcecc35ca9fbddb4a9e21dda..c5404c77383331aa0acca4980248322cb218fd74 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -915,7 +915,7 @@ if all_tests || "Dataloaders" in ARGS
# @test_warn exists but does not catch warnings...
Repository(joinpath(dir, "**", "trx.mat"))
else
- @test_warn "Cannot read labels from file" Repository(joinpath(dir, "**", "trx.mat"))
+ @test_warn "Cannot preload file" Repository(joinpath(dir, "**", "trx.mat"))
end
@test length(repo4) == 1 && basename(dirname(Dataloaders.filepaths(repo4)[1])) == "20140918_170215"
#
@@ -923,7 +923,7 @@ if all_tests || "Dataloaders" in ARGS
index4 = capacitysampling(60; seed=347980)
loader4 = DataLoader(repo4, window4, index4)
buildindex(loader4; verbose=false)
- @test index4.sampler.maxcount == 60 && index4.sampler.seed == 347980
+ @test index4.sampler.maxcount == 60# && index4.sampler.seed == 347980
@test Set(keys(index4.sampler.selectors)) == Set([:back, :back_large, :back_strong, :back_weak, :cast, :cast_large, :cast_strong, :cast_weak, :hunch, :hunch_large, :hunch_strong, :hunch_weak, :roll, :roll_large, :roll_strong, :roll_weak, :run, :run_large, :run_strong, :run_weak, :stop, :stop_large, :stop_strong, :stop_weak, :small_motion])
counts4 = copy(first(values(index4.targetcounts)))
counts4′= pop!(counts4, ["back", "back_weak", "small_motion"])
@@ -938,8 +938,22 @@ if all_tests || "Dataloaders" in ARGS
@test lengths == if VERSION < v"1.7"
[66, 66, 65, 66, 66, 66, 66, 66, 65, 65, 66, 65, 66, 66, 66, 66, 66, 65, 65, 66, 66, 66, 66, 65, 65, 66, 65, 66, 63, 64, 65, 67, 65, 66, 65, 65, 66, 66, 65, 66, 65, 65, 63, 63, 63, 66, 66, 65, 66, 66, 65, 65, 65, 66, 66, 66, 66, 65, 66, 66, 66, 66, 65, 65, 66, 66, 66, 65, 65, 65, 65, 63, 63, 65, 66, 66, 66, 66, 66, 65, 66, 66, 65, 65, 65, 65, 65, 66, 66, 66, 66, 63, 63, 65, 65, 65, 65, 63, 66, 67, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 66, 66, 66, 65, 65, 65, 66, 65, 66, 66, 66, 65, 65, 65, 65, 65, 66, 65, 63, 63, 63, 63, 63, 66, 65, 66, 65, 66, 64, 65, 66, 64, 64, 64, 64, 65, 65, 65, 65, 66, 66, 66, 66, 63, 66, 66, 66, 66, 66, 66, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 66, 66, 66, 66, 64, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 62, 63, 64, 63, 65, 65, 66, 66, 66, 66, 66, 66, 66, 66, 66, 65, 64, 65, 64, 65, 66, 66, 65, 65, 66, 66, 66, 65, 65, 66, 66, 66, 66, 66, 65, 66, 66, 66, 65, 64, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 65, 65, 65, 65, 65, 64, 64, 64, 64, 64, 64, 64, 63, 63, 65, 65, 65, 66, 66, 66, 66, 66, 66, 66, 66, 66, 65, 66, 66, 66, 65, 65, 66, 66, 66, 66, 65, 65, 66, 65, 65, 65, 65, 65, 65, 65, 65, 66, 66, 65, 65, 65, 65, 65, 66, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 66, 66, 66, 65, 66, 66, 65, 65, 65, 64, 64, 64, 65, 65, 65, 65, 65, 65, 65, 63, 63, 63, 63, 65, 63, 63, 64, 64, 64, 64, 64, 66, 66, 66, 65, 65, 65, 66, 66, 66, 66, 66, 64, 64, 64, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 65, 65, 66, 66, 66, 65, 65, 65, 66, 65, 66, 65, 65, 65, 65, 65, 65, 65, 66, 65, 65, 66, 66, 65, 64, 64, 64, 65, 64, 64, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 64, 63, 63, 66, 65, 63, 66, 66, 66, 66, 65, 66, 66, 66, 66, 66, 66, 66, 65, 65, 66, 66, 66, 66, 65, 63, 64, 64, 65, 64, 64, 65, 66, 66, 66, 66, 65, 66, 66, 66, 63, 63, 63, 64, 64, 64, 65, 64, 64, 64, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 62, 62, 62, 62, 63, 63, 64, 64, 66, 64, 64, 65, 65, 63, 65, 64, 64, 66, 65, 64, 64, 63]
else
- [66, 66, 66, 65, 66, 65, 66, 66, 66, 66, 66, 65, 66, 65, 66, 66, 66, 66, 65, 66, 65, 65, 66, 65, 66, 65, 66, 62, 63, 64, 66, 65, 66, 65, 65, 65, 67, 65, 66, 64, 65, 66, 66, 66, 66, 66, 66, 65, 65, 66, 66, 66, 66, 66, 66, 65, 65, 65, 65, 66, 66, 66, 65, 66, 65, 66, 65, 63, 63, 63, 66, 66, 66, 66, 65, 65, 65, 66, 65, 66, 66, 66, 66, 65, 66, 66, 65, 66, 66, 66, 65, 65, 64, 64, 63, 63, 64, 65, 65, 65, 65, 63, 63, 63, 63, 63, 63, 66, 65, 66, 65, 66, 66, 66, 66, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 66, 65, 66, 66, 65, 65, 66, 65, 65, 66, 65, 65, 65, 65, 66, 64, 64, 64, 63, 66, 66, 66, 63, 62, 65, 65, 66, 65, 65, 66, 64, 64, 64, 65, 63, 63, 63, 63, 63, 63, 64, 63, 65, 66, 66, 65, 66, 65, 66, 65, 66, 65, 65, 66, 65, 65, 65, 65, 65, 65, 66, 66, 66, 66, 64, 65, 65, 64, 63, 63, 63, 63, 62, 63, 63, 64, 64, 64, 64, 65, 66, 66, 66, 66, 66, 65, 66, 66, 66, 66, 66, 65, 66, 66, 66, 67, 65, 65, 65, 64, 63, 66, 65, 66, 66, 66, 65, 66, 66, 65, 64, 64, 64, 64, 64, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 64, 64, 65, 65, 65, 65, 64, 64, 64, 64, 64, 64, 63, 62, 66, 65, 66, 67, 66, 65, 65, 66, 66, 66, 66, 66, 66, 66, 66, 66, 66, 65, 65, 65, 66, 66, 66, 66, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 66, 66, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 66, 66, 65, 66, 66, 66, 65, 65, 64, 64, 64, 65, 65, 65, 65, 65, 65, 65, 64, 64, 64, 63, 64, 64, 63, 62, 64, 64, 64, 64, 64, 64, 66, 66, 65, 66, 65, 66, 65, 65, 65, 66, 65, 63, 63, 63, 63, 64, 64, 64, 64, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 64, 65, 65, 66, 66, 65, 65, 65, 65, 65, 65, 65, 66, 66, 66, 66, 64, 63, 63, 64, 65, 65, 64, 64, 64, 63, 63, 63, 63, 64, 64, 64, 64, 64, 64, 64, 64, 63, 63, 63, 64, 66, 66, 66, 65, 65, 66, 66, 65, 66, 66, 66, 65, 66, 66, 65, 65, 65, 66, 66, 66, 66, 66, 63, 63, 65, 66, 66, 63, 63, 63, 64, 64, 64, 64, 65, 65, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 64, 64, 65, 64, 63, 63, 64, 66, 63, 64, 64, 64, 64, 63, 65, 65, 63, 63, 64]
+ [65, 65, 66, 64, 66, 66, 67, 66, 66, 66, 65, 66, 66, 65, 66, 65, 65, 66, 66, 65, 66, 66, 65, 65, 66, 66, 66, 65, 65, 66, 66, 66, 66, 66, 65, 65, 66, 64, 64, 66, 66, 66, 66, 66, 66, 65, 66, 65, 64, 63, 63, 64, 65, 66, 66, 66, 66, 66, 66, 66, 66, 65, 65, 65, 65, 66, 66, 66, 66, 66, 66, 65, 65, 66, 66, 65, 65, 64, 63, 63, 63, 62, 64, 66, 66, 66, 66, 66, 66, 65, 66, 66, 66, 65, 66, 66, 66, 65, 66, 65, 66, 64, 64, 63, 65, 65, 65, 64, 63, 65, 65, 65, 65, 65, 65, 65, 65, 65, 66, 65, 66, 65, 66, 66, 66, 66, 65, 66, 65, 65, 65, 65, 65, 65, 65, 65, 65, 66, 66, 66, 63, 63, 66, 66, 66, 66, 66, 65, 65, 66, 66, 64, 63, 63, 64, 66, 65, 65, 63, 64, 64, 64, 63, 63, 63, 64, 64, 65, 66, 65, 65, 65, 64, 66, 65, 65, 66, 65, 66, 66, 65, 65, 65, 66, 65, 65, 65, 65, 65, 66, 65, 65, 65, 65, 65, 65, 65, 66, 66, 66, 66, 66, 66, 65, 65, 66, 65, 64, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 64, 64, 63, 66, 66, 66, 66, 66, 66, 66, 66, 66, 66, 65, 64, 65, 65, 67, 65, 65, 63, 63, 63, 62, 66, 66, 66, 66, 66, 66, 66, 66, 66, 65, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 65, 65, 65, 65, 64, 64, 64, 64, 64, 64, 64, 63, 63, 63, 63, 63, 63, 64, 66, 65, 66, 65, 66, 66, 66, 66, 66, 66, 66, 66, 66, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 66, 66, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 65, 66, 65, 65, 66, 65, 65, 65, 64, 64, 64, 65, 65, 65, 65, 65, 65, 65, 64, 64, 63, 63, 63, 63, 64, 64, 64, 64, 64, 64, 63, 66, 66, 66, 66, 65, 66, 65, 65, 65, 65, 65, 66, 66, 64, 64, 64, 64, 64, 64, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 64, 64, 64, 63, 65, 66, 65, 66, 66, 66, 66, 65, 65, 66, 65, 65, 64, 63, 64, 64, 64, 64, 64, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 62, 63, 63, 64, 64, 64, 65, 65, 64, 65, 63, 63, 64, 63, 66, 66, 65, 65, 66, 64, 65, 65, 64, 66, 66, 66, 65, 66, 66, 65, 66, 66, 65, 65, 68, 66, 66, 65, 66, 66, 66, 63, 66, 66, 66, 66, 66, 63, 63, 63, 64, 64, 64, 64, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 63, 64, 65, 66, 64, 65, 64, 63]
end
+ #
+ repo5 = Repository(run1)
+ window5 = TimeWindow(1.5, 10)
+ index5 = ratiobasedsampling(["back", "bend", "hunch", "run", "stop", "small_action"], 1.5; seed=347980)
+ loader5 = DataLoader(repo5, window5, index5)
+ buildindex(loader5; verbose=false)
+ original_sample_size = length(loader5.index)
+ samplesize!(loader5.index, 19)
+ @test original_sample_size == 121 && length(loader5.index) == 19
+ lengths′= nothing # set scope
+ Dataloaders.sample(loader5, :spine) do _, _, _, segments
+ lengths′= [length(segment.timeseries) for segment in segments]
+ end
+ @test all(==(16), lengths′) && length(lengths′) == 19
end
end