Added Time Series Container and Block

Project.toml

@@ -4,6 +4,7 @@ authors = ["Lorenz Ohly", "Julia Community"]
 version = "0.3.0"
 
 [deps]
+ARFFFiles = "da404889-ca92-49ff-9e8b-0aa6b4d38dc8"
 Animations = "27a7e980-b3e6-11e9-2bcd-0b925532e340"
 BSON = "fbb218c0-5317-5bc6-957e-2ee96dd4b1f0"
 CSV = "336ed68f-0bac-5ca0-87d4-7b16caf5d00b"

src/FastAI.jl

@@ -109,6 +109,8 @@ export Vision
 include("Tabular/Tabular.jl")
 @reexport using .Tabular
 
+include("TimeSeries/TimeSeries.jl")
+@reexport using .TimeSeries
 
 include("interpretation/makie/stub.jl")
 function __init__()
@@ -155,6 +157,7 @@ export
     TableRow,
     Continuous,
     Image,
+    TimeSeriesRow,
 
     # encodings
     encode,

src/TimeSeries/TimeSeries.jl

@@ -0,0 +1,38 @@
+module TimeSeries
+
+using ..FastAI
+using ..FastAI:
+    # blocks
+    Block, WrapperBlock, AbstractBlock, OneHotTensor, OneHotTensorMulti, Label,
+    LabelMulti, wrapped, Continuous, getencodings, getblocks,
+    # encodings
+    Encoding, StatefulEncoding, OneHot,
+    # visualization
+    ShowText,
+    # other
+    FASTAI_METHOD_REGISTRY, registerlearningmethod!
+
+# for tests
+using ..FastAI: testencoding
+
+# extending
+import ..FastAI:
+    blockmodel, blockbackbone, blocklossfn, encode, decode, checkblock,
+    encodedblock, decodedblock, showblock!, mockblock, setup
+
+import Requires: @require
+import DataFrames: DataFrame, Not, select
+import UnicodePlots
+import ARFFFiles
+
+using FilePathsBase
+using InlineTest
+
+# Blocks
+include("blocks/timeseriesrow.jl")
+
+include("recipes.jl")
+
+export TimeSeriesRow
+
+end

src/TimeSeries/blocks/timeseriesrow.jl

@@ -0,0 +1,38 @@
+"""
+    TimeSeriesRow{M,N}() <: Block
+
+`Block` for a M variate time series with N number of time steps. `obs` is valid for `TimeSeriesRow{M,N}()`
+if it is an (M,N) dimensional Matrix with number element type.
+
+## Examples
+
+Creating a block:
+
+```julia
+TimeSeriesRow{1,51}()  # Univariate time series with length 51.
+TimeSeriesRow{2,51}()  # Multivariate time series with 2 variables and length 51.
+```
+"""
+struct TimeSeriesRow{M,N} <: Block end
+
+function checkblock(::TimeSeriesRow{M,N}, obs::AbstractArray{T,2}) where {M,N,T<:Number}
+    size(obs) == (M,N)
+end
+
+mockblock(::TimeSeriesRow{M,N}) where {M,N} = rand(Float64, (M,N))  
+
+function setup(::Type{TimeSeriesRow}, data)
+    # N,M = size(data[1,:,:])
+    N,M = size(getobs(data, 1))
+    return TimeSeriesRow{N,M}()
+end
+
+# visualization
+
+function showblock!(io, ::ShowText, block::TimeSeriesRow, obs)
+    plot = UnicodePlots.lineplot(obs[1,:])
+    for j=2:size(obs,1)
+        UnicodePlots.lineplot!(plot, obs[j,:])
+    end
+    print(io, plot)
+end

src/TimeSeries/recipes.jl

@@ -0,0 +1,38 @@
+Base.@kwdef struct TimeSeriesDatasetRecipe <: Datasets.DatasetRecipe
+    file = ""
+    targetcol = "target"
+end
+
+Datasets.recipeblocks(::Type{TimeSeriesDatasetRecipe}) = Tuple{TimeSeriesRow, Label} 
+
+function Datasets.loadrecipe(recipe::TimeSeriesDatasetRecipe, path)
+    path = convert(String, path)
+    datasetpath = joinpath(path, recipe.file)
+    df = ARFFFiles.load(DataFrame, datasetpath)
+    labels = Array(df[!, recipe.targetcol])
+    rows = Matrix(select(df, Not(:target)))
+    N,M = size(rows)
+    rows = reshape(rows, (N,1,M))
+    rows = TimeSeriesDataset(rows)
+    data = rows, labels
+    blocks = (
+        setup(TimeSeriesRow,rows),
+        Label(unique(eachobs(labels))),
+    )
+
+    return data, blocks
+end
+
+# Registering recipes
+
+const RECIPES = Dict{String,Vector{Datasets.DatasetRecipe}}(
+    "adiac" => [
+        TimeSeriesDatasetRecipe(file="Adiac_TRAIN.arff")
+    ]
+)
+
+function _registerrecipes()
+    for (name, recipes) in RECIPES, recipe in recipes
+        Datasets.registerrecipe!(Datasets.FASTAI_DATA_REGISTRY, name, recipe)
+    end
+end

src/datasets/Datasets.jl

@@ -63,6 +63,7 @@ export
     # primitive containers
     FileDataset,
     TableDataset,
+    TimeSeriesDataset,
 
     # utilities
     isimagefile,

src/datasets/containers.jl

@@ -91,6 +91,19 @@ LearnBase.nobs(dataset::TableDataset{<:DataFrame}) = nrow(dataset.table)
 LearnBase.getobs(dataset::TableDataset{<:CSV.File}, idx) = dataset.table[idx]
 LearnBase.nobs(dataset::TableDataset{<:CSV.File}) = length(dataset.table)
 
+# TimeSeriesDataset
+
+struct TimeSeriesDataset{}
+    table::AbstractArray{Float64,3}
+end 
+
+function LearnBase.getobs(dataset::TimeSeriesDataset, idx)
+    dataset.table[idx,:,:]
+end
+
+function LearnBase.nobs(dataset::TimeSeriesDataset) 
+    size(dataset.table)[1]
+end
 
 # ## Tests
 

src/datasets/fastaidatasets.jl

@@ -7,18 +7,20 @@ struct FastAIDataset
     datadepname
     subpath
     size
+    root_url
 end
 
-const ROOT_URL = "https://s3.amazonaws.com/fast-ai-"
+# const ROOT_URL = "https://s3.amazonaws.com/fast-ai-"
 
 function FastAIDataset(
         name, subfolder, checksum="";
         extension="tgz",
         description="",
         datadepname=name,
         subpath=name,
-        size="???")
-    return FastAIDataset(name, subfolder, extension, description, checksum, datadepname, subpath, size)
+        size="???",
+        root_url="https://s3.amazonaws.com/fast-ai-")
+    return FastAIDataset(name, subfolder, extension, description, checksum, datadepname, subpath, size, root_url)
 end
 
 DESCRIPTIONS = Dict(
@@ -105,6 +107,9 @@ const DATASETCONFIGS = [
     FastAIDataset("annotations_trainval2017", "coco", datadepname="coco-annotations_trainval2017", extension="zip"),
     FastAIDataset("stuff_annotations_trainval2017", "coco", datadepname="coco-stuff_annotations_trainval2017", extension="zip"),
     FastAIDataset("panoptic_annotations_trainval2017", "coco", datadepname="coco-panoptic_annotations_trainval2017", extension="zip"),
+
+    # time series
+    FastAIDataset("Adiac", "", "9c808bcfc77f3cab0a640bfa23e680ccde91da6981962e63c0f4f0737dcf6390", datadepname="adiac", extension="zip", root_url = "http://www.timeseriesclassification.com/Downloads", subpath="")
 ]
 
 const DATASETS = [d.datadepname for d in DATASETCONFIGS]
@@ -125,7 +130,7 @@ function DataDeps.DataDep(d::FastAIDataset)
 
         Download size: $(d.size)
         """,
-        "$(ROOT_URL)$(d.subfolder)/$(d.name).$(d.extension)",
+        "$(d.root_url)$(d.subfolder)/$(d.name).$(d.extension)",
         d.checksum,
         post_fetch_method=DataDeps.unpack,
     )