output.jl

### Persefone.jl - a model of agricultural landscapes and ecosystems in Europe.
###
### This file includes functions for saving the model output.
###

const LOGFILE = "simulation.log"

## Much of this code was adapted from the GeMM model by Leidinger et al.
## (https://github.com/CCTB-Ecomods/gemm/blob/master/src/output.jl)

"""
    createdatadir(outdir, overwrite)

Creates the output directory, dealing with possible conflicts.
"""
function createdatadir(outdir::String, overwrite::Union{Bool,String})
    if isdir(outdir)
        if overwrite == "ask"
            println("The chosen output directory $(outdir) already exists.")
            println("Type 'yes' to overwrite this directory. Otherwise, the simulation will abort.")
            print("Overwrite? ")
            answer = readline()
            overwrite = (answer == "yes" || answer == "y")
        end
        !overwrite ? Base.error("Output directory exists, will not overwrite. Aborting.") :
            @warn "Overwriting existing output directory $(outdir)."
        #TODO replace with exception
    end
    mkpath(outdir)
end

"""
    modellogger(loglevel, outdir)

Create a logger object that writes output both to screen and to a logfile.
This object is stored as `model.logger` and can then be used with `with_logger()`.
Note: requires [`createdatadir`](@ref) to be run first.
"""
function modellogger(loglevel::String, outdir::String)
    !isdir(outdir) && #TODO replace with exception
        Base.error("$(outdir) does not exist. Call `createdatadir()` before `modellogger()`.")
    loglevel == "debug" ? loglevel = Logging.Debug :
        loglevel == "warn" ? loglevel = Logging.Warn :
        loglevel == "info" ? loglevel = Logging.Info :
        Base.error("Invalid loglevel $loglevel, should be debug/info/warn.") #TODO make exception
    #XXX If this is a parallel run, should we turn off logging to screen?
    logfile = open(joinpath(outdir, LOGFILE), "w+")
    TeeLogger(ConsoleLogger(logfile, loglevel),
              ConsoleLogger(stdout, loglevel))
end

"""
    withtestlogger(model)

Replace the model logger with the currently active logger. This is intended to be used
in the testsuite to circumvent a [Julia issue](https://github.com/JuliaLang/julia/issues/48456),
where `@test_logs` doesn't work with local loggers.
"""
function withtestlogger(model::AgentBasedModel)
    # copied together from https://github.com/JuliaLang/julia/blob/master/base/logging.jl
    logstate = current_task().logstate
    logstate == nothing ? model.logger = global_logger() : model.logger = logstate.logger
    model
end

"""
    saveinputfiles(model)

Copy all input files into the output directory, including the actual parameter
settings used. This allows replicating a run in future.
"""
function saveinputfiles(model::AgentBasedModel)
    #XXX If this is a parallel run, we should save the global config to the top-level
    # output directory
    @debug "Setting up output directory $(@param(core.outdir))."
    currentcommit = read(`git rev-parse HEAD`, String)[1:8]
    open(joinpath(@param(core.outdir), basename(@param(core.configfile))), "w") do f
        println(f, "#\n# --- Persefone configuration parameters ---")
        println(f, "# This file was generated automatically.")
        println(f, "# Simulation run on $(string(Dates.format(Dates.now(), "d u Y HH:MM:SS"))),")
        # Record the current git commit and versions of dependencies for reproducibility
        println(f, "# with Persefone $(pkgversion(Persefone)), git commit $(currentcommit),")
        println(f, "# running on Julia $(VERSION) with Agents.jl $(pkgversion(Agents)).\n#\n")
        if !isempty(strip(read(`git status -s`, String)))
            println(f, "# WARNING: Your repository contains uncommitted changes. This may")
            println(f, "#          compromise the reproducibility of this simulation run.\n")
        end
        TOML.print(f, prepareTOML(model.settings))
    end
    # Copy the map files to the output folder
    lcmap = @param(world.landcovermap)
    ffmap = @param(world.farmfieldsmap)
    #TODO replace errors with exceptions
    !(isfile(lcmap)) && Base.error("The map file $(lcmap) doesn't exist.")
    !(isfile(ffmap)) && Base.error("The map file $(ffmap) doesn't exist.")
    cp(lcmap, joinpath(@param(core.outdir), basename(lcmap)), force = true)
    cp(ffmap, joinpath(@param(core.outdir), basename(ffmap)), force = true)
end

"""
    prepareTOML(dict)

An internal utility function to re-convert the one-dimensional dict created
by [`flattenTOML`](@ref) into the two-dimensional dict needed by `TOML.print`,
and convert any data types into TOML-compatible types where necessary.
"""
function prepareTOML(settings)
    # convert data types
    settings["core.loglevel"] == Logging.Debug ? settings["core.loglevel"] = "debug" :
        settings["core.loglevel"] == Logging.Warn ? settings["core.loglevel"] = "warn" :
        settings["core.loglevel"] = "info"
    # convert dict structure
    fulldict = Dict{String, Dict{String, Any}}()
    for parameter in keys(settings)
        domain, param = split(parameter, ".")
        !(domain in keys(fulldict)) && (fulldict[domain] = Dict{String,Any}())
        fulldict[domain][param] = settings[parameter]
    end
    fulldict
end

"""
    DataOutput

A struct for organising model output. This is used to collect model data
in an in-memory dataframe or for CSV output. Submodels can register their
own output functions using [`newdataoutput!`](@ref).

Struct fields:
    - name: a string identifier for the data collection (used as file name)
    - header: a list of column names
    - outputfunction: a function that takes a model object and returns data values to record (formatted as a vector of vectors)
    - frequency: how often to call the output function (daily/monthly/yearly/end/never)
    - plotfunction: a function that takes a model object and returns a Makie figure object (optional)
"""
struct DataOutput
    name::String
    header::Vector{String}
    outputfunction::Function
    frequency::String
    plotfunction::Union{Function,Nothing}
end

"""
    newdataoutput!(model, name, header, outputfunction, frequency)

Create and register a new data output. This function must be called by all submodels
that want to have their output functions called regularly.
"""
function newdataoutput!(model::AgentBasedModel, name::String, header::Vector{String},
                        outputfunction::Function, frequency::String,
                        plotfunction::Union{Function,Nothing}=nothing)
    if !(frequency in ("daily", "monthly", "yearly", "end", "never"))
        Base.error("Invalid frequency '$frequency' for $name.") #TODO replace with exception
    end
    ndo = DataOutput(name, header, outputfunction, frequency, plotfunction)
    append!(model.dataoutputs, [ndo])
    if frequency != "never"
        if @param(core.csvoutput)
            open(joinpath(@param(core.outdir), name*".csv"), "w") do f
                println(f, join(header, ","))
            end
        end
        if @param(core.storedata)
            df = DataFrame()
            for h in header
                df[!,h] = Any[] #XXX allow specifying types?
            end
            model.datatables[name] = df
        end
    end
end

"""
    outputdata(model, force=false)
 
Cycle through all registered data outputs and activate them according to their
configured frequency. If `force` is `true`, activate all outputs regardless
of their configuration.
"""
function outputdata(model::AgentBasedModel, force=false)
    #XXX enable output every X days, or weekly?
    #XXX all output functions except for "end" are run on the first update
    # -> should they all be run on the last update, too?
    startdate = @param(core.startdate)
    isnextmonth = d -> (day(d) == day(startdate))
    isnextyear = d -> (month(d) == month(startdate) && day(d) == day(startdate))
    for output in model.dataoutputs
        (!force && output.frequency == "never") && continue
        # check if this output should be activated today
        if force || (output.frequency == "daily") ||
            (output.frequency == "monthly" && isnextmonth(model.date)) ||
            (output.frequency == "yearly" && isnextyear(model.date)) ||
            (output.frequency == "end" && model.date == @param(core.enddate))
            data = output.outputfunction(model)
            if @param(core.csvoutput)
                open(joinpath(@param(core.outdir), output.name*".csv"), "a") do f
                    for row in data
                        println(f, join(row, ","))
                    end
                end                
            end
            if @param(core.storedata)
                for row in data
                    push!(model.datatables[output.name], row)
                end
            end
        end
    end
end

"""
    visualiseoutput(model)

Cycle through all data outputs and call their respective plot functions,
saving each figure to file.
"""
function visualiseoutput(model::AgentBasedModel)
    @debug "Visualising output."
    #CairoMakie.activate!() # make sure we're using Cairo
    for output in model.dataoutputs
        isnothing(output.plotfunction) && continue
        figure = output.plotfunction(model)
        save(joinpath(@param(core.outdir), output.name*"."*@param(core.figureformat)),
             figure)
    end
end

"""
    savemodelobject(model, filename)

Serialise a model object and save it to file for later reference.
Includes the current model and Julia versions for compatibility checking.

WARNING: produces large files (>100 MB) and takes a while to execute.
"""
function savemodelobject(model::AgentBasedModel, filename::String)
    object = Dict("model"=>model,
                  "modelversion"=>pkgversion(Persefone),
                  "juliaversion"=>VERSION)
    !endswith(filename, ".dat") && (filename *= ".dat")
    filename = joinpath(@param(core.outdir), filename)
    serialize(filename, object)
    @debug "Saved model object to $(filename)."
end