Error When Adding Random Slopes

[bryorsnef]

I'm trying to fit a model with two grouping factors and one covariate. One has many levels (more than 20,000) and one has only a few hundred. There are ~70,000 data. I'm getting an error whenever I add a random slope of the covariate by the factor with 200 levels. The models fits without issues if I exclude random slopes.

ERROR: MethodError: rmulΛ!(::BlockedSparse{Float64,2,1}, ::ReMat{Float64,1}) is ambiguous. Candidates:
  rmulΛ!(A::M, B::ReMat{T,1}) where {T, M<:AbstractArray{T,2}} in MixedModels at /home/bryor/.juliapro/JuliaPro_v1.3.1-1/packages/MixedModels/qyKOf/src/remat.jl:391
  rmulΛ!(A::BlockedSparse{T,S,P}, B::ReMat{T,P}) where {T, S, P} in MixedModels at /home/bryor/.juliapro/JuliaPro_v1.3.1-1/packages/MixedModels/qyKOf/src/remat.jl:405
Possible fix, define
  rmulΛ!(::Union{M<:BlockedSparse{T,S,1}, M<:BlockedSparse{T,S,1}, M<:(BlockedSparse{T,S,1} where S)}, ::Union{ReMat{T,1}, ReMat{T,1}, ReMat{T,1}})

Stacktrace:
 [1] updateL!(::LinearMixedModel{Float64}) at /home/bryor/.juliapro/JuliaPro_v1.3.1-1/packages/MixedModels/qyKOf/src/linearmixedmodel.jl:678
 [2] #fit!#38(::Bool, ::Bool, ::typeof(fit!), ::LinearMixedModel{Float64}) at /home/bryor/.juliapro/JuliaPro_v1.3.1-1/packages/MixedModels/qyKOf/src/linearmixedmodel.jl:251
 [3] (::StatsBase.var"#kw##fit!")(::NamedTuple{(:verbose, :REML),Tuple{Bool,Bool}}, ::typeof(fit!), ::LinearMixedModel{Float64}) at ./none:0
 [4] #fit#35(::Array{Any,1}, ::Dict{Symbol,Any}, ::Bool, ::Bool, ::typeof(fit), ::Type{LinearMixedModel}, ::StatsModels.FormulaTerm{StatsModels.Term,Tuple{StatsModels.Term,StatsModels.FunctionTerm{typeof(|),var"#25#27",(:Item,)},StatsModels.FunctionTerm{typeof(|),var"#26#28",(:FreqZipfUS, :Subject)}}}, ::NamedTuple{(:RT, :FreqZipfUS, :Subject, :Item),Tuple{CSV.Column{Float64,Float64},CSV.Column{Union{Missing, Float64},Union{Missing, Float64}},CategoricalArray{Int64,1,UInt32,Int64,CategoricalValue{Int64,UInt32},Union{}},CategoricalArray{String,1,UInt32,String,CategoricalString{UInt32},Union{}}}}) at /home/bryor/.juliapro/JuliaPro_v1.3.1-1/packages/MixedModels/qyKOf/src/linearmixedmodel.jl:116
 [5] (::StatsBase.var"#kw##fit")(::NamedTuple{(:wts, :contrasts, :verbose, :REML),Tuple{Array{Any,1},Dict{Symbol,Any},Bool,Bool}}, ::typeof(fit), ::Type{LinearMixedModel}, ::StatsModels.FormulaTerm{StatsModels.Term,Tuple{StatsModels.Term,StatsModels.FunctionTerm{typeof(|),var"#25#27",(:Item,)},StatsModels.FunctionTerm{typeof(|),var"#26#28",(:FreqZipfUS, :Subject)}}}, ::NamedTuple{(:RT, :FreqZipfUS, :Subject, :Item),Tuple{CSV.Column{Float64,Float64},CSV.Column{Union{Missing, Float64},Union{Missing, Float64}},CategoricalArray{Int64,1,UInt32,Int64,CategoricalValue{Int64,UInt32},Union{}},CategoricalArray{String,1,UInt32,String,CategoricalString{UInt32},Union{}}}}) at ./none:0
 [6] #fit#34(::Array{Any,1}, ::Dict{Symbol,Any}, ::Bool, ::Bool, ::typeof(fit), ::Type{LinearMixedModel}, ::StatsModels.FormulaTerm{StatsModels.Term,Tuple{StatsModels.Term,StatsModels.FunctionTerm{typeof(|),var"#25#27",(:Item,)},StatsModels.FunctionTerm{typeof(|),var"#26#28",(:FreqZipfUS, :Subject)}}}, ::DataFrame) at /home/bryor/.juliapro/JuliaPro_v1.3.1-1/packages/MixedModels/qyKOf/src/linearmixedmodel.jl:110
 [7] fit(::Type{LinearMixedModel}, ::StatsModels.FormulaTerm{StatsModels.Term,Tuple{StatsModels.Term,StatsModels.FunctionTerm{typeof(|),var"#25#27",(:Item,)},StatsModels.FunctionTerm{typeof(|),var"#26#28",(:FreqZipfUS, :Subject)}}}, ::DataFrame) at /home/bryor/.juliapro/JuliaPro_v1.3.1-1/packages/MixedModels/qyKOf/src/linearmixedmodel.jl:110
 [8] top-level scope at REPL[32]:1

[palday]

I know what's going wrong, but won't get a chance to fix it until probably next week.

In the meantime, can you give me the output of:

model = LinearMixedModel(@formula(your_formula_here), your_data_here)
describeblocks(model)

Alternatively, if you are able to share your data, even privately, that would be a great help in checking that I get the fix right. :)

[bryorsnef]

I can share the data.

julia_test.txt

[palday]

Cool, thanks, can you share your formula too? :) I'm guessing Y ~ X + (1|F1) (1|F2)? And there seem to be missing values in X -- am I parsing the file correctly?

[bryorsnef]

Yup. The missing values in X are genuine.

Y ~ X + (1|F1) + (1|F2) works fine.
Y ~ X + (1 + X|F1) + (1|F2) throws the error.

[palday]

@bryorsnef can you install the rankUpdate branch and see if that fixes your problem?

You can install development versions with:

julia>] add MixedModels#rankUpdate