CODEWITHSUNDEEP
pythonmultithreadingmachine-learningjuliapycall
Fatafim
Fatafim

Reputation: 307

Julia multithreading from pycall

Say I have a jupyter notebook:

%%julia

using Pkg
Pkg.add("DecisionTree")
using DecisionTree

X = Vector([1.1,2.2,3.3])
Y = Vector([1.1,2.2,3.3])
X = reshape(X, size(X))

X = Float32.(X)
Y = Float32.(Y)
print(typeof(X))
print(typeof(Y))
model = DecisionTree.build_forest(Y, X')

From what I know DecisionTree.jl uses multithreading, which pycall does not support, which results in the error:

RuntimeError: <PyCall.jlwrap (in a Julia function called from Python)
JULIA: TaskFailedException
Stacktrace:
  [1] wait
    @ .\task.jl:334 [inlined]
  [2] threading_run(func::Function)
    @ Base.Threads .\threadingconstructs.jl:38
  [3] macro expansion
    @ .\threadingconstructs.jl:97 [inlined]
  [4] build_forest(labels::Vector{Float32}, features::LinearAlgebra.Adjoint{Float32, Vector{Float32}}, n_subfeatures::Int64, n_trees::Int64, partial_sampling::Float64, max_depth::Int64,

My question is - is there any way to make it work after all?

Upvotes: 1

Views: 483

Answers (1)

Antonello
Antonello

Reputation: 6431

The problem has nothing to do with calling it from Python, but from the fact that you are trying to make a model where the features is a single record with 3 dimensions and the label is a 3 (records) vector. DecisionTrees expects indeed the input to be a column vector of dimension nRecords for the label and a nRecods by nDimensions matrix for the features.

For example:

julia> X = [1.1,2.2,3.3]
3-element Vector{Float64}:
 1.1
 2.2
 3.3

julia> Y = [1.1,2.2,3.3]
3-element Vector{Float64}:
 1.1
 2.2
 3.3

julia> X = reshape(X,3,1) # reshape to a single column **matrix**
3×1 Matrix{Float64}:
 1.1
 2.2
 3.3

julia> model = DecisionTree.build_forest(Y, X)
Ensemble of Decision Trees
Trees:      10
Avg Leaves: 1.0
Avg Depth:  0.0

Also, to make a vector you don't need to specify "Vector". I suggest you to have a look on my tutorial on Julia or on my course on Scientific Programming and Machine Learning with Julia (I completed it just a couple of days ago, I still need to "clean" it before announcing it)

Upvotes: 2

Related Questions