Need `mode` method for SecondOrder
ElOceanografo opened this issue · comments
I was trying to compare some second-order methods for Hessians using DifferentiationInterfaceTest, but found there is no mode
method defined for SecondOrder
backends:
julia> DifferentiationInterface.mode(SecondOrder(AutoSparseForwardDiff(), AutoReverseDiff()))
ERROR: MethodError: no method matching mode(::SecondOrder{AutoSparseForwardDiff{nothing, Nothing}, AutoReverseDiff})
Closest candidates are:
mode(::AutoChainRules{<:ChainRulesCore.RuleConfig{>:ChainRulesCore.HasReverseMode}})
@ DifferentiationInterfaceChainRulesCoreExt ~/.julia/packages/DifferentiationInterface/9POaB/ext/DifferentiationInterfaceChainRulesCoreExt/DifferentiationInterfaceChainRulesCoreExt.jl:17
mode(::AutoChainRules{<:ChainRulesCore.RuleConfig{>:ChainRulesCore.HasForwardsMode}})
@ DifferentiationInterfaceChainRulesCoreExt ~/.julia/packages/DifferentiationInterface/9POaB/ext/DifferentiationInterfaceChainRulesCoreExt/DifferentiationInterfaceChainRulesCoreExt.jl:16
mode(::Union{AutoSparseFiniteDiff, AutoSparseZygote, AutoSparseFastDifferentiation, AutoSparseSymbolics, AutoSparseReverseDiff, AutoSparseForwardDiff, AutoSparsePolyesterForwardDiff})
@ DifferentiationInterface ~/.julia/packages/DifferentiationInterface/9POaB/src/sparse.jl:37
...
Stacktrace:
[1] top-level scope
@ REPL[5]:1
...which causes an error when creating the table of benchmark results.
Good catch! I'll add this fix to the main branch so you can use it, but it won't be registered until next week or so because v0.3 is a big revamp