Examples failing

[cscherrer]

Hi @NHDaly,

I saw another reference to Salsa today, and it struck me that this could be very effective for probabilistic programming! I found a very nice description of the ML example here:
94db4dd

But when I try this, I get

julia> s1 = let s = SalsaMLIncr.new_lr_runtime()
                  for i in 1:10
                      for _ in 1:100
                          SalsaMLIncr.insert_training_pair!(s, (i,), i + rand(-1.0:1.0))
                      end
                      # Time remains constant: O(the batch size, or, 100)
                      @time SalsaMLIncr.predict(s, (rand(),))
                  end
                  s
              end
ERROR: DerivedFunctionException: Error encountered while executing derived function:
MethodError: no method matching get_map_for_key(::Salsa._DefaultSalsaStorage.DefaultStorage, ::Salsa.DerivedKey{typeof(Main.SalsaMLIncr.lr_δweights_δmse), Tuple{Int64}})
Closest candidates are:
  get_map_for_key(::Salsa._DefaultSalsaStorage.DefaultStorage, ::KT, ::Type{RT}) where {TT, KT<:(Salsa.DerivedKey{<:Any, TT}), RT} at ~/git/Salsa.jl/src/default_storage.jl:101
  get_map_for_key(::Salsa._DefaultSalsaStorage.DefaultStorage, ::Salsa.InputKey) at ~/git/Salsa.jl/src/default_storage.jl:119

------ Salsa Trace -----------------
[3] @derived lr_learned_weights(::Runtime)
[2] @derived lr_learned_weights_unrolled(::Runtime, 1::Int64)
[1] @derived lr_δweights_δmse(::Runtime, 0::Int64)
------------------------------------

Stacktrace:
  [1] macro expansion
    @ ~/git/Salsa.jl/src/default_storage.jl:136 [inlined]
  [2] macro expansion
    @ ./lock.jl:228 [inlined]
  [3] _previous_output_internal(runtime::Salsa._TracingRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage}, key::Salsa.DerivedKey{typeof(Main.SalsaMLIncr.lr_δweights_δmse), Tuple{Int64}})
    @ Salsa._DefaultSalsaStorage ~/git/Salsa.jl/src/default_storage.jl:135
  [4] previous_output(rt::Salsa._TracingRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage})
    @ Salsa ~/git/Salsa.jl/src/runtime_tracing.jl:98
  [5] var"%%__user_lr_δweights_δmse"(s::Salsa._TracingRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage}, iteration::Int64)
    @ Main.SalsaMLIncr ~/git/Salsa.jl/examples/ML-LR-IVM/linear-regression-incremental.jl:68
  [6] _memoized_lookup_internal(runtime::Salsa._TracingRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage}, key::Salsa.DerivedKey{typeof(Main.SalsaMLIncr.lr_δweights_δmse), Tuple{Int64}})
    @ Salsa._DefaultSalsaStorage ~/git/Salsa.jl/src/default_storage.jl:257
  [7] memoized_lookup(rt::Salsa._TracingRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage}, dependency_key::Salsa.DerivedKey{typeof(Main.SalsaMLIncr.lr_δweights_δmse), Tuple{Int64}})
    @ Salsa ~/git/Salsa.jl/src/runtime_generic.jl:20
  [8] memoized_lookup_unwrapped
    @ ~/git/Salsa.jl/src/runtime_generic.jl:43 [inlined]
  [9] lr_δweights_δmse
    @ ~/git/Salsa.jl/src/derived_macro.jl:109 [inlined]
 [10] var"%%__user_lr_learned_weights_unrolled"(s::Salsa._TracingRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage}, iteration::Int64)
    @ Main.SalsaMLIncr ~/git/Salsa.jl/examples/ML-LR-IVM/linear-regression-incremental.jl:58
 [11] _memoized_lookup_internal(runtime::Salsa._TracingRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage}, key::Salsa.DerivedKey{typeof(Main.SalsaMLIncr.lr_learned_weights_unrolled), Tuple{Int64}})
    @ Salsa._DefaultSalsaStorage ~/git/Salsa.jl/src/default_storage.jl:257
 [12] memoized_lookup(rt::Salsa._TracingRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage}, dependency_key::Salsa.DerivedKey{typeof(Main.SalsaMLIncr.lr_learned_weights_unrolled), Tuple{Int64}})
    @ Salsa ~/git/Salsa.jl/src/runtime_generic.jl:20
 [13] memoized_lookup_unwrapped(rt::Salsa._TracingRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage}, dependency_key::Salsa.DerivedKey{typeof(Main.SalsaMLIncr.lr_learned_weights_unrolled), Tuple{Int64}})
    @ Salsa ~/git/Salsa.jl/src/runtime_generic.jl:43
 [14] lr_learned_weights_unrolled
    @ ~/git/Salsa.jl/src/derived_macro.jl:109 [inlined]
 [15] var"%%__user_lr_learned_weights"(s::Salsa._TracingRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage})
    @ Main.SalsaMLIncr ~/git/Salsa.jl/examples/ML-LR-IVM/linear-regression-incremental.jl:40
 [16] _memoized_lookup_internal(runtime::Salsa._TracingRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage}, key::Salsa.DerivedKey{typeof(Main.SalsaMLIncr.lr_learned_weights), Tuple{}})
    @ Salsa._DefaultSalsaStorage ~/git/Salsa.jl/src/default_storage.jl:257
 [17] memoized_lookup(rt::Salsa._TopLevelRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage}, dependency_key::Salsa.DerivedKey{typeof(Main.SalsaMLIncr.lr_learned_weights), Tuple{}})
    @ Salsa ~/git/Salsa.jl/src/runtime_generic.jl:20
 [18] memoized_lookup_unwrapped
    @ ~/git/Salsa.jl/src/runtime_generic.jl:43 [inlined]
 [19] lr_learned_weights(s::Salsa._TopLevelRuntime{Salsa.EmptyContext, Salsa._DefaultSalsaStorage.DefaultStorage})
    @ Main.SalsaMLIncr ~/git/Salsa.jl/src/derived_macro.jl:109
 [20] predict
    @ ~/git/Salsa.jl/examples/ML-LR-IVM/linear-regression-incremental.jl:28 [inlined]
 [21] macro expansion
    @ ./timing.jl:220 [inlined]
 [22] top-level scope
    @ ./REPL[45]:7

I guess something in the code changed to make this no longer work? Is there a way to get it going again?

BTW Salsa is really cool, I wish I had learned more about it when we worked together :)