make MLJBase.default_logger() the default logger in TunedModel(...)

src/tuned_models.jl

@@ -257,6 +257,10 @@ key                 | value
   regular [`PerformanceEvaluation`](@ref) objects to the history (accessed via the
   `:evaluation` key); the compact form excludes some fields to conserve memory.
 
+- `logger=default_logger()`: a logger for externally reporting model performance
+  evaluations, such as an `MLJFlow.Logger` instance. On startup,
+  `default_logger()=nothing`; use `default_logger(logger)` to set a global logger. 
+
 """
 function TunedModel(
     args...;
@@ -281,7 +285,7 @@ function TunedModel(
     check_measure=true,
     cache=true,
     compact_history=true,
-    logger=nothing
+    logger=MLJBase.default_logger()
     )
 
     # user can specify model as argument instead of kwarg:

test/tuned_models.jl

@@ -526,4 +526,29 @@ end
     @test first(evaluations) isa MLJBase.PerformanceEvaluation
 end
 
+struct  DummyLogger
+    buffer
+end
+
+MLJBase.log_evaluation(logger::DummyLogger, performance_evaluation) =
+    write(logger.buffer, performance_evaluation.measurement[1])
+
+@testset "default logger" begin
+    buffer = IOBuffer()
+    logger = DummyLogger(buffer)
+    default_logger(logger)
+    model1 = KNNRegressor(K=5)
+    model2 = KNNRegressor(K=3)
+    tmodel = TunedModel(models=[model1, model2], measure=l2)
+    mach = machine(tmodel, make_regression(10)...)
+    fit!(mach, verbosity=0)
+    seekstart(buffer)
+    @test all(report(mach).history) do entry
+        logger_measurement = read(buffer, Float64)
+        logger_measurement == entry.evaluation.measurement[1]
+    end
+    default_logger(nothing)
+    close(buffer)
+end
+
 true