add catch for empty resampling

Author: ablaom

src/resampling.jl

@@ -3,7 +3,7 @@
 const AbstractRow = Union{AbstractVector{<:Integer}, Colon}
 const TrainTestPair = Tuple{AbstractRow, AbstractRow}
 const TrainTestPairs = Union{
-    NTuple{<:Any,TrainTestPair},
+    Tuple{Vararg{TrainTestPair}},
     AbstractVector{<:TrainTestPair},
 }
 
@@ -102,6 +102,12 @@ const ERR_BAD_RESAMPLING_OPTION = ArgumentError(
         "of the form `(train_rows, test_rows)`"
 )
 
+const ERR_EMPTY_RESAMPLING_OPTION = ArgumentError(
+    "`resampling` cannot be emtpy. It must be an "*
+        "`MLJ.ResamplingStrategy` or a vector (or tuple) of tuples "*
+        "of the form `(train_rows, test_rows)`"
+)
+
 
 # ==================================================================
 ## RESAMPLING STRATEGIES
@@ -1440,6 +1446,7 @@ function evaluate!(
     # Note: `rows` and `repeats` are only passed to the final `PeformanceEvaluation`
     # object to be returned and are not otherwise used here.
 
+    isempty(resampling) && throw(ERR_EMPTY_RESAMPLING_OPTION)
     resampling isa TrainTestPairs || throw(ERR_BAD_RESAMPLING_OPTION)
 
     X = mach.args[1]()

test/resampling.jl

@@ -276,11 +276,15 @@ end
         model = DeterministicConstantRegressor()
         mach  = machine(model, X, y, cache=cache)
 
-        # check catch for bad `resampling` option:
+        # check catch for bad `resampling` options:
         @test_throws(
             MLJBase.ERR_BAD_RESAMPLING_OPTION,
             evaluate(model, X, y; resampling="junk", verbosity=0, acceleration=accel),
         )
+        @test_throws(
+            MLJBase.ERR_EMPTY_RESAMPLING_OPTION,
+            evaluate(model, X, y; resampling=[], verbosity=0, acceleration=accel),
+        )
 
         # check we can provide tuples of pairs, instead of vectors of pairs, in
         # `resampling` option: