24929: Updates the reduce_data algorithm, MAJOR

howso/remove_cases.amlg

@@ -270,52 +270,6 @@
 			)
 		)
 
-		(declare (assoc re_derivation_series_case_ids (null) ))
-
-		;check if time series dataset, if so, need to pull all cases from affected series so that their
-		;derived features can be-rederived below after cases are removed
-		(if (and (!= (null) !tsTimeFeature) (size !derivedFeaturesMap))
-			(let
-				(assoc series_id_features (get !tsFeaturesMap "series_id_features") )
-
-				;for each case pull the list of series id values for all the series id features
-				(declare (assoc
-					series_id_lists
-						(values (map
-							(lambda (retrieve_from_entity (current_value) series_id_features))
-							cases
-						) .true)
-				))
-
-				;for each series affected, get the cases so they can be re-derived
-				(declare (assoc
-					per_series_cases
-						(map
-							(lambda (let
-								(assoc
-									series_id_values (current_value 1)
-								)
-
-								(contained_entities
-									(values (map
-										(lambda
-											(query_equals (current_index) (current_value))
-										)
-										(zip series_id_features series_id_values)
-									))
-								)
-							))
-							series_id_lists
-						)
-				))
-
-				;collapse all lists of cases into one list of unique case ids
-				(assign (assoc
-					re_derivation_series_case_ids (apply "append" per_series_cases)
-				))
-			)
-		)
-
 		;remove all the cases
 		(apply "destroy_entities" cases)
 
@@ -370,16 +324,6 @@
 		;dataset has changed so clear out these cached value
 		(call !ClearCachedCountsAndEntropies)
 
-		;if there are cases that need to have features re-derived, do it here
-		(if re_derivation_series_case_ids
-			(call !DeriveTrainFeatures (assoc
-				features !trainedFeatures
-				;keep and derive only those features that are not in the features list
-				derived_features (get !tsFeaturesMap "ts_derived_features")
-				case_ids re_derivation_series_case_ids
-			))
-		)
-
 		(call !UpdateHasNulls (assoc features !trainedFeatures))
 
 		(call !UpdateRegionalMinSize (assoc

howso/scale.amlg

@@ -539,9 +539,6 @@
 						(size !hyperparameterMetadataMap)
 					)
 					(call reduce_data (assoc
-						abs_threshold_map !autoAblationAbsThresholdMap
-						delta_threshold_map !autoAblationDeltaThresholdMap
-						rel_threshold_map !autoAblationRelThresholdMap
 						skip_auto_analyze skip_auto_analyze
 					))
 

howso/train.amlg

@@ -982,9 +982,6 @@
 				)
 				(call reduce_data (assoc
 					features features
-					abs_threshold_map !autoAblationAbsThresholdMap
-					delta_threshold_map !autoAblationDeltaThresholdMap
-					rel_threshold_map !autoAblationRelThresholdMap
 					skip_auto_analyze skip_auto_analyze
 				))
 

unit_tests/ut_h_derive_start_end.amlg

@@ -139,7 +139,7 @@
 	))
 
 
-	;get the earlier case and check that its derived values are updated
+	;get the earlier case and check that its derived values stay the same
 	(assign (assoc
 		result
 			(get
@@ -158,21 +158,21 @@
 			)
 	))
 
-	(print "Validate case derived features are updated correctly: ")
+	(print "Validate case derived features don't change: ")
 	(call assert_approximate (assoc
 		obs mike_case_3
 		exp
 			(assoc
-				".series_index" 2
+				".series_index" 3
 				".reverse_series_index" 1
 				".session_training_index" 4
-				".time_delta_1" 1800
-				".time_lag_1" 1600
+				".time_delta_1" 1400
+				".time_lag_1" 2000
 				"name" "mike"
 				"time" 3400
 			)
 	))
-	(call exit_if_failures (assoc msg "Re-derived features correctly after case removal."))
+	(call exit_if_failures (assoc msg "derived features remain as they were after case removal."))
 
 	(call exit_if_failures (assoc msg unit_test_name ))
 )

unit_tests/ut_h_reduce_data.amlg

@@ -175,7 +175,7 @@
 		obs (get (call_entity "howso" "get_num_training_cases") (list 1 "payload" "count"))
 	))
 
-	(print "reduce_data with no thresholds reduces down to minimum model size: ")
+	(print "reduce_data reduces down to minimum model size: ")
 	(call_entity "howso" "reduce_data")
 	(call assert_same (assoc
 		exp
@@ -186,43 +186,9 @@
 		obs
 			(get (call_entity "howso" "get_num_training_cases") (list 1 "payload" "count"))
 	))
-	(call exit_if_failures (assoc msg "reduce_data with no thresholds reduced as expected"))
+	(call exit_if_failures (assoc msg "reduce_data reduced as expected"))
 	(call !EmptyAndTrain)
 
-	(print "reduce_data with very liberal thresholds reduces down to minimum model size: ")
-	(call_entity "howso" "reduce_data" (assoc
-		abs_threshold_map (assoc accuracy (assoc species 0.1))
-		delta_threshold_map (assoc accuracy (assoc species 0.5))
-		rel_threshold_map (assoc accuracy (assoc species 0.5))
-	))
-	(call assert_same (assoc
-		exp
-			(get
-				(call_entity "howso" "get_auto_ablation_params")
-				(list 1 "payload" "min_num_cases")
-			)
-		obs
-			(get (call_entity "howso" "get_num_training_cases") (list 1 "payload" "count"))
-	))
-	(call exit_if_failures (assoc msg "reduce_data with liberal thresholds reduced as expected"))
-	(call !EmptyAndTrain)
-
-	(print "reduce_data with conservative thresholds reduces to above minimum model size: ")
-	(call_entity "howso" "reduce_data" (assoc
-		abs_threshold_map (assoc accuracy (assoc species 1))
-		delta_threshold_map (assoc accuracy (assoc species 0.001))
-		rel_threshold_map (assoc accuracy (assoc species 0.001))
-	))
-	(call assert_same (assoc
-		exp
-			(-
-				(size training_data)
-				(call_entity "howso""debug_label" (assoc label "!ablationBatchSize"))
-			)
-		obs
-			(get (call_entity "howso" "get_num_training_cases") (list 1 "payload" "count"))
-	))
-	(call exit_if_failures (assoc msg "reduce_data with conservative thresholds reduced as expected"))
 
 	(print "a basic train automatically calls reduce_data when the threshold is low enough, resulting in min. model size w/ no threshold maps: ")
 	(call !EmptyAndTrain (assoc enable_auto_ablation .true max_num_cases 120))

unit_tests/ut_h_scale_ablation.amlg

@@ -33,11 +33,12 @@
     ))
 
     (print "set_auto_ablation_params\n")
-    (call_entity "howso" "set_auto_ablation_params" (assoc
-        auto_ablation_enabled .true
-        min_num_cases 100
-        max_num_cases 200
-    ))
+	(call_entity "howso" "set_auto_ablation_params" (assoc
+		auto_ablation_enabled .true
+		min_num_cases 100
+		max_num_cases 200
+		reduce_max_cases 100
+	))
 
     (print "train\n")
     (call_entity "howso" "train" (assoc
@@ -63,32 +64,20 @@
         obs (first first_pass_cases)
         exp 1
     ))
-    (print "only '998' and cases under '100' remain: ")
+    (print "only 100 cases remain: ")
     (call assert_true (assoc
-        obs
-			(apply "and"
-				(map
-					(lambda (or (< (first (current_value)) 100) (= (first (current_value)) 998)))
+        obs (=
+				100
+				(size
 					(get first_pass_cases (list 1 "payload" "cases"))
 				)
 			)
     ))
 
-    ;train some more cases with medium numbers.
-
-    (print "train\n")
-    (call_entity "howso" "train" (assoc
-        cases (call !CreateSquareCases (assoc xs (range 400 498)))
-        features (list "x" "y")
-        session "unit_test"
-    ))
-
-    ; set up training a specific number of medium-valued cases, anything near here was dropped in the first
-	; reduce_data call, but we've loaded in some duplicate cases.
-
+	;would be payload for cases in the 400-500 range
     (declare (assoc
         train_payload (call_entity "howso" "compute_train_payload" (assoc
-            cases (call !CreateSquareCases (assoc xs (range 450 455)))
+            cases (call !CreateSquareCases (assoc xs (range 400 500)))
             features (list "x" "y")
             session "unit_test"
         ))
@@ -100,12 +89,16 @@
     ))
 
     ; With that payload in hand, let's reduce_data again.
-
+	(call_entity "howso" "set_auto_ablation_params" (assoc
+		auto_ablation_enabled .true
+		min_num_cases 25
+		max_num_cases 100
+		reduce_max_cases 25
+	))
     (print "reduce_data\n")
-    (call_entity "howso" "reduce_data" (assoc))
-
-    ; This will again drop a lot of the medium-valued cases, so we're going to fail committing the payload.
+    (call_entity "howso" "reduce_data")
 
+    ; This will drop a lot of the cases we need, so we're going to fail committing the payload.
     (print "process_train_payload failed: ")
     (call assert_same (assoc
         obs (call_entity "howso" "process_train_payload" (get train_payload (list 1 "payload")))

version.json

@@ -1,6 +1,6 @@
 {
   "version": "0.0.0",
   "dependencies": {
-    "amalgam": "71.0.4"
+    "amalgam": "71.0.6"
   }
 }