added random sort of training data for better training results (#48)

.env.example

@@ -24,6 +24,7 @@ ML_SPAM_DELETION_ENABLED=false
 ML_SPAM_AUTOBAN_ENABLED=true
 ML_SPAM_AUTOBAN_SCORE_THRESHOLD=-5.0
 ML_SPAM_AUTOBAN_CHECK_LAST_MSG_COUNT=10
+ML_TRAIN_RANDOM_SORT_DATA=true
 ML_TRAIN_INTERVAL_DAYS=30
 ML_TRAIN_CRITICAL_MSG_COUNT=5
 ML_TRAINING_SET_FRACTION=0.2

src/VahterBanBot.Tests/ContainerTestBase.fs

@@ -96,6 +96,7 @@ type VahterTestContainers() =
             .WithEnvironment("CLEANUP_OLD_MESSAGES", "false")
             .WithEnvironment("ML_ENABLED", "true")
             .WithEnvironment("ML_SEED", "42")
+            .WithEnvironment("ML_TRAIN_RANDOM_SORT_DATA", "false")
             .WithEnvironment("ML_SPAM_DELETION_ENABLED", "true")
             .WithEnvironment("ML_SPAM_THRESHOLD", "1.0")
             .WithEnvironment("ML_STOP_WORDS_IN_CHATS", """{"-42":["2"]}""")

src/VahterBanBot/ML.fs

@@ -75,7 +75,11 @@ type MachineLearning(
                       createdAt = x.created_at
                       lessThanNMessagesF = if x.less_than_n_messages then 1.0f else 0.0f }
                 )
-
+                |> fun x ->
+                    if botConf.MlTrainRandomSortData then
+                        Array.sortInPlaceBy (fun _ -> Guid.NewGuid()) x
+                    x
+
             let dataView = mlContext.Data.LoadFromEnumerable data
             let trainTestSplit = mlContext.Data.TrainTestSplit(dataView, testFraction = botConf.MlTrainingSetFraction)
             let trainingData = trainTestSplit.TrainSet

src/VahterBanBot/Program.fs

@@ -61,6 +61,7 @@ let botConf =
       MlSpamAutobanEnabled = getEnvOr "ML_SPAM_AUTOBAN_ENABLED" "false" |> bool.Parse
       MlSpamAutobanCheckLastMsgCount = getEnvOr "ML_SPAM_AUTOBAN_CHECK_LAST_MSG_COUNT" "10" |> int
       MlSpamAutobanScoreThreshold = getEnvOr "ML_SPAM_AUTOBAN_SCORE_THRESHOLD" "-5.0" |> double
+      MlTrainRandomSortData = getEnvOr "ML_TRAIN_RANDOM_SORT_DATA" "true" |> bool.Parse
       MlTrainInterval = getEnvOr "ML_TRAIN_INTERVAL_DAYS" "30" |> int |> TimeSpan.FromDays
       MlTrainCriticalMsgCount = getEnvOr "ML_TRAIN_CRITICAL_MSG_COUNT" "5" |> int
       MlTrainingSetFraction = getEnvOr "ML_TRAINING_SET_FRACTION" "0.2" |> float

src/VahterBanBot/Types.fs

@@ -34,6 +34,7 @@ type BotConfiguration =
       MlSpamAutobanEnabled: bool
       MlSpamAutobanCheckLastMsgCount: int
       MlSpamAutobanScoreThreshold: double
+      MlTrainRandomSortData: bool
       MlTrainInterval: TimeSpan
       MlTrainCriticalMsgCount: int
       MlTrainingSetFraction: float