Merge branch 'feature/iteration63' into develop

rust_project/arc-competition/payload/.loda-rust/analytics-arc/analytics_log.txt

@@ -32,4 +32,4 @@ optimize: 0, dontoptimize: 1, optimize/total: 0.0%
 AnalyzeProgramModified
 timestamps: 88
 
-subcommand_analytics finished, elapsed: 20 ms
+subcommand_analytics finished, elapsed: 21 ms

rust_project/arc-competition/payload/.loda-rust/analytics-arc/last_analytics_timestamp.txt

@@ -1 +1 @@
-2023-11-23T01:19:08Z
+2023-11-25T18:24:46Z

rust_project/loda-rust-cli/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "loda-rust-cli"
-version = "2023.11.23"
+version = "2023.11.25"
 authors = ["Simon Strandgaard <[email protected]>"]
 description = "Command line interface for LODA Rust"
 repository = "https://github.com/loda-lang/loda-rust"

rust_project/loda-rust-cli/src/arc/arc_work_model_impl_pair.rs

@@ -820,19 +820,19 @@ impl arc_work_model::Pair {
         let output_least_popular_color: Option<u8> = self.output.image_meta.histogram_all.least_popular_color_disallow_ambiguous();
 
         if input_most_popular_color.is_some() && input_most_popular_color == output_most_popular_color {
-            let label = ActionLabel::InputMostPopularColorIsOutputMostPopularColor;
+            let label = ActionLabel::PairInputMostPopularColorIsOutputMostPopularColor;
             self.action_label_set.insert(label);
         }
         if input_most_popular_color.is_some() && input_most_popular_color == output_least_popular_color {
-            let label = ActionLabel::InputMostPopularColorIsOutputLeastPopularColor;
+            let label = ActionLabel::PairInputMostPopularColorIsOutputLeastPopularColor;
             self.action_label_set.insert(label);
         }
         if input_least_popular_color.is_some() && input_least_popular_color == output_most_popular_color {
-            let label = ActionLabel::InputLeastPopularColorIsOutputMostPopularColor;
+            let label = ActionLabel::PairInputLeastPopularColorIsOutputMostPopularColor;
             self.action_label_set.insert(label);
         }
         if input_least_popular_color.is_some() && input_least_popular_color == output_least_popular_color {
-            let label = ActionLabel::InputLeastPopularColorIsOutputLeastPopularColor;
+            let label = ActionLabel::PairInputLeastPopularColorIsOutputLeastPopularColor;
             self.action_label_set.insert(label);
         }
 

rust_project/loda-rust-cli/src/arc/arcathon_solution_coordinator.rs

@@ -27,18 +27,24 @@ pub enum PredictionType {
 
     SolveSplit,
     SolveGenetic,
+
+    /// Number of unique colors the output images across all train pairs, is s a single color.
+    SolveOneColor,
 }
 
 impl PredictionType {
     #[allow(dead_code)]
     fn sort_weight(&self) -> u32 {
         match self {
             // Split tries out lots of things deterministic, so it's high priority.
-            Self::SolveSplit => 0, 
+            Self::SolveSplit => 0,
+
+            // The SolveOneColor is a very simple solver, so it's high priority.
+            Self::SolveOneColor => 1, 
 
             // The LODA programs that have been manually been coded are somewhat good and deals with many edge cases.
             // The mutated LODA programs, may not deal with edge cases, but they are still good, since all train+test pairs gets evaluated.
-            Self::SolveGenetic => 1, 
+            Self::SolveGenetic => 2, 
 
             // Logistic regression is rarely correct, so it's low priority.
             // It's best at tasks where `input_size == output_size`.

rust_project/loda-rust-cli/src/arc/histogram.rs

@@ -327,7 +327,7 @@ impl Histogram {
         self.least_popular().count_allow_ambiguous()
     }
 
-    /// The pairs ordered by their color value.
+    /// The pairs ordered by their color value. Ignoring their count.
     /// 
     /// The lowest color value comes first.
     /// 
@@ -342,6 +342,16 @@ impl Histogram {
         pairs
     }
 
+    /// The colors ordered by their color value. Ignoring their count.
+    /// 
+    /// The lowest color value comes first.
+    /// 
+    /// The highest color value comes last.
+    #[allow(dead_code)]
+    pub fn color_vec(&self) -> Vec<u8> {
+        self.pairs_ordered_by_color().iter().map(|pair| pair.1).collect()
+    }
+
     /// The least frequent occurring comes first.
     /// 
     /// The medium frequent occurring comes middle.
@@ -850,7 +860,24 @@ mod tests {
     }
 
     #[test]
-    fn test_50001_pairs_descending() {
+    fn test_50001_color_vec() {
+        // Arrange
+        let mut h = Histogram::new();
+        let values: [u8; 8] = [3, 42, 42, 3, 2, 3, 4, 5];
+        for value in values {
+            h.increment(value);
+        }
+
+        // Act
+        let actual: Vec<u8> = h.color_vec();
+
+        // Assert
+        let expected: Vec<u8> = vec![2, 3, 4, 5, 42];
+        assert_eq!(actual, expected);
+    }
+
+    #[test]
+    fn test_50002_pairs_descending() {
         // Arrange
         let mut h = Histogram::new();
         let values: [u8; 12] = [3, 1, 1, 42, 7, 42, 7, 3, 2, 3, 4, 5];
@@ -867,7 +894,7 @@ mod tests {
     }
 
     #[test]
-    fn test_50002_pairs_ascending() {
+    fn test_50003_pairs_ascending() {
         // Arrange
         let mut h = Histogram::new();
         let values: [u8; 12] = [3, 1, 1, 42, 7, 42, 7, 3, 2, 3, 4, 5];

rust_project/loda-rust-cli/src/arc/label.rs

@@ -389,16 +389,16 @@ pub enum ActionLabel {
     OutputImageColorsComesFromInputImage,
 
     /// What input colors explains the output colors. The most popular stays the most popular.
-    InputMostPopularColorIsOutputMostPopularColor,
+    PairInputMostPopularColorIsOutputMostPopularColor,
 
     /// What input colors explains the output colors. The least popular stays the least popular.
-    InputLeastPopularColorIsOutputLeastPopularColor,
+    PairInputLeastPopularColorIsOutputLeastPopularColor,
 
     /// What input colors explains the output colors. The most popular becomes the least popular.
-    InputMostPopularColorIsOutputLeastPopularColor,
+    PairInputMostPopularColorIsOutputLeastPopularColor,
 
     /// What input colors explains the output colors. The least popular becomes the most popular.
-    InputLeastPopularColorIsOutputMostPopularColor,
+    PairInputLeastPopularColorIsOutputMostPopularColor,
 
     /// The output size is the same as the input size.
     /// Each pixel have the same number of identical pixels as in the input.
@@ -468,6 +468,7 @@ pub enum ActionLabel {
     // ObjectsOnlyMoveInDirectionY,
     // CropWhenFinishedDrawingInsideSingleColorObjectWithColor { color: u8 },
     // OutputImageCropOutSingleColorObject { color: u8 },
+    // OutputSizeIsHalfOfInputSizeDependingOnInputOrientation,
     // OutputSizeIsTheSameAsBoundingBoxOfSingleColorObject { color: u8 },
     // OutputSizeIsTheSameAsHoleInSingleColorObject { color: u8 },
     // OutputSizeIsTheSameAsInputSmallestSingleColorObjectRectangle,

rust_project/loda-rust-cli/src/arc/mod.rs

@@ -148,6 +148,7 @@ mod single_color_object;
 mod single_color_object_satisfies_label;
 mod single_color_object_to_label;
 mod solve_logisticregression;
+mod solve_one_color;
 mod solve_split;
 mod split;
 mod split_to_label;
@@ -291,6 +292,7 @@ pub use single_color_object::{SingleColorObject, SingleColorObjectRectangle, Sin
 pub use single_color_object_satisfies_label::SingleColorObjectSatisfiesLabel;
 pub use single_color_object_to_label::SingleColorObjectToLabel;
 pub use solve_logisticregression::SolveLogisticRegression;
+pub use solve_one_color::SolveOneColor;
 pub use solve_split::{Operation, SolveSplit, SolveSplitFoundSolution};
 pub use split::{EvenSplit, Split, SplitCandidate, SplitCandidateContainer};
 pub use split_to_label::SplitToLabel;

rust_project/loda-rust-cli/src/arc/solve_logisticregression.rs

@@ -1,6 +1,10 @@
 //! Performs logistic regression of each input pixel with the corresponding classification for the output pixel.
 //! 
 //! These older commits solves some of the tasks from the hidden ARC dataset, using logistic regression:
+//! commit 2023-Nov-25: solves 2 of the hidden ARC tasks. This uses variant=0 and variant=1 and variant=2. Only variant=0 does something useful.
+//! Since it uses multiple variants and doesn't solve more tasks than the previous commit, then it is not an improvement.
+//! https://github.com/loda-lang/loda-rust/commit/d08069b56c92c54140c27d13f8e6ad0897824d5a
+//!
 //! commit 2023-Nov-23: solves 2 of the hidden ARC tasks. This uses variant=0 and variant=1 and variant=2. Only variant=0 does something useful.
 //! Since it uses multiple variants and doesn't solve more tasks than the previous commit, then it is not an improvement.
 //! https://github.com/loda-lang/loda-rust/commit/2e6577cd66af0c00fdd65533d4770a40ed2ccf3c
@@ -70,6 +74,7 @@
 //! However I don't see any improvement of the prediction accuracy.
 //! 
 //! Future experiments:
+//! * Use confusion matrix for determining what parameters have impact on the precision. And enable/disable features based on the confusion matrix.
 //! * Transform the `train` pairs: rotate90, rotate180, rotate270, flipx, flipy.
 //! * Transform the `test` pairs: rotate90, rotate180, rotate270, flipx, flipy.
 //! * Provide `weight` to logistic regression, depending on how important each parameter is.
@@ -388,6 +393,7 @@ impl ProcessTaskContext {
 struct ProcessedTaskWithOneTestPair {
     test_index: u8,
     cropped_image: Image,
+    #[allow(dead_code)]
     inspect_internal_image_vec: Vec<Image>,
 }
 
@@ -397,12 +403,10 @@ struct ProcessedTask {
 }
 
 pub struct SolveLogisticRegression {
-    #[allow(dead_code)]
     tasks: Vec<Task>,
 }
 
 impl SolveLogisticRegression {
-    #[allow(dead_code)]
     pub fn new(tasks: Vec<Task>) -> Self {
         // println!("loaded {} tasks", tasks.len());
         Self {
@@ -415,6 +419,7 @@ impl SolveLogisticRegression {
     /// This can be run with the public ARC dataset contains expected output for the test pairs.
     /// 
     /// This cannot be run with the hidden ARC dataset, which doesn't contain expected output for the test pairs.
+    #[allow(dead_code)]
     pub fn run_and_verify(&self) -> anyhow::Result<()> {
         let run_and_verify_htmllog = true;
         let run_and_verify_ignore_already_solved = false;
@@ -559,7 +564,7 @@ impl SolveLogisticRegression {
     /// This code is intended to run with the hidden ARC dataset, which doesn't contain expected output for the test pairs.
     pub fn run_predictions(&self) -> anyhow::Result<TaskNameToPredictionVec> {
         let number_of_tasks: u64 = self.tasks.len() as u64;
-        println!("{} - run start - will process {} tasks with logistic regression", human_readable_utc_timestamp(), number_of_tasks);
+        println!("{} - run start - will process {} tasks with SolveLogisticRegression", human_readable_utc_timestamp(), number_of_tasks);
         let count_solved = AtomicUsize::new(0);
         let pb = ProgressBar::new(number_of_tasks as u64);
         pb.set_style(ProgressStyle::default_bar()
@@ -676,10 +681,10 @@ impl SolveLogisticRegression {
         }
 
         let mut prediction_vec = Vec::<arcathon_solution_coordinator::Prediction>::new();
-        for (test_index, ptwotp) in ptwotp_vec.iter().enumerate() {
+        for ptwotp in &ptwotp_vec {
             let grid: arc_json_model::Grid = arc_json_model::Grid::from_image(&ptwotp.cropped_image);
             let prediction = arcathon_solution_coordinator::Prediction {
-                output_id: test_index.min(255) as u8,
+                output_id: ptwotp.test_index.min(255) as u8,
                 output: grid,
                 prediction_type,
             };