Add explicit for-item mapping to nearest-neighbor set

Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "recoreco"
 description = "Fast item-to-item recommendations on the command line."
-version = "0.1.9"
+version = "0.1.10"
 authors = ["Sebastian Schelter <ssc [at] apache [dot] org>"]
 license = "GPL-3.0-or-later"
 homepage = "https://github.com/sscdotopen/recoreco"
@@ -22,4 +22,4 @@ serde = "1.0"
 serde_derive = "1.0"
 serde_json = "1.0"
 getopts = "0.2"
-rayon = "1.0"
+rayon = "1.0"

src/bin/indicators.rs

@@ -94,7 +94,7 @@ fn compute_indicators(
     interactions_path: &str,
     n: usize,
     indicators_path: Option<String>,
-) -> Result<(), Box<Error>> {
+) -> Result<(), Box<dyn Error>> {
 
     // We use constants here for the moment, these should result in a good runtime/quality ratio.
     const F_MAX: u32 = 500;

src/io.rs

@@ -97,7 +97,7 @@ struct Indicators<'a> {
 /// `{ "for_item": "michael jackson", "indicated_items": ["justin timberlake", "queen"] }`
 ///
 pub fn write_indicators(
-    indicators: &[FnvHashSet<u32>],
+    indicators: &[(u32, FnvHashSet<u32>)],
     renaming: &Renaming,
     indicators_path: Option<String>,
 ) -> io::Result<()> {
@@ -107,13 +107,13 @@ pub fn write_indicators(
         _ => boxed_writer(stdout()),
     };
 
-    for (item_index, indicated_item_indices) in indicators.into_iter().enumerate() {
+    for (_index, indicator_tuple) in indicators.into_iter().enumerate() {
 
-        let for_item = renaming.item_name(item_index as u32);
+        let for_item = renaming.item_name(indicator_tuple.0 as u32);
 
-        let indicated_items: FnvHashSet<&str> = indicated_item_indices
+        let indicated_items: FnvHashSet<&str> = (&indicator_tuple.1)
             .into_iter()
-            .map(|item_index| renaming.item_name(*item_index as u32))
+            .map(|item_index| renaming.item_name(*item_index))
             .collect();
 
         let indicators_as_json = json!(
@@ -130,9 +130,9 @@ pub fn write_indicators(
 
 fn boxed_writer<T>(
     destination: T
-) -> Box<Write>
+) -> Box<dyn Write>
 where
     T: Write + 'static
 {
     Box::new(BufWriter::new(destination))
-}
+}

src/lib.rs

@@ -54,7 +54,7 @@ pub mod stats;
 mod usage_tests;
 
 use llr::ScoredItem;
-use types::{SparseVector, SparseMatrix, SparseBinaryMatrix};
+use types::{SparseVector, SparseMatrix, IndicatorSet};
 use stats::DataDictionary;
 
 /// Compute item indicators from a stream of interactions.
@@ -135,7 +135,7 @@ pub fn indicators<T>(
     num_indicators_per_item: usize,
     f_max: u32,
     k_max: u32
-) -> SparseBinaryMatrix
+) -> IndicatorSet
 where
     T: Iterator<Item = (String, String)>
 {
@@ -181,7 +181,7 @@ where
         if item_interaction_counts[item_idx] < f_max {
 
             // Retrieve current history sample for interacting user
-            let mut user_history = &mut samples_of_a[user_idx];
+            let user_history = &mut samples_of_a[user_idx];
             let num_items_in_user_history = user_history.len();
 
             // Check whether we have seen enough interactions for this user yet
@@ -259,9 +259,10 @@ where
                 num_cooccurrences_observed,
                 num_indicators_per_item,
                 &precomputed_logarithms,
+                //&renaming
             )
         })
-        .collect();
+        .collect::<Vec<(u32, FnvHashSet<u32>)>>();
 
     let duration = to_millis(start.elapsed());
     println!(
@@ -285,20 +286,21 @@ fn rescore(
     num_cooccurrences_observed: u64,
     n: usize,
     logarithms_table: &[f64],
-) -> FnvHashSet<u32> {
+) -> (u32, FnvHashSet<u32>) {
 
     // We can skip the scoring if we have seen less than n items
     if cooccurrence_counts.len() <= n {
-        cooccurrence_counts
+        (item, cooccurrence_counts
             .keys()
             .cloned()
-            .collect::<FnvHashSet<_>>()
+            .collect::<FnvHashSet<_>>())
     } else {
         // We'll use a heap to keep track of the current top-n scored items
         let mut top_indicators: BinaryHeap<ScoredItem> = BinaryHeap::with_capacity(n);
 
         for (other_item, num_cooccurrences) in cooccurrence_counts.iter() {
             if *other_item != item {
+
                 // Compute counts of contingency table
                 let k11 = u64::from(*num_cooccurrences);
                 let k12 = u64::from(row_sums_of_c[item as usize]) - k11;
@@ -327,6 +329,6 @@ fn rescore(
             .map(|scored_item| scored_item.item)
             .collect();
 
-        indicators_for_item
+        (item, indicators_for_item)
     }
-}
+}

src/llr.rs

@@ -184,4 +184,4 @@ mod tests {
         assert_eq!(top_k[2].item, 2);
         assert!(within_epsilon(top_k[2].score, 1.5));
     }
-}
+}

src/stats.rs

@@ -258,4 +258,4 @@ mod tests {
         assert_eq!(renaming.item_name(1), "item_b");
         assert_eq!(renaming.item_name(2), "item_c");
     }
-}
+}

src/types.rs

@@ -47,3 +47,5 @@ pub fn new_dense_vector(dimensions: usize) -> DenseVector {
 pub fn new_sparse_matrix(num_rows: usize) -> SparseMatrix {
     vec![FnvHashMap::with_capacity_and_hasher(0, Default::default()); num_rows]
 }
+
+pub type IndicatorSet = Vec<(u32, FnvHashSet<u32>)>;