neighbor 0.1.1 → 0.1.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 01df817b036ee9b4d0c54ddadfb3be7fb31b36bb81504321b8dfd3d7a1ba83a4
-  data.tar.gz: f7a1757914f2e1226bc3b95fe5e0df9837a530315d6e627062df565e331b1708
+  metadata.gz: e64a3292445759187b7c08fd9cc54fe0da340ea3e3ab5de909620de6395b4984
+  data.tar.gz: 274d73ed464a0503973c5549022dca8820257fe470c9746b5599da9a153d46e3
 SHA512:
-  metadata.gz: 8381834f92092cb13d2c8898b588f0758e99e48f84e3b506aab15acf54fb0b3f72449734ca68db10de466eecde4a7b277cceca72192240ce21567125ad0af52f
-  data.tar.gz: 79ddd953bf0c134d73c92c61a8051524c024e2e1c95710e346a191a8a7c098f55562f81263752ea2e2ccaef9ced8e376dbf6431a4e745931d256ee2ca854d309
+  metadata.gz: 17df9f8a5848337fc570c6fa685f54f4aa9d49bea8e52e74fbec098e0bc9e450b655c7d96cb66c6e8b6aaae4824f2c831ea26a14f7f46b50ef64c955cfb9839b
+  data.tar.gz: a47165d174ec86ebfdcd128e62c50b85d097aac535b6b6d424ea31988428a57237cdade2829d8e22a15e566ac7597371c95fabafad42cdbc5d56f063abab55e4

data/CHANGELOG.md

@@ -1,3 +1,7 @@
+## 0.1.2 (2021-02-21)
+
+- Added `nearest_neighbors` scope
+
 ## 0.1.1 (2021-02-16)
 
 - Fixed `Could not dump table` error

data/README.md

@@ -50,12 +50,18 @@ item.update(neighbor_vector: [1.0, 1.2, 0.5])
 
 > With cosine distance (the default), vectors are normalized before being stored
 
-Get the nearest neighbors
+Get the nearest neighbors to a record
 
 ```ruby
 item.nearest_neighbors.first(5)
 ```
 
+Get the nearest neighbors to a vector
+
+```ruby
+Item.nearest_neighbors([1, 2, 3])
+```
+
 ## Distance
 
 Specify the distance metric
@@ -73,6 +79,8 @@ Supported values are:
 - `taxicab`
 - `chebyshev`
 
+For inner product, see [this example](examples/disco_user_recs.rb)
+
 Records returned from `nearest_neighbors` will have a `neighbor_distance` attribute
 
 ```ruby
@@ -86,7 +94,7 @@ By default, Postgres limits the `cube` data type to 100 dimensions. See the [Pos
 
 ## Example
 
-You can use Neighbor for online item recommendations with [Disco](https://github.com/ankane/disco). We’ll use MovieLens data for this example.
+You can use Neighbor for online item-based recommendations with [Disco](https://github.com/ankane/disco). We’ll use MovieLens data for this example.
 
 Generate a model
 
@@ -126,7 +134,7 @@ movie = Movie.find_by(name: "Star Wars (1977)")
 movie.nearest_neighbors.first(5).map(&:name)
 ```
 
-[Complete code](examples/disco.rb)
+[Complete code](examples/disco_item_recs.rb)
 
 ## History
 

data/lib/neighbor/model.rb

@@ -11,8 +11,10 @@ module Neighbor
       class_eval do
         attribute attribute_name, Neighbor::Vector.new(dimensions: dimensions, distance: distance)
 
-        define_method :nearest_neighbors do
-          return self.class.none if neighbor_vector.nil?
+        scope :nearest_neighbors, ->(vector) {
+          return none if vector.nil?
+
+          quoted_attribute = "#{connection.quote_table_name(table_name)}.#{connection.quote_column_name(attribute_name)}"
 
           operator =
             case distance
@@ -24,11 +26,10 @@ module Neighbor
               "<->"
             end
 
-          quoted_attribute = "#{self.class.connection.quote_table_name(self.class.table_name)}.#{self.class.connection.quote_column_name(attribute_name)}"
-
           # important! neighbor_vector should already be typecast
           # but use to_f as extra safeguard against SQL injection
-          order = "#{quoted_attribute} #{operator} cube(array[#{neighbor_vector.map(&:to_f).join(", ")}])"
+          vector = Neighbor::Vector.cast(vector, dimensions: dimensions, distance: distance)
+          order = "#{quoted_attribute} #{operator} cube(array[#{vector.map(&:to_f).join(", ")}])"
 
           # https://stats.stackexchange.com/questions/146221/is-cosine-similarity-identical-to-l2-normalized-euclidean-distance
           # with normalized vectors:
@@ -38,11 +39,15 @@ module Neighbor
           neighbor_distance = distance == "cosine" ? "POWER(#{order}, 2) / 2.0" : order
 
           # for select, use column_names instead of * to account for ignored columns
-          self.class
-            .select(*self.class.column_names, "#{neighbor_distance} AS neighbor_distance")
-            .where.not(self.class.primary_key => send(self.class.primary_key))
+          select(*column_names, "#{neighbor_distance} AS neighbor_distance")
             .where.not(attribute_name => nil)
             .order(Arel.sql(order))
+        }
+
+        define_method :nearest_neighbors do
+          self.class
+            .where.not(self.class.primary_key => send(self.class.primary_key))
+            .nearest_neighbors(send(attribute_name))
         end
       end
     end

data/lib/neighbor/vector.rb

@@ -6,13 +6,11 @@ module Neighbor
       @distance = distance
     end
 
-    def cast(value)
-      return if value.nil?
-
+    def self.cast(value, dimensions:, distance:)
       value = value.to_a.map(&:to_f)
-      raise Error, "Expected #{@dimensions} dimensions, not #{value.size}" unless value.size == @dimensions
+      raise Error, "Expected #{dimensions} dimensions, not #{value.size}" unless value.size == dimensions
 
-      if @distance == "cosine"
+      if distance == "cosine"
         norm = Math.sqrt(value.sum { |v| v * v })
         value.map { |v| v / norm }
       else
@@ -20,6 +18,10 @@ module Neighbor
       end
     end
 
+    def cast(value)
+      self.class.cast(value, dimensions: @dimensions, distance: @distance) unless value.nil?
+    end
+
     def serialize(value)
       "(#{cast(value).join(", ")})" unless value.nil?
     end

data/lib/neighbor/version.rb

@@ -1,3 +1,3 @@
 module Neighbor
-  VERSION = "0.1.1"
+  VERSION = "0.1.2"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: neighbor
 version: !ruby/object:Gem::Version
-  version: 0.1.1
+  version: 0.1.2
 platform: ruby
 authors:
 - Andrew Kane
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2021-02-17 00:00:00.000000000 Z
+date: 2021-02-22 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: activerecord