Make depth first search for Knapsack non-recursive and more efficient. Solves all problems in reasonable time now.

knapsack/dp_3

@@ -1,4 +0,0 @@
-3 10
-45 5
-48 8
-35 3

knapsack/knapsack.py

@@ -3,6 +3,7 @@ from collections import namedtuple
 Result = namedtuple("Result", ['objective', 'is_optimal', 'xs'])
 Item = namedtuple("Item", ['index', 'value', 'weight'])
 Knapsack = namedtuple("Knapsack", ['count', 'capacity', 'items'])
+Node = namedtuple("Node", ['index', 'value', 'room', 'estimate', 'path'])
 
 
 def solve_knapsack_greedy(knapsack):
@@ -62,57 +63,69 @@ def solve_knapsack_dynamic(knapsack):
 
 
 def solve_knapsack_depth_first_search(knapsack):
-    knapsack.items.sort(key=lambda item: item.value / item.weight,
+    knapsack.items.sort(key=lambda item: item.value / float(item.weight),
                         reverse=True)
-    num_items = len(knapsack.items)
-    result = {"objective": 0,
-              "path": []}
 
-    def get_max_value(from_index, capacity):
-        value = 0
-        for i in range(from_index, num_items):
-            item = knapsack.items[i]
-            if item.weight <= capacity:
-                value += item.value
-                capacity -= item.weight
+    def get_estimate(current_value, current_index, remaining_capacity):
+        estimated_value = current_value
+        for item in knapsack.items[current_index:]:
+            if item.weight <= remaining_capacity:
+                estimated_value += item.value
+                remaining_capacity -= item.weight
             else:
-                value += int((capacity / item.weight) * item.value)
-                break
-        return value
+                v = int((remaining_capacity / float(item.weight)) * item.value)
+                estimated_value += v
+                return estimated_value
+        return estimated_value
 
-    def search(index, capacity, value, path, result):
-        if value > result["objective"]:
-            result["objective"] = value
-            result["path"] = list(path)
+    estimate = get_estimate(0, 0, knapsack.capacity)
+    nodes = [Node(0, 0, knapsack.capacity, estimate, [])]
+    num_items = len(knapsack.items)
+    best_value = 0
+    best_path = []
 
-        if capacity <= 0:
-            return
+    while nodes:
+        current_node = nodes.pop()
+        current_index = current_node.index
 
-        if index == num_items:
-            return
+        if current_node.room < 0:
+            continue
 
-        # print("--- search")
-        # print("Index: {} capacity: {}, value: {}".format(index, capacity, value))
-        # print("Path {}".format(path))
+        if current_node.estimate < best_value:
+            continue
 
-        # Take current item.
-        max_value = get_max_value(index, capacity)
-        print(max_value)
-        item = knapsack.items[index]
-        # print("max_value: {}".format(max_value))
-        max_value_not = get_max_value(index + 1, capacity)
-        # print("max_value_not: {}".format(max_value_not))
-        if item.weight <= capacity and value + max_value > result["objective"]:
-            path.append(1)
-            search(index + 1, capacity - item.weight,
-                   value + item.value, path, result)
-            path.pop()
+        if current_node.value > best_value:
+            best_value = current_node.value
+            best_path = list(current_node.path)
 
-        # Do not take current item.
-        if value + max_value_not > result["objective"]:
-            path.append(0)
-            search(index + 1, capacity, value, path, result)
-            path.pop()
+        if current_index == num_items:
+            continue
+
+        current_item = knapsack.items[current_index]
+
+        # Right child - do not take current_item
+        new_index = current_index + 1
+        new_value = current_node.value
+        new_room = current_node.room
+        new_estimate = get_estimate(new_value, new_index, new_room)
+        if new_estimate > best_value:
+            new_path = current_node.path + [0]
+            r = Node(new_index, new_value, new_room, new_estimate, new_path)
+            nodes.append(r)
+
+        # Left child - take current_item
+        new_index = current_index + 1
+        new_room = current_node.room - current_item.weight
+        if new_room >= 0:
+            new_value = current_node.value + current_item.value
+            new_estimate = get_estimate(new_value, new_index, new_room)
+            new_path = current_node.path + [1]
+            n = Node(new_index, new_value, new_room, new_estimate, new_path)
+            nodes.append(n)
+
+        # If we sort the items by estimate here it becomes
+        # best first search.
+        # nodes.sort(key=lambda node: node.estimate)
 
     def correct_path(path):
         path = path + [0] * (num_items - len(path))
@@ -120,9 +133,7 @@ def solve_knapsack_depth_first_search(knapsack):
         path = [v[0] for v in sorted(values, key=lambda value: value[1].index)]
         return path
 
-    search(0, knapsack.capacity, 0, [], result)
-
-    return Result(result["objective"], 1, correct_path(result["path"]))
+    return Result(best_value, 0, correct_path(best_path))
 
 
 def input_data_to_knapsack(input_data):

knapsack/solver.py

@@ -1,15 +1,13 @@
-#!/usr/bin/python3
+#!/usr/bin/pypy
 # -*- coding: utf-8 -*-
 
 import knapsack
 
 
 def solve_it(input_data):
-    # Modify this code to run your optimization algorithm
     k = knapsack.input_data_to_knapsack(input_data)
     # r = knapsack.solve_knapsack_dynamic(k)
     r = knapsack.solve_knapsack_depth_first_search(k)
-    # r = knapsack.solve_knapsack_greedy(k)
     return knapsack.result_to_output_data(r)