henri24 commited on
Commit
95fb11f
·
verified ·
1 Parent(s): 477d767

Upload 2 files

Browse files

reasoning-datasets-competition

Files changed (2) hide show
  1. README.md +48 -3
  2. llm_combinatorial_test.json +0 -0
README.md CHANGED
@@ -1,3 +1,48 @@
1
- ---
2
- license: cc-by-sa-4.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Combined Combinatorial Optimization Dataset
2
+
3
+ ## Overview
4
+ This dataset is created for evaluating the effectiveness of Large Language Models (LLMs) on various combinatorial optimization problems. Each instance in the dataset represents a problem instance and includes all the necessary attributes to learn and evaluate problem solutions generated by Google's OR Tools.
5
+
6
+ ## Problem Types and Attributes
7
+ Each instance in the dataset is expected to have the following attributes:
8
+ - **input**: The actual problem data (e.g., graph data for VRP, jobs and machines for JSSP, etc.).
9
+ - **instruction**: A general description of the problem. This attribute contains problem-specific instructions or details (for example, the number of machines and jobs in a jssp problem). The description varies for each problem type.
10
+ - **output_list_of_list**: Expected feasible solution provided as a list of lists.
11
+ - **output_starjob**: Expected solution in a human-readable format. This includes detailed explanations such as how the makespan is calculated in JSSP or how the knapsack capacity is updated step by step.
12
+
13
+ Additionally, instances are tagged with a specific **problem_type** (e.g., 'vrp', 'jssp', etc.) which may have additional attributes, depending on the specific requirements of the problem type. For example, instances where `problem_type` equals `'vrp'` might include extra attributes relevant only to Vehicle Routing Problems.
14
+
15
+ ## Dataset Statistics
16
+ - **Total Instances**: 150
17
+
18
+ ### Instance Counts by Problem Type
19
+ | Problem Type | Count |
20
+ |--------------|-------|
21
+ | vrp | 30 |
22
+ | knapsack | 30 |
23
+ | tsp | 30 |
24
+ | binpack | 30 |
25
+ | jssp | 30 |
26
+
27
+ ## Attribute Details
28
+ Below are the attributes extracted from the first instance along with their inferred data types. Note that the dataset may contain additional problem-specific attributes that vary by problem type:
29
+ ```
30
+ {'capacity': 'int',
31
+ 'city_size': 'int',
32
+ 'demands': 'list',
33
+ 'input': 'str',
34
+ 'instance_id': 'int',
35
+ 'instruction': 'str',
36
+ 'max_interval': 'int',
37
+ 'num_cities': 'int',
38
+ 'num_vehicles': 'int',
39
+ 'output_list_of_list': 'str',
40
+ 'output_starjob': 'str',
41
+ 'paired_distances': 'str',
42
+ 'problem_type': 'str',
43
+ 'time': 'float',
44
+ 'vehicle_count': 'int'}
45
+ ```
46
+
47
+ ## Conclusion
48
+ This dataset provides both structured problem definitions and two types of solution representations to facilitate a comprehensive evaluation of LLM-based approaches across different combinatorial optimization problems.
llm_combinatorial_test.json ADDED
The diff for this file is too large to render. See raw diff