Datasets:
Upload dataset.md
Browse files- table_data/dataset.md +84 -0
table_data/dataset.md
ADDED
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Introduction to Datasets
|
2 |
+
|
3 |
+
Presolved Data is stored in `.\instance`. The folder structure after the datasets are set up looks as follows
|
4 |
+
|
5 |
+
```bash
|
6 |
+
instances/
|
7 |
+
MIPLIB/ -> 1065 instances
|
8 |
+
set_cover/ -> 3994 instances
|
9 |
+
independent_set/ -> 1604 instances
|
10 |
+
nn_verification/ -> 3104 instances
|
11 |
+
load_balancing/ -> 2286 instances
|
12 |
+
```
|
13 |
+
|
14 |
+
### Dataset Description
|
15 |
+
|
16 |
+
#### MIPLIB
|
17 |
+
|
18 |
+
Heterogeneous dataset from [MIPLIB 2017](https://miplib.zib.de/), a well-established benchmark for evaluating MILP solvers. The dataset includes a diverse set of particularly challenging mixed-integer programming (MIP) instances, each known for its computational difficulty.
|
19 |
+
|
20 |
+
#### Set Covering
|
21 |
+
|
22 |
+
This dataset consists of instances of the classic Set Covering Problem, which can be found [here](https://github.com/ds4dm/learn2branch/tree/master). Each instance requires finding the minimum number of sets that cover all elements in a universe. The problem is formulated as a MIP problem.
|
23 |
+
|
24 |
+
#### Maximum Independent Set
|
25 |
+
|
26 |
+
This dataset addresses the Maximum Independent Set Problem, which can be found [here](https://github.com/ds4dm/learn2branch/tree/master). Each instance is modeled as a MIP, with the objective of maximizing the size of the independent set.
|
27 |
+
|
28 |
+
#### NN Verification
|
29 |
+
|
30 |
+
This “Neural Network Verification” dataset is to verify whether a neural network is robust to input perturbations can be posed as a MIP. The MIP formulation is described in the paper [On the Effectiveness of Interval Bound Propagation for Training Verifiably Robust Models (Gowal et al., 2018)](https://arxiv.org/abs/1810.12715). Each input on which to verify the network gives rise to a different MIP.
|
31 |
+
|
32 |
+
#### Load Balancing
|
33 |
+
|
34 |
+
This dataset is from [NeurIPS 2021 Competition](https://github.com/ds4dm/ml4co-competition). This problem deals with apportioning workloads. The apportionment is required to be robust to any worker’s failure. Each instance problem is modeled as a MILP, using a bin-packing with an apportionment formulation.
|
35 |
+
|
36 |
+
### Dataset Spliting
|
37 |
+
|
38 |
+
Each dataset was split into a training set $D_{\text{train}}$ and a testing set $D_{\text{test}}$, following an approximate 80-20 split. Moreover, we split the dataset by time and "optimality", which means according to the proportion of optimality for each parameter is similar in training and testing sets. This ensures a balanced representation of both temporal variations and the highest levels of parameter efficiency in our data partitions.
|
39 |
+
|
40 |
+
To split the datasets and create different folds for cross validation, run
|
41 |
+
|
42 |
+
```bash
|
43 |
+
python extract_feature/split_fold.py \
|
44 |
+
--dataset_name "your_dataset_name" \
|
45 |
+
--time_path "/your/path/to/soving times" \
|
46 |
+
--feat_path "/your/path/to/features" \
|
47 |
+
```
|
48 |
+
|
49 |
+
## Handcraft Feature Extraction
|
50 |
+
|
51 |
+
**Folder:** ```extract_feature```
|
52 |
+
|
53 |
+
**Problem format:** ```your_file.mps.gz``` or ```your_file.lp```
|
54 |
+
|
55 |
+
**Log format:** ```your_file.log```
|
56 |
+
|
57 |
+
1. Static feature extraction for MIP problems: run
|
58 |
+
|
59 |
+
```bash
|
60 |
+
python extract_feature/extract_problem.py \
|
61 |
+
--problem_folder "/your/path/to/instances" \
|
62 |
+
--dataset_name "your_dataset_name" \
|
63 |
+
```
|
64 |
+
|
65 |
+
2. Extract other features from COPT solution log: run
|
66 |
+
|
67 |
+
```bash
|
68 |
+
python extract_feature/extract_log_feature.py \
|
69 |
+
--log_folder "/your/path/to/solving logs" \
|
70 |
+
--dataset_name "your_dataset_name" \
|
71 |
+
```
|
72 |
+
|
73 |
+
3. Feature combination and preprocessing: run
|
74 |
+
|
75 |
+
```bash
|
76 |
+
python extract_feature/combine.py \
|
77 |
+
--dataset_name "your_dataset_name" \
|
78 |
+
```
|
79 |
+
4. Label extraction from COPT solution log: run
|
80 |
+
|
81 |
+
```bash
|
82 |
+
python extract_feature/extract_time.py \
|
83 |
+
--log_folder "/your/path/to/solving logs" \
|
84 |
+
--dataset_name "your_dataset_name" \
|