-
Notifications
You must be signed in to change notification settings - Fork 0
/
README
34 lines (25 loc) · 1.59 KB
/
README
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
- You can run the code directly using the .jar file or the .class executables, or you can access and compile the source code yourself from the "source" folder.
- The algorithm needs the Loandra solver to be installed.
- There needs to be 6 folders called "clauses", "consts", "labels", "logs", "pairs", and "solutions" next to the JAR file.
Parameters:
-d [value]: tree depth (specifying no depth results in flat clustering)
-print: to print the results
-norm: to normalize the features in the dataset
-obj [mode] [eps value]: the clustering objective, "wcm" mode for MD only and "wcm+bcm_p" mode for Pareto objective (optional: add "-Smart" after [eps value] to use smart pairs)
-k [value]: number of clusters
-seed [value]: randomization seed
-mcl_c n[value]: number of must and cannot-links to be generated based on the size of the dataset (kappa value)
-sto [value]: timeout limit (in minutes)
-path [string]: the path to the loandra solver
-post [string]: a postfix to distinguish the name of the files stored in the six folders
Dataset:
- The name of the dataset should come at the end of the command
- The dataset file should be named "instance_[dataset name]"
- The dataset format is:
"flags: l" if labels appear at the end of each line or "flags: lb" if they appear at the beginning
[n] (size of the dataset)
[f] (number of features)
[c] (number of labels)
each line: first the features and then the label, separated by space
Example:
java com.company.Main -norm -print -seed 1234 -post _test -obj wcm+bcm_p 0.1 -smart -k 2 -d 2 -mcl_c n0.5 -sto 30 -path ~/SAT_Project/loandra-master/loandra_static iris