BIRCH (Balanced Iterative Reducing and Clustering Using Hierarchies) is an unsupervised data mining algorithm which is used to perform hierarchical clustering over particularly large data-sets.

• Compare with jbirch's results;
• Fix memory leaks;
• Make array.c more memory efficient by allowing setting the size of the elements;
• Analyze the ideal size for every Array initialization;
• Implement a more efficient way of reading csv files (intance.c);
• Make sure that jbirch is correct, since it is being used as reference for this implementation;
• Implement jbirch's automatic rebuilding feature.
• Output the entire tree, not only the leafs.

• gcc

To compile, execute the following command line from the root of the project:

gcc ./src/main.c ./src/*/*.c -o ./bin/main -lm

After compiling it, you can run the algorithm by executing the ./bin/main binary with the following arguments, which must be informed in this order and separated by whitespace:

1. Branching Factor: An integer value;
2. Threshold: A float value;
3. Apply Merging Refinement: 1 for yes, 0 for no;
4. Dataset File Path: String;
5. Dataset Delimiters: String;
6. Ignore Last Column of the Dataset: 1 for yes, 0 for no;
7. Output File Name: String.

Example of valid command line to run the program (execute it from the root of the project):

./bin/main 100 0.8 1 IRIS.csv , 1 output.csv

Once the execution finishes, the output file containing the result of the clustering will be created with the name specified in the "Output File Name" argument. The output file contains the cluster identifier of each instance, in the same order that the instances appears in the input file.

The program takes as input a csv. The column's delimiter is passed as an argument as described in the How to Run section. Labels can only appear in the last column of the dataset. Make sure to set the "Ignore Last Column of the Dataset" accordingly.

In order to test our implementation of BIRCH we compared it with Roberto Perdisci's implementation, called jbirch. The comparison was made using the covtype.csv dataset, which is able to cover all the relevant code from both implementations. Important to notice that in the comparison we disabled jbirch's automatic rebuilding feature as it isn't supported by our implementation yet.

To reproduce the tests, follow these steps:

1. Run jbirch on the covtype.csv dataset by following the instructions in the README.md file of this branch of the project;
2. Run our implementation, the C implementation, over the same dataset and with the same configurations by executing this command line from the root of the project (edit the dataset file path as needed): ./bin/main 5 0.5 1 covtype.csv , 1 cbirch_output.csv;
3. Compare the output file of both implementations. For the test to be succesfull both files must have the same content.

Valgrind must be installed in your machine.

To compile birch with ggdb3 flag execute the following command line from the root of the project:

gcc -ggdb3 ./src/main.c ./src/*/*.c -o ./bin/main -lm

To run the valgrind over birch with the covtype.csv dataset execute the following command line from the root of the project (edit the dataset file path as needed):

valgrind --leak-check=full --show-leak-kinds=all --track-origins=yes --verbose ./bin/main 100 0.8 1 covtype.csv , 1

The covtype dataset triggers all the functions of the implementation, allowing Valgrind to test every piece of code.