# ocsmm

**Repository Path**: murphysong/ocsmm

## Basic Information

- **Project Name**: ocsmm
- **Description**: ocmssddfffdf
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2023-06-16
- **Last Updated**: 2023-06-16

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Fast, Efficient One Class Support Measure Machines

Repository intended to explore a fast and memory efficient implementation of One Class Support Measure Machines (OCSMM).
Ideas from these papers are explored:
1. [One-Class Support Measure Machines for Group Anomaly Detection, 2013, Proceedings of the Twenty-Ninth](https://arxiv.org/abs/1408.2064)
2. [Large Scale Online Kernel Learning. Journal of Machine Learning Research 2016](http://www.jmlr.org/papers/v17/14-148.html)

<p float="center">
  <img src="results/plots/mixture_scatter.png" width="720" />
</p>

<p float="center">
  <img src="results/plots/roc_curves.png" width="240" />
  <img src="results/plots/sample_size_times.png" width="240" /> 
  <img src="results/plots/rolling_window_times.png" width="240" /> 
</p>

# Conda environment
```console
local_user@local_host$ bash setup.sh
```

# Run OCSMM experiments
```console
local_user@local_host$ python src/utils_experiment.py
local_user@local_host$ jupyter notebook src/group_outlier_detection.ipynb 
```