The synthetic dataset is composed of 10000 amount of 2 dimensional points belonging to 5 different classes. The data points are randomly sampled from Gaussian distributions with given mean and variance values. We have 6 different datasets with different difficulty levels. The difficulty of the datasets are determined by the mean and variance of the Gaussians the samples are taken from. The whole dataset is divided in two parts with the same size as training and test set. The color-coded scatter plots with increasing difficulty can be seen here.

Results of Ranking SGD | Results of Weighted Average Ranking SGD |

a) Dataset5 Ranking | a) Dataset5 Average Ranking |

b) Dataset6 Ranking | b) Dataset6 Average Ranking |

c) Dataset1 Ranking | c) Dataset1 Average Ranking |

d) Dataset2 Ranking | d) Dataset2 Average Ranking |

e) Dataset3 Ranking | e) Dataset3 Average Ranking |

f) Dataset4 Ranking | f) Dataset4 Average Ranking |

i. Comparison between different SGD methods

5 | 6 | 1 | 2 | 3 | 4 | |

w-OVR | 100 | 96.30 | 47.04 | 33.08 | 27.78 | 24.48 |

MUL | 100 | 95.70 | 44.96 | 26.70 | 23.32 | 22.36 |

OWA | 100 | 96.32 | 47.48 | 32.98 | 27.62 | 24.62 |

RNK | 100 | 96.28 | 47.76 | 33.24 | 27.66 | 24.30 |