https://github.com/ysig/kml_2021

Submission code for kml_2021
https://github.com/ysig/kml_2021

Last synced: 27 days ago
JSON representation

Submission code for kml_2021

• Host: GitHub
• URL: https://github.com/ysig/kml_2021
• Owner: ysig
• Created: 2021-03-24T16:30:03.000Z (almost 4 years ago)
• Default Branch: main
• Last Pushed: 2021-03-25T00:43:38.000Z (almost 4 years ago)
• Last Synced: 2024-10-05T21:06:44.520Z (3 months ago)
• Language: Python
• Homepage:
• Size: 1.01 MB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# kernels-2021

Kaggle-Ranking of Kernels (**Public Leaderboard**):

| Method  | Kaggle Accuracy |

| ------------- | ------------- |

| Linear-Regression  |  0.60466  |

| SVM  | 0.55866  |

| Kernel-Ridge-Regression | 0.61933 |

| Spectrum (k=6) | 0.62600 |

| Gappy (q=8, l=6) | 0.62866  |

| mismatch (m=1, k=8) | 0.65400 |

| mismatch (m=1, k=9) | 0.67333 |

**Final Ranking**  

Private Leaderboard: 0.67733  

(mismatch (m=1, k=9))

# Details about this code-base

Dependencies: cvxopt, scipy, numpy

You can run main.py as:

    `python main.py`

or with:

    `python main.py --dry`

for validation. In the latter case it needs `sciki-learn` for `train_test_split` and `accuracy_report`.

All the submitted kaggle files are in folder `predictions` and all the logs from validation are in the folder `logs`.

You can find the extensive evaluation logs with the most promising of the tested kernel parameters and C from 2^{-2}, ..., 2^{2} in 

The only kernel that is not in validation and in predictions and which has been implemented is the substring kernel, which

can be found at `substring.py`.

Files corresponding to kernels:

 - `spectrum.py`

 - `svm.py` and `gaussian.py`

 - `gappy.py`

 - `mismatch.py`

 - `substring.py`

Files corresponding to classifiers:

 - `kernel_svm.py`

 - `ridge_regression.py`

 - `kernel_ridge_regression.py`

Other files:

 - `training.py` corresponds to all the train and predict pipeline, for all methods.

 - `classes.py` used to define base classes for kernels and classifiers.

 - `argparser.py` used to define the simple command line argument of dry.

 - `utils.py` used to define a thresholding function.