- SVMs use kernel functions to transform data into alternate dimensions.
- Seek a linearly separable plane where one does not exist in the input dimensions.
- SVMs do not require prior knowledge of the problem domain.
"Support Vector" - points closest to the boundary.
The key insight of SVMs is that some example are more important that others, and that paying attention to them can lead to better generalization.
- Maximum Margin Separator - decision bounder with the largest margin between support vectors.
- Kernel Trick - Use kernel functions on pairs of input data in different dimensions or feature spaces.
- Nonparametric - No need to tune hyperparameters.
$ aiw svm -h
usage: aiw svm [-h] -f DATASET_FILENAME
options:
-h, --help show this help message and exit
-f DATASET_FILENAME, --file DATASET_FILENAME
.csv file containing your dataset
(.venv-ai)
Dataset used was coronavirus prediction dataset here.
Upon first loading the dataset, this warning is output:
DtypeWarning: Columns (2,3,4,5,6,8) have mixed types. Specify dtype option on import or set low_memory=False.
This tells us that we need to clean whatever dataset we run an SVM model on.