How to invent your own new ML algorithm: A novel ensembling technique
Fig. 1: The diagrammatic representation of weighted average
Fig. 2: A visualization of the curse of dimensionality. A dataset of 20 samples was generated from a random uniform distribution and the distance of each sample from one another was mapped resulting in this 20x20 distance matrix. As the number of dimensions increase, the data points move further apart. This can be seen above as the cells in the matrix transition from a deeper color set (indicating smaller distance) to a lighter color set (indicating larger distance). Thus, for higher dimensions, an exponential number of samples are required.
Fig.3: A visualization of the decision tree classifier on the Iris dataset. Clearly the model is creating partitions in sample space
Code 1: Notebook for dynamic weighted averaging (dynamic blending) on AnalyticVidhya’s Janatahack: Customer Segmentation Hackathon dataset
Code 2: Notebook for dynamic weighted averaging (dynamic blending) on AnalyticVidhya’s Janatahack: ML for Banking Hackathon dataset
Table 1: The first row scores are in terms of accuracy while the second row scores are in terms of F1-micro. Bold values indicate the best performance
References
[1] Wolpert, David H., and William G. Macready. "No free lunch theorems for optimization." IEEE transactions on evolutionary computation 1, no. 1 (1997): 67-82.
Cite us
For attribution in academic contexts or open source libraries, please cite this work as:
@article{hassanbasu2021blending,
title={How to invent your own new ML algorithm: A novel ensembling technique},
author={Hassan, Atif and Basu, Sayantan},
year={2021},
howpublished = {\url{https://www.deepwizai.com/projects/how-to-invent-your-own-new-ml-algorithm-a-novel-ensembling-technique}}
}