Sample Size Estimation and Model Selection Technique library affords methods for preliminary data analysis and effective dataset size estimation with respect to small data subset.
Key advantages:
- fast data analysis
- for 50 features SampleSizeEstimation requires around 30 examples for accurate prediction
- visualization of hypothetic train error and data features provided
Sample size estimation methods are in code/lib/m_models.py. Each method takes a dataset with labels X, y and returns a dictionary with results of the estimation. Estimated optimal sample size is m*. Demonstration of usage of one of the methods can be seen in Demo.ipynb.