plan.md

Goal

1. Model a classic benchmark dataset
2. Reusable script design
3. Good metadata logging

Backlog

1. DONE - grids of outlier digits
2. DONE - grid of digit means
3. DONE - cumulative distribution plot
4. DONE - neural net modeling

High level procedure

1. Tune on 150000
2. Model on 240000

File structure

1. Exploratory data analysis script
2. Tuning script
3. Modeling script
4. figures/
5. tests/

Models

1. MLP
   • Variance Threshold, MinMaxScaler, Gridsearch hidden_layer_sizes, init_learning_rate
2. RandomForest
   • Variance Threshold, Gridsearch min_samples_leaf

Saved data

Contents	Filetype
metadata	meta_*.json
fitted estimator	grid_result*.pkl
images	*.png
log data	log_*.md

Future Work

1. @ROBUSTNESS Improve checking for whether files exist
2. @SIMPLIFY Simplify number of config parameters in tuning.py and model.py