次に読むべき論文は? - Louppe, Gilles. "Understanding random forests: From theory to practice." arXiv preprint arXiv:1407.7502 (2014). - Chzhen, E., Hebiri, M., Salmon, J., et al. On lasso refitting strategies. Bernoulli 25, 4A (2019), 3175–3200. - Zhou, Z., and Hooker, G. Unbiased measurement of feature importance in tree-based methods. arXiv preprint arXiv:1903.05179 (2019). - Nori, H., Jenkins, S., Koch, P., and Caruana, R. Interpretml: A unified framework for machine learning interpretability. arXiv preprint arXiv:1909.09223 (2019). 要約
- 広く知られている話でscikit-learnにも解説有🐜 - Permutation Importance vs Random Forest Feature Importance (MDI) - 選ばれる特徴量が少ないほどuninformativeな奴の順 位が下落 → Unbiasedできてる Discussion and future work - Limits of LASSO: big p, small m - LASSO 〜 最大m特徴量 - Zou, H., and Hastie, T. Regularization and variable selection via the elastic net. Journal of the royal statistical society: series B (statistical methodology) 67, 2 (2005), 301–320. - Controlburn 〜 最大m^2特徴量(Sparse性能に差が!) - Controlburnだとグループとして採用したい特徴量も1つになる - Group Elastic Net(L2も入れるということ)で解決する? - Münch, M. M., Peeters, C. F., Van Der Vaart, A. W., and Van De Wiel, M. A. Adaptive group-regularized logistic elastic net regression. Biostatistics (2018)
Z., and Hooker, G. Unbiased measurement of feature importance in tree-based methods. arXiv preprint arXiv:1903.05179 (2019). - Louppe, Gilles. "Understanding random forests: From theory to practice." arXiv preprint arXiv:1407.7502 (2014). - Nori, H., Jenkins, S., Koch, P., and Caruana, R. Interpretml: A unified framework for machine learning interpretability. arXiv preprint arXiv:1909.09223 (2019). - 各種古典系論文(Brieman 2001, 2002, Strobl 2007, 2008) 所感