http://www.cs.berkeley.edu/~bartlett/courses/2009fall-cs281a/slides22.pdf Semiconjugate prior 찾다가 발견한 문서인데 monte carlo 에대해 정리가 잘 되어있어 올려둡니다.

https://communities.sas.com/t5/SAS-Statistical-Procedures/PROC-LOGISTIC-vs-PROC-GENMOD-different-results/td-p/27599 추정된 계수가 서로 다르게 나온다는 점에 유의.

http://ais.informatik.uni-freiburg.de/teaching/ss11/seminar_pgm/pgm-basics.pdf Factor graphs, Undirected graphs / Markov networks, Directed graphs / Bayesian networks 에 대한 쉬운 설명이 되어 있습니다.

https://www.quora.com/What-is-Least-Angle-Regression-and-when-should-it-be-used LARS, then, is essentially forward stagewise made fast. Instead of making tiny hops in the direction of one variable at a time, LARS makes optimally-sized leaps in optimal directions. These directions are chosen to make equal angles (equal correlations) with each of the variables currently in our model.

Philosophy and the practice of Bayesian statistics 베이지안 통계는 사후확률로 모델을 서로 비교하는 귀납적 추론이 아니라 모델 검토를 통해 모델 확장 및 변경을 수행하는 가설연역법이다.

https://en.m.wikipedia.org/wiki/Hypothetico-deductive_model The hypothetico-deductive model or method is a proposed description of scientific method. According to it, scientific inquiry proceeds by formulating a hypothesis in a form that could conceivably be falsified by a test on observable data. A test that could and does run contrary to predictions of the hypothesis is taken as a falsification…

http://www.johndcook.com/blog/2010/03/30/statistical-rule-of-three/ N/3. 귀무가설이 확률 = p 라고 할때 관찰된 사실이 귀무가설을 기각하지 않게 하면 N/3이 됩니다. 베이지안의 경우도 마찬가지 결과.

keras Kaggle 블로그 보다가 발견한 라이브러리 입니다. 설명은 다음과 같습니다. Keras is a minimalist, highly modular neural network library in the spirit of Torch, written in Python, that uses Theano under the hood for optimized tensor manipulation on GPU and CPU. It was developed with a focus on enabling fast experimentation. Kaggle의 competition에서 1등한 분들이…

Elements of Statistical Learning을 읽다가 도무지 이해가 안가서 유튜브에서 찾아서 모아봤습니다. 순서대로 보시면 됩니다. 그외에도 유튜브에서 spectral clustering을 검색해보면 좋은 강의가 많이 있습니다.

Boruta algorithm is a feature selection algorithm. Its purpose is to find all relevant features, and it does that by comparing Z score of original features and shuffled features. Shuffling here is similar to the idea in permutation tests. Z score of a variable in random forest is average loss when values of a feature…