07/08/2020 - machine learning

i have tried andrew ng's coursera stanford and sanjoy dasgupta dse220x machine learning. despite making handwritten notes, looking at ng's notes on github, it didn't get me anywhere. the background theory is too complicated even though i know linear algebra, calculus, statistics etc. i spent weeks and i barely understand linear regression which is extremely basic. 

originally i wanted to understand the math concept to get an idea of how to tune. but it is too complex to visualise. i rather treat it like a blackbox first, and just learn the code and then deal with it later. anyway there are other important tasks in ML i have yet to learn and i think data cleaning and coding is way more practical vs knowing ML math in depth.

i am now looking at the scikit learn/sklearn docs and also at the kaggle titanic dataset walkthrough code. just by looking at the sklearn algorithms can get an idea of what people usually use. there's some explanation on the rationale for the algorithms too.

i have already found a medium article comparing various algorithms, so i just gonna tap on that. scikit learn also has a decision map for choosing algos.

since i have some coding knowledge, and some knowledge of linear regression, i think i will go back to the python ML related libraries first

wes mckinney's python for data analysis is pretty good. numpy, pandas etc

author filters and compiles commonly used functions in the book T-T

i love, love, cheatsheets and reference tables.

data science for business - foster provest et al

great book minus the math.

am also using powerpoint/google sheets to make ppt slides so as to teach myself.

cost and optimisation function (mse + gradient descent)

this is the clearest explanation i found

https://chelseatroy.com/2017/02/07/machine-learning-intuition-cost-function-optimization/

kdnuggets is a good website