[數撚圍爐區] 理想與現實 (11)

metoo

聽過人講係因為中和作用

但係generalised likelihood ratio test個df就好合理

-2 log likelihood ratio 嗰個
去到advanced啲嘅course你就會知道佢係統計學三大test（wald、score、likelihood ratio）之一
定係你哋依家2602已經有教
幾時講到exponential family

exponential family係將啲常見嘅distribution（beta gamma poisson normal exponential 等等）擺入同一個framework度，好多時啲stat theorem都係apply晒所有exponential family distribution，所以唔駛逐個derive（例如你可以望下generalised linear model嘅literature）

score test就係字面意思，用score function做test

首先你要知道個normal distribution只係影響你hypothesis test、ci嗰啲inference嘢，唔會影響model fitting

應該話如果sample size大嘅時候，堆test、ci就robust to violations of normality

如果sample size細嘅話咪搵其他計囉例如用nonparametric bootstrap去test

等等先。。。你話2602而家有講linear regression？？？！！！？？？

bayesian得yin一個做，佢做咗dept head唔得閒教書，所以咪冇晒bayesian

係咪我哋以前喺tg講嘢個陣有啲溝通問題

我讀過1600, 1603, 2602, 3600, 3602, 3603, 3612, 4601, 4602, 而家讀緊3955同3617...

同埋做research有好多嘢要自己學自己睇

係已經...我final sem做緊directed study

cls寫一萬字