4.0.1 Cng xy dựng Model Machine Learning với Bta v Bt

Thi c, tt c cc model d on (predictive models) u trng c v n, nhng lm cch no chng ta c th xy dng cc model mt cch ng tin cy hn? y chnh l cu hi ti thng c hi bi cc nh khoa hc d liu nhng mc kinh nghim khc nhau. Cu hi c v n gin, nhng ng thi cng rt thch thc bi v c rt nhiu lung quan im v mi quan tm khc nhau t nhng ngi lin quan.

Nhng ngi xy dng model thng tp trung vo vic hun luyn model (training model) mt cch t ng, kim sot hiu nng, debug v nhng hot ng duy tr c h thng t d liu n ti u hiu nng phn cng v ci tin model (MLOps). Ngi s dng model phn nhiu thng hng th hn vi nhng cu hi v t u m model cho ra kt qu d on nh vy (explainability) hoc nhng i hi nht nh hiu c bn trong model hot ng nh th no (transparency) v mc "phng th" (security) ca m hnh trc nhng tn cng, v d nh t d liu km cht lng. V pha x hi, nhng mi quan tm thng l model c cng bng hay thin v (fairness hay bias), v d m hnh c u tin cho mt nhm ngi nh trong x hi, hoc thin v v gii tnh khng; hay kt qu d on t model c ph hp chun mc o c hay khng (ethics).

Quyn sch ny tng ha tt c cc gc nhn trn. Bn c s tm thy nhng k thut machine learning chn lc v c nhng cch hiu trc quan ngn gn. Cc k thut u c gn lin vi cc on code trong ngn ng R. Bn c s cng hai nhn vt Bta v Bt i do vo khu vn, khng phi ca k hoa d tho, m l ca cc model ML th v. Trong khu vn c ngn n c thp ln bi nhng gc nhn thu o v model t kinh nghim thc t.

Cuc bn lun gia Bta v Bt cng chnh l nhng g din ra trong thc t m cc nh khoa hc d liu thng gp, c nn th mt model khc khng, hoc c nn th mt k thut khc khai ph d liu, hoc mt tp d liu khc chng? --- v ti t cu hi nh: lm cch no so snh gia nhng model hoc lm th no kim tra hiu nng.

Qu trnh pht trin model i hi s cn thn v trch nhim, nhng rt th v. Thng thng, nhiu quyn sch ch ch tm vo mt kin thc m qun mt i nim vui v s th v. Nhng, hi vng quyn sch ny, chng ta s c c c hai.