Module 7 Test: 20 MCQs
Ye tumhara progress tracker hai. 20 sawal: 5 Easy, 10 Moderate, 5 Hard . Pehle sab sawal khud hal karo (code ko apne computer par chalane se pehle apne dimagh se output sochna), phir Answer Key se check karo.
| Score | Matlab |
|---|---|
| 18 se 20 | Zabardast! Module 7 par tumhari grip mazboot hai. |
| 14 se 17 | Bohat acha! Kamzor chapter dobara parh lo. |
| 10 se 13 | Theek hai. Code khud chala kar dobara dekho. |
| 10 se kam | Fikr nahi. Chapter 34 se dobara shuru karo aur har code apne computer par chalao. |
EASY
Q1. MLOps ka sab se sahi matlab kya hai?
A) Sirf Docker chalana
B) ML model ko bharosemand tareeqe se banana, chalana aur sambhalna
- C) Model ki accuracy barhana
D) Sirf data saaf karna
Q2. FastAPI me data bhej kar prediction lene ke liye aam tor par kaun sa HTTP method use hota hai?
- A) GET
B) DELETE
C) POST
D) HEAD
Q3. Docker me image kya hoti hai?
A) App aur environment ka tayyar packet (saancha) jis se containers bante hain
- B) Ek chalta hua container
C) Ek tasveer ki file
D) Ek GitHub repository
Q4. MLflow me ek single training ki koshish ko kya kehte hain?
A) Experiment
B) Alias
C) Artifact
D) Run
Q5. Python logging me level INFO rakhne par kaun si line nazar nahi aati?
A) error wali line
B) debug wali line (logger.debug)
-
C) warning wali line
-
D) critical wali line
MODERATE
Q6. Container ke andar chalte FastAPI ko apne browser se kholne ke liye kya zaroori hai?
A) Sirf Dockerfile me EXPOSE 8000 likhna
B) Server ko --reload ke sath chalana
C) Server ko --host 0.0.0.0 par chalana aur docker run me -p se port mapping dena
- D) Container ko do baar chalana
Q7. Pydantic me Field(gt=0, lt=15) lagane ka nateeja kya hota hai?
A) Qeemat 0 se bari aur 15 se chhoti ho, warna FastAPI 422 deta hai
B) Qeemat hamesha 15 ho jati hai
-
C) Model tez ho jata hai
-
D) Ye sirf documentation ke liye hai
Q8. Dockerfile me requirements.txt ko code se pehle copy karke libraries install kyun karte hain?
-
A) Docker ka sakht qanoon hai ke txt files pehle aati hain
-
B) Isse image sab se chhoti ban jati hai
-
C) Taake image ki tarteeb alphabet ke mutabiq ho
-
D) Taake sirf code badalne par libraries ka bhaari install step dobara na chale (layer cache)
Q9. Workflow me docker-build job ke andar needs: test likha hai. Iska matlab kya hai?
-
A) Test job do baar chalegi
-
B) docker-build job tab hi chalegi jab test job pass ho
-
C) Dono jobs ek hi machine par chalengi
-
D) test job optional ho jati hai
Q10. Tum ne mlflow ui chalaya lekin koi run nazar nahi aaya, jabke training chali thi. Sab se mumkin wajah?
-
A) UI ko galat folder ya galat tracking URI se chalaya (training me sqlite:///mlflow.db tha)
-
B) Internet slow hai
-
C) Python ka version purana hai
-
D) Browser cache band hai
Q11. Neeche ka code chalane par kaun si lines chhapengi?
import logging
logging.basicConfig(level=logging.WARNING, format="%(levelname)s: %(message)s")
logger = logging.getLogger("t")
logger.info("a")
logger.warning("b")
logger.error("c")
logger.debug("d")
- A) Chaaron: a, b, c, d
B) Sirf b, c aur d
C) Sirf WARNING: b aur ERROR: c
D) Sirf ERROR: c
**Q12. Live umar vs training umar par KS test ka
p = 0.0000
aaya. Iska sahi matlab kya hai?**
-
A) Test kharab ho gaya
-
B) Model ki accuracy 100% hai
-
C) Dono data bilkul ek jaise hain
D) Taqseem me bohat wazeh farq hai: drift ka saboot
Q13. Medical API ke logs me kya likhna sab se munasib hai?
-
A) User ka naam aur poori symptoms ki list
-
B) Symptoms ki ginti, top prediction aur latency (naam, phone ya ID nahi)
-
C) User ka password
D) User ka phone number
Q14. MLflow Model Registry me champion alias ka sab se bara faida kya hai?
-
A) Docker image ki zaroorat khatam ho jati hai
-
B) Model ki accuracy khud barh jati hai
C) Deployment code models:/naam@champion hi rehta hai; alias naye version par muntaqil karna kaafi hai
- D) Model ka size chhota ho jata hai
Q15. FastAPI me model ko predict() function ke andar load karna kyun theek nahi?
- A) Har request par dobara load hoga aur API bohat sust ho jayegi
B) Python ye allow nahi karta
C) Model kharab ho jata hai
D) Swagger /docs nahi chalta
HARD
Q16. Neeche ke code me 5 jodiyan add ho chuki hain. Aakhri (5th) add ke baad m.check() kya dega?
m = AccuracyMonitor(window=4, threshold=0.75)
for p, a in [(1, 1), (2, 2), (1, 2), (3, 3), (1, 3)]:
m.add(p, a)
print(m.check())
# (AccuracyMonitor wohi class hai jo Chapter 39 me hai)
A) OK: accuracy 60%
B) ALERT: accuracy 60% threshold 75% se neeche hai C) OK: accuracy 50%
D) ALERT: accuracy 50% threshold 75% se neeche hai
Q17. Tests local par pass hain, lekin GitHub Actions me ModuleNotFoundError: sklearn aaya. Sab se mumkin wajah?
A) GitHub ka Python version galat hai
B) Library requirements.txt me likhi hi nahi thi; local par pehle se install thi
C) pytest ka naam galat hai
D) YAML file ka rang galat hai
Q18. Model scikit-learn 1.8 se train hua, lekin Docker image me requirements me version pin nahi tha aur naya version install ho gaya. Model load par ajeeb error ya warning aayi. Sab se behtar hal?
A) Docker hata do aur sirf local chalao
B) Model ko dobara save kar do aur bas umeed rakho
C) requirements.txt me wohi scikit-learn version pin karo jis se model train hua tha
D) Sirf pytest dobara chalao
Q19. Do MLflow runs ki test accuracy barabar 0.900 hai: ek 25 trees ka chhota model, doosra 100 trees ka bara model. Test set sirf 30 samples ka hai. Sab se samajhdar faisla?
A) Chhota aur tez model chuno aur zyada data ya cross-validation se dobara check karo, kyunke 30 samples me 1 galti 0.033 ka farq banati hai
B) 100 trees wala hamesha behtar hota hai
C) Dono bekaar hain
D) Jo pehle train hua usay chuno
Q20. Live app me umar ka average training (35) se 48 ho gaya. Asal sahi jawab (ground truth) 2 mahine baad milta hai. Sab se munasib pehla qadam kya hai?
A) KS test ko nazar-andaz karo
B) Logs band kar do
C) Foran naya model deploy kar do
D) Pehle tasdeeq karo ke data me bug to nahi, phir naya data jama kar ke retrain karo aur MLflow/CI/CD se deploy karo
Module 7 Answer Key (Jawabat aur wazahat)
| # | Level | Jawab | Wazahat |
|---|---|---|---|
| 1 | Easy | B) ML model ko bharosemand tareeqe se banana, chalana aur sambhalna | MLOps = ML + DevOps + Data Engineering: poora lifecycle (data, training, deployment, monitoring). Docker iska sirf ek hissa hai. |
| 2 | Easy | C) POST | Prediction ke liye features bhejne parte hain, isliye /predict POST hota hai. GET sirf maloomat lene ke liye (jese /health). |
| # | Level | Jawab | Wazahat |
|---|---|---|---|
| 3 | Easy | A) App aur environment ka tayyar packet (saancha) jis se containers bante hain | Image saancha hai (sirf padhne wala), container us ki chalti hui copy hai. Ek image se kai containers chal sakte hain. |
| 4 | Easy | D) Run | Run = ek training ki koshish. Experiment un runs ka folder hai, aur Artifact run ke sath rakhi hui file (maslan model). |
| 5 | Easy | B) debug wali line ( logger.debug) | Level ka matlab: us darje ya us se oopar ke messages dikhao. DEBUG, INFO se neeche hai, isliye chhup jati hai. |
| 6 | Moderate | C) Server ko --host 0.0.0.0 par chalana aur docker run me -p se port mapping dena | Agar server sirf 127.0.0.1 par sune to bahar se koi nahi pohanch sakta. EXPOSE sirf documentation hai; asal rasta -p se khulta hai. |
| 7 | Moderate | A) Qeemat 0 se bari aur 15 se chhoti ho, warna FastAPI 422 deta hai | gt = greater than, lt = less than. Galat qeemat model tak pohanchti hi nahi; FastAPI foran 422 wapas kar deta hai. |
| 8 | Moderate | D) Taake sirf code badalne par libraries ka bhaari install step dobara na chale (layer cache) | Docker har line ka nateeja cache karta hai. Libraries pehle install hon to code badalne par sirf neeche wale steps dobara chalte hain, aur build tez hoti hai. |
| 9 | Moderate | B) docker-build job tab hi chalegi jab test job pass ho | needs jobs ki tarteeb tay karta hai. Agar test fail ho to docker-build chalti hi nahi, aur kharab code image tak nahi pohanchta. |
| 10 | Moderate | A) UI ko galat folder ya galat tracking URI se chalaya (training me sqlite:///mlflow.db tha) | Record mlflow.db (ya mlruns) me hota hai. UI ko wohi --backend-store- uri dena parta hai jo training me tha, warna wo khali jagah dekhta hai. |
| 11 | Moderate | C) Sirf WARNING: b aur ERROR: c | Level WARNING hai, to WARNING aur us se oopar (ERROR, CRITICAL) dikhte hain. info (a) aur debug (d) chhup jate hain. |
| 12 | Moderate | D) Taqseem me bohat wazeh farq hai: drift ka saboot | p chhota (< 0.05) ka matlab hai ke itna farq ittefaq se nahi aaya. Lekin KS ki qeemat aur asal averages bhi dekhne chahiyen. |
| 13 | Moderate | B) Symptoms ki ginti, top prediction aur latency (naam, phone ya ID nahi) | Health data bohat hassas hota hai. Logs me sirf utna likho jitna masla dhoondne ke liye zaroori ho; nijji maloomat kabhi nahi. |
| 14 | Moderate | C) Deployment code models:/naam@champion hi rehta hai; alias naye version par muntaqil karna kaafi hai | Alias se production wala model tay hota hai. Naya model behtar ho to sirf alias badalte hain; API ka code nahi badalna parta. |
| 15 | Moderate | A) Har request par dobara load hoga aur API bohat sust ho jayegi | Model ko module ke upar ek baar load karo. Phir har request usi load hue model ko istemal karti hai. |
| 16 | Hard | D) ALERT: accuracy 50% threshold 75% se neeche hai | window=4 hai, to aakhri 4 nateeje yaad rehte hain: [True, False, True, False] = 50%. Window bhar chuki hai aur 50% < 75% hai, isliye ALERT aata hai. |
| 17 | Hard | B) Library requirements.txt me likhi hi nahi thi; local par pehle se install thi | CI ka robot khali machine hota hai: wahan sirf wohi install hota hai jo requirements.txt me likha ho. Local par library pehle se thi. |
| 18 | Hard | C) requirements.txt me wohi scikit-learn version pin karo jis se model train hua tha | Pickle/joblib models library version se jure hote hain. Training aur deployment dono me ek hi version pin karna chahiye. |
| 19 | Hard | A) Chhota aur tez model chuno aur zyada data ya cross- validation se dobara check karo, kyunke 30 samples me 1 galti 0.033 ka farq banati hai | Barabar accuracy par chhota model sasta aur tez hota hai. 30 samples bohat kam hain, isliye faisla zyada data ya cross-validation par karna chahiye. |
| 20 | Hard | D) Pehle tasdeeq karo ke data me bug to nahi, phir naya data jama kar ke retrain karo aur MLflow/CI/CD se deploy karo | Drift sach me bhi ho sakta hai aur data pipeline ka bug bhi. Pehle tasdeeq, phir naya data, dobara training, MLflow me compare, aur CI/CD se deploy. |
Tumhara score: ______ / 20