(۱۶ اسفند ۱۳۹۳ ۰۲:۴۳ ق.ظ)cavalier نوشته شده توسط:  بله از همون فرمول به دست اوردم -۴/۳ میشد و -۲
که تو گزینه ها نبود.. sw-1 هم که این طوری حساب کردم.. sw=sig1^2+sig2^2 که میشه جمع عناصر قطری مماتریبس کواریانس
میشد
[۳,۰;۰,۲] که معکوس میکردی و در تفاضل میانگینها ضرب تو گزینه ها نبود!!
احتمالات پریوری هم ریختم تو سطل اشغال خخخخخخخخخ. تو فیشر که احتمالات پریوری نیست مگه بیزینه!!
فیشر دسته بندی رو بر اساس فاصله میانکینها از هم (زیاد باشن)و واریانس درون کلاسی (کم باشن) بر روی خط تابیده شدش به دست میاره.

---

رامین دمت گرمممم! موفق باشی مشارکت کن

(۱۶ اسفند ۱۳۹۳ ۰۱:۰۹ ق.ظ)mhghna نوشته شده توسط:  سلام
امسال هم مثل پارسال خیلی از سوالها تخصصی ، ترجمه شده سوالهای زبان اصلی دانشگاه های دیگه هست
اونهایی که یادم میاد پیدا می کنم اینجا میزارم تا سوالات بزان تو سایت

(a) [1 point] We can get multiple local optimum solutions if we solve a linear regression problem by
minimizing the sum of squared errors using gradient descent.
True False
Solution:
False

گزینه ۴ بود درسته ؟

---

اینم یه سوال دیگه :
© [1 point] When the hypothesis space is richer, over fitting is more likely.
True False
Solution:
True فکر کنم گزینه ۱ می شد (که گزینه غلط رو می خواست گفته بود فضای فرضیه کوچکتر بیش برازش بیشتر)
(d) [1 point] When the feature space is larger, over fitting is more likely.
True False
Solution:
True
(e) [1 point] We can use gradient descent to learn a Gaussian Mixture Model.
True False
Solution:
True

اینم یه سوال دیگه بود
(۲ points) Which of the following statements are true for k-NN classiers (circle all answers
that are correct).
۱/ The classication accuracy is better with larger values of k.
۲/ The decision boundary is smoother with smaller values of k.
۳/ k-NN is a type of instance-based learning. که دنبال این گزینه بود
۴/ k-NN does not require an explicit training step.
۵/ The decision boundary is linear.
Solution: 3 and 4

---

اینم یکی دسگه اش که واسه خودم متاسفم و اشتباه زدم
© (2 points) For the data above we are now comparing the following two models:
A: y = w2
۱x + w2x
B: y = wx
Note that model A now uses two parameters (though both multiply the same input value,
x). Again we assume unlimited data. Which of the following is correct (chose the answer
that best describes the outcome):
۱/ There are datasets for which A would perform better than B
۲/ There are datasets for which B would perform better than A
۳/ Both 1 and 2 are correct.
۴/ They would perform equally well on all datasets.
Solution: 4. Since we have unlimited data, the parameters learned for model B would
be equal to the some of the parameters learned for model A : w = w2
۱ + w2.

---

اینم یه سوال دیگه اش
(a) (2 points) Which of the following procedures is sucient and necessary and most ecient
for proving that the VC dimension of a learner is N?
۱/ Show that the classier can shatter all possible dichotomies with N points.
۲/ Show that the classier can shatter a subset of all possible dichotomies with N points.
۳/ Show that the classier can shatter all possible dichotomies with N points and that it
cannot shatter any of the dichotomies with N+1 points.
۴/ Show that the classier can shatter all possible dichotomies with N points and that it
cannot shatter one of the dichotomies with N+1 points.
۵/ Show that the classier can shatter a subset of all possible dichotomies with N points
and that it cannot shatter one of the dichotomies with N+1 points.
Solution: 4

---

سوال بعدی:
(۲ points) Which of the following statements are true for BOTH decision trees and Nave
Bayes classiers (you may chose more than one statement):
۱/ In both classiers a pair of features are assumed to be independent
۲/ In both classiers a pair of features are assumed to be dependent
۳/ In both classiers a pair of features are assumed to be independent given the class
label
۴/ In both classiers a pair of features are assumed to be dependent given the class label

Solution: 2. In both classiers features are not assumed to be independent. In Nave
Bayes they are assumed to be independent only when given the class label. For decision
trees they are not assumed to be independent even if the class label is provided.