D****o
发帖数: 6 | 1 Last week in one phone interview, I am asked: in classification, random
forest is very good algorithm, so why do we need other methods?
Now the answer I can imagine is: in some cases, RF may be overqualified. For
example, if the classes are linearly separable, using logistic regression
can give the same accuracy and higher training efficiency.
So anyone has a better answer? | w******d
发帖数: 120 | 2 Regulation/ governance 要求，random forest 不行，比如银行
For
【在 D****o 的大作中提到】
: Last week in one phone interview, I am asked: in classification, random
: forest is very good algorithm, so why do we need other methods?
: Now the answer I can imagine is: in some cases, RF may be overqualified. For
: example, if the classes are linearly separable, using logistic regression
: can give the same accuracy and higher training efficiency.
: So anyone has a better answer?
| w******d
发帖数: 120 | 3 Unsupervised clustering 我觉得也可以看成classification ptobelm, RF也不行
For
【在 D****o 的大作中提到】
: Last week in one phone interview, I am asked: in classification, random
: forest is very good algorithm, so why do we need other methods?
: Now the answer I can imagine is: in some cases, RF may be overqualified. For
: example, if the classes are linearly separable, using logistic regression
: can give the same accuracy and higher training efficiency.
: So anyone has a better answer?
| w******d
发帖数: 120 | 4 AlphaGo 估计也不是RF
For
【在 D****o 的大作中提到】
: Last week in one phone interview, I am asked: in classification, random
: forest is very good algorithm, so why do we need other methods?
: Now the answer I can imagine is: in some cases, RF may be overqualified. For
: example, if the classes are linearly separable, using logistic regression
: can give the same accuracy and higher training efficiency.
: So anyone has a better answer?
| C**********t
发帖数: 24 | | N**N
发帖数: 1713 | 6 支持这个，另外就是rf也不一定是最好的，boosting在有的数据上会更好
也可以提一下这个：
https://en.wikipedia.org/wiki/No_free_lunch_theorem
【在 C**********t 的大作中提到】
: Interpretability?
| a*******e
发帖数: 253 | 7 Sometimes you need to explain the model
For
【在 D****o 的大作中提到】
: Last week in one phone interview, I am asked: in classification, random
: forest is very good algorithm, so why do we need other methods?
: Now the answer I can imagine is: in some cases, RF may be overqualified. For
: example, if the classes are linearly separable, using logistic regression
: can give the same accuracy and higher training efficiency.
: So anyone has a better answer?
| t******k
发帖数: 5617 | 8 people will prefer a model with 90% accuracy but tell you how and why than a
RF model with 95% accuracy but only tell you how
For
【在 D****o 的大作中提到】
: Last week in one phone interview, I am asked: in classification, random
: forest is very good algorithm, so why do we need other methods?
: Now the answer I can imagine is: in some cases, RF may be overqualified. For
: example, if the classes are linearly separable, using logistic regression
: can give the same accuracy and higher training efficiency.
: So anyone has a better answer?
| s*********h
发帖数: 6288 | 9 1.速度啊速度。
2.解释啊解释
3.软件支持
4.不够好啊不够好。NN beat RF的 情况很多的。
For
【在 D****o 的大作中提到】
: Last week in one phone interview, I am asked: in classification, random
: forest is very good algorithm, so why do we need other methods?
: Now the answer I can imagine is: in some cases, RF may be overqualified. For
: example, if the classes are linearly separable, using logistic regression
: can give the same accuracy and higher training efficiency.
: So anyone has a better answer?
| s***n
发帖数: 678 | 10 这是为什么？监管者为什么要管银行用什么算法？
是指这模型不能给出明确insights么，可能在未来出问题么。
【在 w******d 的大作中提到】
: Regulation/ governance 要求，random forest 不行，比如银行
:
: For
| l******n
发帖数: 9344 | 11 fed，occ的report都是有规定的，你只能在他们规定的框架内干事情，这就是
regulation， no why question.
【在 s***n 的大作中提到】
: 这是为什么？监管者为什么要管银行用什么算法？
: 是指这模型不能给出明确insights么，可能在未来出问题么。
|
|
