jinzishuai / learn2deeplearn

A repository of codes in learning deep learning

GNU General Public License v3.0

13 stars 1 forks source link

Closed jinzishuai closed 6 years ago

jinzishuai commented 6 years ago

I think 2 is wrong since mini-batch can be batch when it is the biggest one.

only 1 wrong

jinzishuai commented 6 years ago

I didn't realize actually (2) is larger oscillation, which means smaller beta.

jinzishuai commented 6 years ago

no reason is given

jinzishuai commented 6 years ago

B is certainly true
C is not necessarily true. For a small sample-sized problem, the time might be the same doing it with mini-batch or a single batch with one epoch.
- Mini-batch might reach convergence sooner (without reaching full epoch)
- For very large problems, batch might be much slower due to limitations of CPU/RAM.