狠狠撸

?? ???? ?????~?
https://www.youtube.com/watch?v=V1eYniJ0Rnk

????
- ??? ??
- ??? ????.
- ???? ??
- ?... ??? ?? ?? ???, ??? ?? ??? 32?? ??? ??????? ?.
- https://wikidocs.net/2825
- ? ??? python 3.6? ?? tensorflow-gpu? ?? ? ??? python 3.5? 4.2.0? ??.
- 4.2.0? tensorflow-gpu? ?? ??…
- 64?? ???? ??, anaconda x86 ??? ???? tensorflow cpu only ?? ?
? ??.

???
- ???? ? ? ?? ???...
- ?????? 32??? 64??? ??? ???.
(32?? ????? ??? tensorflow? ??? ?? ???…)
- http://stackoverflow.com/questions/33709391/using-multiple-python-engines-32bit-64bit-
and-2-7-3-5
- >set CONDA_FORCE_32BIT=1
- >conda create -n py35_32 python=3.5
- >set CONDA_FORCE_32BIT=
- >conda create -n py35_64 python=3.5
- python 3.5 32bit ??
- >set CONDA_FORCE_32BIT=1
- ?? ??...

tensorflow GPU ??
- GTX1070? ??? ?? ?? ?? ??
- cuda ??
- http://jaejunyoo.blogspot.com/2017/02/start-tensorflow-gpu-window-10.html
- https://www.tensorflow.org/install/install_windows
- cuda 8.0.61_win10 ??
- cudnn-8.0-windows10-x64-v6.0 ??
- python 3.5 64bit ???? ??
- hello world ?? ??

IDE
- pyCharm ??
- ??: ?? ? ??? pyCharm? ?? ? ???...
- https://www.jetbrains.com/pycharm/download/#section=windows
- ??? ????
- https://www.jetbrains.com/help/pycharm/2016.3/conda-support-creating-conda-
environment.html

?? ??
- ???? ?? (2017/3/20)
- ??: https://wikidocs.net/4231
- Open api ??
- Open api+ ?? ??
- Koa studio ??
- ?? ?? ??
- ?? 3???? ???? ?? ??

?? ??
- ??? ???? ???? ??
- ??? ?? ? ?? ??? ???? ?? 1? ???.

??? ??
- ???? ??
- ??? pandas datareader? ????? ??.
- ??? ? ?? ????… ?? ?? ? ????? ???? pandas datareader? ???.
- https://github.com/dspshin/stock-analyzer/blob/master/noti.py
- ???
- ??? ?? ??.
- ??? ??. ?? ? ???? ?? ????. --;
?? ?, ???? ???
?.

def set(self, *args):
return self.kiwoom.dynamicCall("SetInputValue(QString, QString)", *args)
STOCK_COLUMNS = ["???", "???", "??", "?????", "?????", "?????"]
?????? ????

?? ???? ?? ? DB? ??...
- ???? 1? delay? ?????…
- ???? 1? ????? ?? ???? ??? ?? ???? ???.
- ??? ????? ??.

???? ??? ?? ???
- sqlite3 to pandas
- http://stackoverflow.com/questions/36028759/how-to-open-and-convert-sqlite-database-to-
pandas-dataframe
- pandas tutorial
- https://github.com/jvns/pandas-cookbook/tree/master/cookbook
- ?? ?? ??
- ???? ??? ?? ?? ???, stockstats ??? ??
- https://github.com/jealous/stockstats
- ????? ?? ?? ??? ?? ?? ? ??

Test log #1 - ???? ???? ?? ??? ??? ?? ??.
- kospi + 34, 100, 20, 2 + 10 epoch + 0.001 learning rate =
- 2297 sec, 0.971548 accuracy
- kospi + 34, 100, 100, 20, 2 + 10 epoch + 0.001 learning rate =
- 2286 sec, 0.9713 accuracy
- kospi + 34, 200, 100, 20, 2 + 10 epoch + 0.001 learning rate =
- 2347 sec, 0.9699 accuracy
- kospi+kosdaq + 34, 200, 100, 20, 2 + 10 epoch + 0.001 learning rate =
- 5586 sec, 0.95725 accuracy

???
- accuracy? ??? ??? ? ???,
?? accuracy? ?? ? ???, ?? ? ? ?? ?? ? ???.
BUT ?? ???? ??? ?? ?? ?? ?? ? ?.
- ???, ?? ? ? ??? ??? ??? ??.
- ? ??? ????? ?? ??? ??? ??
- ?? ??? ?? ?, ????? ? ??? ???.

Test log #2
- ???? ???? ??? ????? ?? ??? ??.
- 200? ??? ???? ?? ??? ??.
- network : 35, 200, 3
- 20170502??? ???? ?? ? ?, 0504??? ??? ??.

??? ????? ?? ??
- ??? ?? ???? ?? ?? ?? ?? ??
- OOM ??
- Batch? ??? ??? ??? ??.
- ????? ? batch? ???? ??? ???. --;
- Np.save, load ???? ??? ???? ?? ??? ???.

Test log #3
- ~2017/05/04??? ???? ???, 0508? ??
- all + 34, 512, 1024, 512, 64, 3 + 0.0001 learning rate + 500 epoch
- 2.5 hours, accuracy 0.789907, 112 cands(all 1)
- all + 34, 512, 512, 512, 64, 3 + 0.0001 + 500
- 2.2 hours, accuracy 0.839576, 1 cands
- all + 34, 1024, 1024, 1024, 64, 3 + 0.0001 + 100
- 2.0 hours, accuracy 0.835347, 19 cands
- 001470 0
- 005740 0
- 005745 0
- 009275 0
- 023350 0
- 049770 0

...
- all + 34, 1024, 1024, 1024, 1024, 1024, 64, 3 + 0.0001 + 50
- 0.818191, 2+52 cands.
- …
- ? deep + wide ?? ?? ??? ?? ? ??...
- ??? ???? :
- ? ??? ?? ??? ??? ????

??? ??? ? ??
- ?? ???? ?? ???? ??? ????? ?????, ??? ? ??
? ?????? ??+?? ???? ?????
- 35 → 35 x 3 days
- 35 → 35 x 5 days
- Normalization? ??? ????
- ????: ? ??? ???? ??

?? ??
- ??? ???? ?? → 3? ?? ???? ?? (binary classification)
- layer: 35*5 , 512, 512, 512, 512, 512, 64, 2
- 0.0001 learning rate, 1000 epoch
- 20150504 ???? ??? ??

Accuracy ???? ??
- Accuracy ???? ??? ??? ???? ? ?? ??.
- recall, precision? ?? ??.
- recall = tp / (tp+fn)
- ?????, ?? True?? ??? True? ??? ?? ??
- precision = tp / (tp+fp)
- ?, ???? True? ??? ? ??? ?? True? ?? ??
????
Positive Negative
???? positive True Positive False Positive
???? negative False Negative True Negative

CNN ?
- ??? stockstat?? ??? ??????? ?? ??? ??? ?????
…
- ?? ??? ??
- ??? ?? ??? ?? ??.

RNN
- ?? ?? ??? ???? ???? ?? ???...
- LSTM ??.

LSTM ??
- ??? ??? ?? ??
- precision? ? ???? ??.
- facebook tensorflow KR ?????? ??
??? ???…
- https://www.facebook.com/groups/TensorFlowKR/
?multi_permalinks=473913729616357
- ???? ?? ? ?????.

Re-try LSTM
- facebook ?? ?, 3? ?? ???? ?? ?? ?? ??? training?? ??
?? ??? ????? ??.
- ?… ??? ?? ???...

News
- ??? ??? ? ??. ??? ???? ??????
- ??? ???? ????? ??? NLP? ??.
- konlpy ??
- http://konlpy.org/ko/stable/install/
- ????
- ???? ?/?? ????? scoring? ????.

Crawling News
- ?? ???? ??? ??, ???? ???? ??~
- ??? ?? ????? crawling
- ?? : http://finance.naver.com/news/market_special.nhn?&page=1
- field : datetime, title, url, content, …
- logic
- page=1 ?? ?? ??? ??? ???, ?? article? ?? ??? ???? ?? ? ?
?
- selecting articles : $(".boardList2 tbody tr")
- url? ?? ??? ??? ???? ??.

?? ??
- ????? histogram ????,
- ??? ??? ??? ????,
- +/- ??? ?? ???? ?? ????
??? ??? ??? ???.

To be continued...
- ? 3??? ??(+?? ?? ? ?)? ? ??? ???????.
- ?? ????, ?? ?? ????.
- ?? ?? ?? ?? ??;
- ???? ??? ???? ?? ? ????
- ?? ???? ??? ????.

狠狠撸

???? + ?? ???

Recommended

More Related Content

What's hot (20)

Similar to ???? + ?? ??? (20)

More from HoChul Shin (9)

???? + ?? ???