data.fetcher_assets issue

https://www.quantopian.com/posts/data-dot-fetcher-assets-doesnt-work

May 24, 2016

same issue here. It seems that data.fetcher_assets keeps appending new stocks.

May 24, 2016

Hi Zhi,
I saw from your post that youre doing something really similar to me (and to the help we site).
Were you able to fully upload your table? Im unable to get all three columns (ie date, stock, and score) into Quantopian in one go, so now ive split my data into two tables: one with dates and stocks and one with dates and score. Then im calling fetch_csv twice, once for each table (which is ok for now but might get compliacated later, ie id like to do the whole thing in one fetch_csv).
Were you able to access your signal the way you imported it?
Thanks

May 24, 2016

Hi Dario,
The csv file is like this.
symbol start_date stock_score
AA 2/13/2012 0.876536739
WFM 2/13/2012 0.056418644
FDX 2/14/2012 0.976608748
M 2/15/2012 0.352700668
AAPL 2/16/2012 0.718324572

I am just doing a little experiment using example(with some changes) from quantopian (under fetcher)
https://www.quantopian.com/help

To access your signal, you can iterate through data.fetcher_assets and use data.current(asset,your_signal):

for asset in data.fetcher_assets:  
      if data.can_trade(asset):  
            order_target_percent(asset,data.current(asset,'score'))

You will find even if there is no score for AA on 2/14, It will still be re-balanced based on its prior day's score. Currently I am trying to find a way to close the stock if there is no score shown on that day. I think the issue comes from data.fetcher_assets.

Hey Zhi,
ok i figured out what was wrong with my import (was importing the wrong file).
So i did (a very ugly, brute force, work-around the date issue) that works for me for now (unitl the quantopian support gets back to me about the fetcher_assets) which seems to work.

In case your'e interested what i did is created a list with two date columns, so it looks like this:
Date,symbol,Signal,Date1
07.01.2014,AIG,-1,7012014
07.02.2014,CBI,1,7022014
... I add a second date column so i can work with these dates directly, which i wasnt really able to do with the column i set to date_column + fetch_csv doesnt let me upload a file without a date column (as far as i can tell) it keeps giving me an error if I try to.

Then in my algorithm i brute force check if the current date matches with the date i extract from the Date1 column from the table (quite tedious and slow), if it does i then add this security to a list of securities i want to trade...
definitely not pretty and will definitely try to change the code soon, but at least i can check some things already.

Zhi,
so i talked to the quantopian support: it seems this is an issue with the fetcher_assets, ie its not working correctly.

Simon Thornington

Ouch.

Yeah that works! Another solution would be to keep the same stock universe for each day:

07.01.2014,AIG,-1
07.01.2014,CBI,-1
07.01.2014,FMCC,1
07.02.2014,AIG,-1
07.02.2014,CBI,1
07.02.2014,FMCC,1
07.03.2014,AIG,1
07.03.2014,CBI,-1
07.03.2014,FMCC,1

Then you can get correct score for symbols on simulation date.

for asset in data.fetcher_assets:  
    print data.current(asset,'Signal')

May 26, 2016

Hi Zhi,
do you know if instead of using this double date column trick I am doing theres a way to access the trade dates directly?
Like for example data.current(data.fetcher_assets, 'tradeDate') or something?
thanks,
Dario

May 27, 2016

I don't know actually.. I think the date and symbol columns will become DataFrame index after fetch and will no longer be able to extract.

Mark Cheng

Oct 19, 2016

Hi folks - I ran into the same problem lately. Wondering if anyone found a more efficient way to solve this?

Thanks in advance