API change rationale?

[wilki2021]

So not necessarily an issue, but curious why you made the recent API change to eliminate the index that talib returns by default. Is there a way to get that value - it seems like it may be useful in certain contexts.

[mrjbq7]

I was a little conflicted -- the old way allocated only as much space as was needed to store the results from TA-Lib, which should be fast and exact. On the other hand, the new way makes it easy to integrate TA-Lib with pandas if all arrays share a common index (i.e., are the same length and the lookback period in the beginning is padded with nan), for example:

d = {}
d['MACD'] = talib.MACD(...)
d['EMA'] = talib.EMA(...)
d['RSI'] = talib.RSI(...)

columns = d.keys()
data = np.column_stack([d[key] for key in columns])
df = pandas.DataFrame(data, columns=columns)

If you've got API suggestions, I'm all for it. I could even generate both API's easily enough if there is interest...

[mrjbq7]

Also, I forgot the mention:

To get the "lookback" value that was previously returned, you can just find the index of the first non-nan value.

rsi = talib.RSI(...)
lookback = np.where(~np.isnan(rsi))[0][0]

[wilki2021]

Thanks. In hindsight, I was too quick too submit the question and think you
probably made the right decision. I am only doing some basic plotting with
matplotlib, but after I got the code changed for the new API I actually liked it
better. Thanks for the work to create this - it was very helpful to get ta-lib
easily integrated in Python.

---

From: John Benediktsson
reply@reply.github.com

To: wilki2021 bob.wilkinson@sbcglobal.net
Sent: Sun, July 1, 2012 8:07:08 PM
Subject: Re: [ta-lib] API change rationale? (#3)

I was a little conflicted -- the old way allocated only as much space as was
needed to store the results from TA-Lib, which should be fast and exact. On the
other hand, the new way makes it easy to integrate TA-Lib with pandas if all
arrays share a common index (i.e., are the same length and the lookback period
in the beginning is padded with nan), for example:

d = {}
d['MACD'] = talib.MACD(...)
d['EMA'] = talib.EMA(...)
d['RSI'] = talib.RSI(...)

columns = d.keys()
data = np.column_stack([d[key] for key in columns])
df = pandas.DataFrame(data, columns=columns)

If you've got API suggestions, I'm all for it. I could even generate both API's
easily enough if there is interest...

---

Reply to this email directly or view it on GitHub:
#3 (comment)