Intellipaat Back

Explore Courses Blog Tutorials Interview Questions
0 votes
2 views
in Data Science by (17.6k points)

This is the setup:

arrays = [["2010-01-01","2010-01-01","2010-01-02","2010-01-02","2010-01-03","2010-01-03"],

                 ["MSFT", "AAPL", "MSFT", "AAPL","MSFT", "AAPL"]]

tuples = list(zip(*arrays))

index = pd.MultiIndex.from_tuples(tuples, names=["date", "symbol"])

df = pd.DataFrame(data=np.random.randn(6, 4), index=index, columns=["high", "low", "open", "close"])

def fn_sum(close, high, low):

    return close+high+low

def fn_plus(close):

        return close+1

The DF looks like this:

date       symbol   high        low        open        close

2010-01-01  MSFT  1.144042   0.889603   -0.193715   1.005927

            AAPL  0.433530  -0.291510    1.420505   0.326206

2010-01-02  MSFT -1.509419  -0.273476   -0.620735  -0.205946

            AAPL  0.454401  -0.085008    0.686485   1.309894

2010-01-03  MSFT  1.487588  -0.777500   -0.218993  -1.242664

            AAPL -0.456024  -0.819463   -2.224953   1.263124

I want to use technical analysis functions on all symbols with a groupby(), apply() fashion like this:

df["1"] = df.groupby(level="symbol").apply(lambda x: fn_sum(x["close"], x["high"], x["low"]))

This results in a broadcasting error:

ValueError: operands could not be broadcast together with shapes (6,2) (3,) (6,2)

Performing the same on a singular column works though:

df["2"] = df.groupby(level="symbol").close.apply(lambda x: fn_plus(x))

Questions:

So how do I get this to work when using apply on multiple columns and combining them back to a DataFrame without broadcasting issues?

Also I'm very grateful for a better implementation that works with MultiIndex DFs like above.

For more context: I want to use technical analysis functions from the TA-lib package. See:https://mrjbq7.github.io/ta-lib/func_groups/volatility_indicators.html

The functions look like this (for example):

ATR(high, low, close[, timeperiod=?])

Average True Range (Volatility Indicators)

Inputs: prices: ['high', 'low', 'close'] Parameters: timeperiod: 14 Outputs: real

I get the same broadcasting error as above in the contrived example.

1 Answer

0 votes
by (41.4k points)

Use DataFrame.join or DataFrame.assign for multiple columns:

s = (df.groupby(level="symbol", group_keys=False)

       .apply(lambda x: fn_sum(x["close"], x["high"], x["low"])))

df = df.join(s.rename('new'))

#alternative

#df = df.assign(new=s)

print (df)

                       high       low      open     close       new

date       symbol                                                  

2010-01-01 MSFT   -1.085631  0.997345  0.282978 -1.506295 -1.594580

           AAPL   -0.578600  1.651437 -2.426679 -0.428913  0.643924

2010-01-02 MSFT    1.265936 -0.866740 -0.678886 -0.094709  0.304487

           AAPL    1.491390 -0.638902 -0.443982 -0.434351  0.418136

2010-01-03 MSFT    2.205930  2.186786  1.004054  0.386186  4.778903

           AAPL    0.737369  1.490732 -0.935834  1.175829  3.403930

In case of only one column, use GroupBy.transform and specify column after groupby:

df['new1'] = df.groupby(level="symbol")['close'].transform(fn_plus) print (df) high low open close new1 date symbol 2010-01-01 MSFT -1.085631 0.997345 0.282978 -1.506295 -0.506295 AAPL -0.578600 1.651437 -2.426679 -0.428913 0.571087 2010-01-02 MSFT 1.265936 -0.866740 -0.678886 -0.094709 0.905291 AAPL 1.491390 -0.638902 -0.443982 -0.434351 0.565649 2010-01-03 MSFT 2.205930 2.186786 1.004054 0.386186 1.386186 AAPL 0.737369 1.490732 -0.935834 1.175829 2.175829

31k questions

32.8k answers

501 comments

693 users

Browse Categories

...