3

Following is what my dataframe looks like:

  symbol      time    open    high     low   close
0   AAPL  09:35:00  219.19  219.67  218.38  218.64
1   AAPL  09:40:00  218.63  219.55  218.62  218.93
2   AAPL  09:45:00  218.91  219.09  218.27  218.44
3   AAPL  09:50:00  218.44  218.90  218.01  218.65
4   AAPL  09:55:00  218.67  218.79  218.08  218.59
5   AAPL  10:00:00  218.59  219.20  218.16  219.01

I am trying to apply a function from talib package that takes two arguments, high & low. Following is my attempt which returns all NaN:

import pandas as pd
import numpy as np
import talib as ta

def f(x):    
   return ta.SAR(df.high, df.low, acceleration=0.05, maximum=0.2)

df['PSAR1'] = df.groupby(['symbol']).apply(f)

However, the function works fine without a groupby clause and returns a number for the following:

df['PSAR2'] = ta.SAR(df.high,df.low, acceleration=0.05, maximum=0.2)

    symbol  time         open   high    low    close    PSAR1   PSAR2
0   AAPL     09:35:00   219.190 219.670 218.380 218.640 NaN     NaN
1   AAPL     09:40:00   218.630 219.550 218.620 218.930 NaN 218.380000
2   AAPL     09:45:00   218.910 219.090 218.270 218.440 NaN 219.550000
3   AAPL     09:50:00   218.440 218.900 218.010 218.650 NaN 219.550000
4   AAPL     09:55:00   218.670 218.790 218.080 218.590 NaN 219.396000
5   AAPL     10:00:00   218.590 219.200 218.160 219.010 NaN 219.257400

What am I doing wrong with apply with multiple arguments & groupby ?

EDIT: With @bsmith89's help, the following worked. 

def f(df):    
   return pd.DataFrame(ta.SAR(df.high, df.low, acceleration=0.05, maximum=0.2),columns= ['PSAR'])
y = df.groupby(['symbol']).apply(f)

df['PSAR'] = y.PSAR.reset_index(drop=True)
gibbz00
  • 1,947
  • 1
  • 19
  • 31

1 Answers1

3

You have written your function to take x as an argument, but then you operate on df instead.

I haven't tested it, but try rewriting as

def f(df):    
   return ta.SAR(df.high, df.low, acceleration=0.05, maximum=0.2)
bsmith89
  • 223
  • 2
  • 6
  • Thank you for noticing that. I thought your solution was going to do it. :) However, I still see NaN. – gibbz00 Nov 01 '18 at 03:54
  • When I'm troubleshooting out a "groupby" scheme, I like to see what comes back when I run the function with just one of the groups, so something like: `f(df.groupby('symbol').get_group('AAPL'))`. – bsmith89 Nov 03 '18 at 21:15
  • What ended up working? I'm not sure I actually answered your question, but I'm happy to update my answer to help others with the same problem. – bsmith89 Nov 09 '18 at 18:55
  • Edited the post to show what worked. Thank you for your help. – gibbz00 Nov 12 '18 at 15:29