calling a function on dataframe data

Question

I am not sure what I am doing wrong here, I am simply trying to call a function with a if-then-else filter in it and apply to a dataframe.

In [7]: df.dtypes Out[7]: Filler float64 Spot float64 Strike float64 cp object mid float64 vol float64 usedvol float64 dtype: object In [8]: df.head() Out[8]: Filler Spot Strike cp mid vol 0 0.0 100 50 c 0.0 25.0 1 0.0 100 50 p 0.0 25.0 2 1.0 100 55 c 1.0 24.5 3 1.0 100 55 p 1.0 24.5 4 2.5 100 60 c 2.5 24.0

I have the below function:

def badvert(df): if df['vol']>24.5: df['vol2'] = df['vol']*2 else: df['vol2'] = df['vol']/2 return(df['vol2'])

Which I call here:

df['vol2']=badvert(df)

Which generates this error message:

--------------------------------------------------------------------------- ValueError Traceback (most recent call last) <ipython-input-14-bbf7a11a17c9> in <module>() ----> 1 df['vol2']=badvert(df) <ipython-input-13-6132a4be33ca> in badvert(df) 1 def badvert(df): ----> 2 if df['vol']>24.5: 3 df['vol2'] = df['vol']*2 4 else: 5 df['vol2'] = df['vol']/2 C:\Users\camcompco\AppData\Roaming\Python\Python34\site-packages\pandas\core\generic.py in __nonzero__(self) 712 raise ValueError("The truth value of a {0} is ambiguous. " 713 "Use a.empty, a.bool(), a.item(), a.any() or a.all()." --> 714 .format(self.__class__.__name__)) 715 716 __bool__ = __nonzero__ ValueError: The truth value of a Series is ambiguous. Use a.empty, a.bool(), a.item(), a.any() or a.all().

My gut tells me that this is a simple "syntax" issue but I am at a loss. Any help would be greatly appreciated

Padraic Cunningham · Accepted Answer · 2015-06-27 16:51:38Z

You want to apply it to each row, this will do what you want using apply and a lambda:

df["vol2"] = df.apply(lambda row: row['vol'] * 2 if row['vol'] > 24.5 else row['vol'] / 2, axis=1) print(df)

Which should output something like:

 Filler Spot Strike cp mid vol vol2 0 0.0 100 50 c 0.0 25.0 50.00 1 0.0 100 50 p 0.0 25.0 50.00 2 1.0 100 55 c 1.0 24.5 12.25 3 1.0 100 55 p 1.0 24.0 12.00 4 2.5 100 60 c 2.5 24.0 12.00

Or using your own function:

def badvert(df): if df['vol']>24.5: df['vol2'] = df['vol']*2 else: df['vol2'] = df['vol']/2 return df['vol2'] df["vol2"] = df.apply(badvert,axis=1)

axis=0 applies the function to each column, axis=1 applies function to each row.

unutbu · Accepted Answer · 2015-06-27 18:30:29Z

df.apply has performance comparable to a Python for-loop. Sometimes using apply or a for-loop to compute row-by-row is unavoidable, but in this case a quicker alternative would be express the calculation as one done on whole columns.

Because of the way the underlying data is stored in a DataFrame, and since there are usually many more rows than columns, calculations done on whole columns is usually quicker than calculations done row-by-row:

df['vol2'] = np.where(df['vol']>24.5, df['vol']*2, df['vol']/2)

thanks for the added color . . . .I just checked, in my application, this is 5.3 times faster than my "def" solution and 2.8 times faster than the df,apply approach. Thank you

Collectives™ on Stack Overflow

calling a function on dataframe data

2 Answers 2

1 Comment

1 Comment

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

1 Comment

Related