How to deal with "divide by zero" with pandas dataframes when manipulating columns?
Question:
I’m working with hundreds of pandas dataframes. A typical dataframe is as follows:
import pandas as pd
import numpy as np
data = 'filename.csv'
df = pd.DataFrame(data)
df
one two three four five
a 0.469112 -0.282863 -1.509059 bar True
b 0.932424 1.224234 7.823421 bar False
c -1.135632 1.212112 -0.173215 bar False
d 0.232424 2.342112 0.982342 unbar True
e 0.119209 -1.044236 -0.861849 bar True
f -2.104569 -0.494929 1.071804 bar False
....
There are certain operations whereby I’m dividing between columns values, e.g.
df['one']/df['two']
However, there are times where I am dividing by zero, or perhaps both
df['one'] = 0
df['two'] = 0
Naturally, this outputs the error:
ZeroDivisionError: division by zero
I would prefer for 0/0 to actually mean “there’s nothing here”, as this is often what such a zero means in a dataframe.
(a) How would I code this to mean “divide by zero” is 0 ?
(b) How would I code this to “pass” if divide by zero is encountered?
Answers:
Two approaches to consider:
Prepare your data so that never has a divide by zero situation, by explicitly coding a “no data” value and testing for that.
Wrap each division that might result in an error with a try
/except
pair, as described at https://wiki.python.org/moin/HandlingExceptions (which has a divide by zero example to use)
(x,y) = (5,0)
try:
z = x/y
except ZeroDivisionError:
print "divide by zero"
I worry about the situation where your data includes a zero that’s really a zero (and not a missing value).
df['one'].divide(df['two'])
Code:
import pandas as pd
import numpy as np
df = pd.DataFrame(np.random.rand(5,2), columns=list('ab'))
df.loc[[1,3], 'b'] = 0
print(df)
print(df['a'].divide(df['b']))
Result:
a b
0 0.517925 0.305973
1 0.900899 0.000000
2 0.414219 0.781512
3 0.516072 0.000000
4 0.841636 0.166157
0 1.692717
1 inf
2 0.530023
3 inf
4 5.065297
dtype: float64
It would probably be more useful to use a dataframe that actually has zero in the denominator (see the last row of column two
).
one two three four five
a 0.469112 -0.282863 -1.509059 bar True
b 0.932424 1.224234 7.823421 bar False
c -1.135632 1.212112 -0.173215 bar False
d 0.232424 2.342112 0.982342 unbar True
e 0.119209 -1.044236 -0.861849 bar True
f -2.104569 0.000000 1.071804 bar False
>>> df.one / df.two
a -1.658442
b 0.761639
c -0.936904
d 0.099237
e -0.114159
f -inf # <<< Note division by zero
dtype: float64
When one of the values is zero, you should get inf
or -inf
in the result. One way to convert these values is as follows:
df['result'] = df.one.div(df.two)
df.loc[~np.isfinite(df['result']), 'result'] = np.nan # Or = 0 per part a) of question.
# or df.loc[np.isinf(df['result']), ...
>>> df
one two three four five result
a 0.469112 -0.282863 -1.509059 bar True -1.658442
b 0.932424 1.224234 7.823421 bar False 0.761639
c -1.135632 1.212112 -0.173215 bar False -0.936904
d 0.232424 2.342112 0.982342 unbar True 0.099237
e 0.119209 -1.044236 -0.861849 bar True -0.114159
f -2.104569 0.000000 1.071804 bar False NaN
You can always use a try statement:
try:
z = var1/var2
except ZeroDivisionError:
print ("0") #As python-3's rule is: Parentheses
OR…
You can also do:
if var1==0:
if var2==0:
print("0")
else:
var3 = var1/var2
Hope this helped! Choose whichever choice you desire (they’re both the same anyways).
I’m working with hundreds of pandas dataframes. A typical dataframe is as follows:
import pandas as pd
import numpy as np
data = 'filename.csv'
df = pd.DataFrame(data)
df
one two three four five
a 0.469112 -0.282863 -1.509059 bar True
b 0.932424 1.224234 7.823421 bar False
c -1.135632 1.212112 -0.173215 bar False
d 0.232424 2.342112 0.982342 unbar True
e 0.119209 -1.044236 -0.861849 bar True
f -2.104569 -0.494929 1.071804 bar False
....
There are certain operations whereby I’m dividing between columns values, e.g.
df['one']/df['two']
However, there are times where I am dividing by zero, or perhaps both
df['one'] = 0
df['two'] = 0
Naturally, this outputs the error:
ZeroDivisionError: division by zero
I would prefer for 0/0 to actually mean “there’s nothing here”, as this is often what such a zero means in a dataframe.
(a) How would I code this to mean “divide by zero” is 0 ?
(b) How would I code this to “pass” if divide by zero is encountered?
Two approaches to consider:
Prepare your data so that never has a divide by zero situation, by explicitly coding a “no data” value and testing for that.
Wrap each division that might result in an error with a try
/except
pair, as described at https://wiki.python.org/moin/HandlingExceptions (which has a divide by zero example to use)
(x,y) = (5,0)
try:
z = x/y
except ZeroDivisionError:
print "divide by zero"
I worry about the situation where your data includes a zero that’s really a zero (and not a missing value).
df['one'].divide(df['two'])
Code:
import pandas as pd
import numpy as np
df = pd.DataFrame(np.random.rand(5,2), columns=list('ab'))
df.loc[[1,3], 'b'] = 0
print(df)
print(df['a'].divide(df['b']))
Result:
a b
0 0.517925 0.305973
1 0.900899 0.000000
2 0.414219 0.781512
3 0.516072 0.000000
4 0.841636 0.166157
0 1.692717
1 inf
2 0.530023
3 inf
4 5.065297
dtype: float64
It would probably be more useful to use a dataframe that actually has zero in the denominator (see the last row of column two
).
one two three four five
a 0.469112 -0.282863 -1.509059 bar True
b 0.932424 1.224234 7.823421 bar False
c -1.135632 1.212112 -0.173215 bar False
d 0.232424 2.342112 0.982342 unbar True
e 0.119209 -1.044236 -0.861849 bar True
f -2.104569 0.000000 1.071804 bar False
>>> df.one / df.two
a -1.658442
b 0.761639
c -0.936904
d 0.099237
e -0.114159
f -inf # <<< Note division by zero
dtype: float64
When one of the values is zero, you should get inf
or -inf
in the result. One way to convert these values is as follows:
df['result'] = df.one.div(df.two)
df.loc[~np.isfinite(df['result']), 'result'] = np.nan # Or = 0 per part a) of question.
# or df.loc[np.isinf(df['result']), ...
>>> df
one two three four five result
a 0.469112 -0.282863 -1.509059 bar True -1.658442
b 0.932424 1.224234 7.823421 bar False 0.761639
c -1.135632 1.212112 -0.173215 bar False -0.936904
d 0.232424 2.342112 0.982342 unbar True 0.099237
e 0.119209 -1.044236 -0.861849 bar True -0.114159
f -2.104569 0.000000 1.071804 bar False NaN
You can always use a try statement:
try:
z = var1/var2
except ZeroDivisionError:
print ("0") #As python-3's rule is: Parentheses
OR…
You can also do:
if var1==0:
if var2==0:
print("0")
else:
var3 = var1/var2
Hope this helped! Choose whichever choice you desire (they’re both the same anyways).