I have a csv file. I read it:
import pandas as pd data = pd.read_csv('my_data.csv', sep=',') data.head() It has output like:
id city department sms category 01 khi revenue NaN 0 02 lhr revenue good 1 03 lhr revenue NaN 0 I want to remove all the rows where sms column is empty/NaN. What is efficient way to do it?
Use dropna with parameter subset for specify column for check NaNs:
data = data.dropna(subset=['sms']) print (data) id city department sms category 1 2 lhr revenue good 1 Another solution with boolean indexing and notnull:
data = data[data['sms'].notnull()] print (data) id city department sms category 1 2 lhr revenue good 1 Alternative with query:
print (data.query("sms == sms")) id city department sms category 1 2 lhr revenue good 1 Timings
#[300000 rows x 5 columns] data = pd.concat([data]*100000).reset_index(drop=True) In [123]: %timeit (data.dropna(subset=['sms'])) 100 loops, best of 3: 19.5 ms per loop In [124]: %timeit (data[data['sms'].notnull()]) 100 loops, best of 3: 13.8 ms per loop In [125]: %timeit (data.query("sms == sms")) 10 loops, best of 3: 23.6 ms per loop 
You can use the method dropna for this:
data.dropna(axis=0, subset=('sms', )) See the documentation for more details on the parameters.
Of course there are multiple ways to do this, and there are some slight performance differences. Unless performance is critical, I would prefer the use of dropna() as it is the most expressive.
import pandas as pd import numpy as np i = 10000000 # generate dataframe with a few columns df = pd.DataFrame(dict( a_number=np.random.randint(0,1e6,size=i), with_nans=np.random.choice([np.nan, 'good', 'bad', 'ok'], size=i), letter=np.random.choice(list('abcdefghijklmnop'), size=i)) ) # using notebook %%timeit a = df.dropna(subset=['with_nans']) #1.29 s ± 112 ms per loop (mean ± std. dev. of 7 runs, 1 loop each) # using notebook %%timeit b = df[~df.with_nans.isnull()] #890 ms ± 59.8 ms per loop (mean ± std. dev. of 7 runs, 1 loop each) # using notebook %%timeit c = df.query('with_nans == with_nans') #1.71 s ± 100 ms per loop (mean ± std. dev. of 7 runs, 1 loop each) 