I'm trying to filter a pandas DataFrame and I'm getting different results using a test case and the real data. Using real data I'm getting NaN values, while on the test case I'm getting what I expect.
Test case:
The test case I created has following code:
import pandas as pd
df1 = pd.DataFrame([
["2014-08-06 12:10:00", 19.85, 299.96, 17.5, 228.5, 19.63, 571.43],
["2014-08-06 12:20:00", 19.85, 311.55, 17.85, 248.68, 19.78, 547.21],
["2014-08-06 12:30:00", 20.06, 355.27, 18.35, 224.82, 19.99, 410.68],
["2014-08-06 12:40:00", 20.14, 405.95, 18.49, 247.33, 20.5, 552.79],
["2014-08-06 12:50:00", 20.14, 352.87, 18.7, 449.33, 20.86, 616.44],
["2014-08-06 13:00:00", 20.28, 356.96, 18.92, 307.57, 21.15, 471.18]],
columns=["date_time","t1", "1", "t4", "4", "t6", "6"])
df1 = df1.set_index(["date_time"])
df1 = pd.to_datetime(df1)
filter1 = pd.DataFrame(["2014-08-06 12:20:00","2014-08-06 13:00:00"])
df1_filtered = df1.ix[filter1[filter1.columns[0]][0:2]]
As you may expect, the result is:
>>> df1_filtered
t1 1 t4 4 t6 6
2014-08-06 12:20:00 19.85 311.55 17.85 248.68 19.78 547.21
2014-08-06 13:00:00 20.28 356.96 18.92 307.57 21.15 471.18
Using real data:
Real data comes from a txt file and looks like this:
Fecha_hora t1 1 t4 4 t6 6
2014-08-06 12:10:00 19.85 299.96 17.5 228.5 19.63 571.43
2014-08-06 12:20:00 19.85 311.55 17.85 248.68 19.78 547.21
2014-08-06 12:30:00 20.06 355.27 18.35 224.82 19.99 410.68
2014-08-06 12:40:00 20.14 405.95 18.49 247.33 20.5 552.79
2014-08-06 12:50:00 20.14 352.87 18.7 449.33 20.86 616.44
2014-08-06 13:00:00 20.28 356.96 18.92 307.57 21.15 471.18
However when I read the real data, and use same filter as before this way:
df2 = pd.read_csv(r"D:/tmp/data.txt", sep='\t', parse_dates=True, index_col=0)
df2_filtered = df2.ix[filter1[filter1.columns[0]][0:2]]
I get following results with values as NaN:
>>> df2_filtered
t1 1 t4 4 t6 6
2014-08-06 12:20:00 NaN NaN NaN NaN NaN NaN
2014-08-06 13:00:00 NaN NaN NaN NaN NaN NaN
But I can still get the values from a certain row like this:
>>> df2.ix["2014-08-06 12:20:00"]
t1 19.85
1 311.55
t4 17.85
4 248.68
t6 19.78
6 547.21
Name: 2014-08-06 12:20:00
Question:
How can I filter my real data in order to get same results as in my test case? May there be a better way to achieve what I'm looking for?
Note: My pandas version is 0.9.0 used under python 2.5. Means I have no loc function.
Note 2: I even tried this using python 2.7 under pythonanywhere.com with same different results. However if I check for df1==df2 I get True for every single value.
Hopefully goes without saying, but if at all possible, upgrade your python/pandas!
In this case, on a recent version (0.20.3) I get missing values in both cases - I need to convert the lookup keys to datetimes and I'm guessing it will work for you too.
The convenience string based date indexing only works with scalars / slices.
In [174]: lookup = pd.to_datetime(filter1[filter1.columns[0]][0:2])
In [175]: df2.ix[lookup]
Out[175]:
t1 1 t4 4 t6 6
Fecha_hora
2014-08-06 12:20:00 19.85 311.55 17.85 248.68 19.78 547.21
2014-08-06 13:00:00 20.28 356.96 18.92 307.57 21.15 471.18