How to extend the logic of selecting from a DataFrame based on the first N-1 levels when N > 2?
As an example, consider a DataFrame:
midx = pd.MultiIndex.from_product([[0, 1], [10, 20, 30], ["a", "b"]])
df = pd.DataFrame(1, columns=midx, index=np.arange(3))
In[11]: df
Out[11]:
0 1
10 20 30 10 20 30
a b a b a b a b a b a b
0 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1
2 1 1 1 1 1 1 1 1 1 1 1 1
Here, it is easy to select columns where 0 or 1 are in the first level:
df[[0, 1]]
But the same logic does not extend to selecting columns with 0 or 1 in the first and 10 or 20 in the second level:
In[13]: df[[(0, 10), (0, 20), (1, 10), (1, 20)]]
ValueError: operands could not be broadcast together with shapes (4,2) (3,) (4,2)
The following works:
df.loc[:, pd.IndexSlice[[0, 1], [10, 20], :]]
but is cumbersome, especially when the selector needs to be extracted from another DataFrame with a 2-level MultiIndex:
idx = df.columns.droplevel(2)
In[16]: idx
Out[16]:
MultiIndex(levels=[[0, 1], [10, 20, 30]],
labels=[[0, 0, 0, 0, 0, 0, 1, 1, 1, ... 1, 2, 2]])
In[17]: df[idx]
ValueError: operands could not be broadcast together with shapes (12,2) (3,) (12,2)
EDIT: Ideally, I would also like to be able to order columns this way, not just select them — again, in the spirit of df[[1, 0]] being able to order columns based on the first level.
If possible, you can filter by boolean indexing with get_level_values and isin:
m1 = df.columns.get_level_values(0).isin([0,1])
m2 = df.columns.get_level_values(1).isin([10,20])
print (m1)
[ True True True True True True True True True True True True]
print (m2)
[ True True True True False False True True True True False False]
print (m1 & m2)
[ True True True True False False True True True True False False]
df1 = df.loc[:, m1 & m2]
print (df1)
0 1
10 20 10 20
a b a b a b a b
0 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1
2 1 1 1 1 1 1 1 1
df.columns = df.columns.droplevel(2)
print (df)
0 1
10 10 20 20 30 30 10 10 20 20 30 30
0 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1
2 1 1 1 1 1 1 1 1 1 1 1 1
df2 = df.loc[:, m1 & m2]
print (df2)
0 1
10 10 20 20 10 10 20 20
0 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1
2 1 1 1 1 1 1 1 1
Related
in this dataframe:
Feat1 Feat2 Feat3 Feat4 Labels
-46.220314 22.862856 -6.1573067 5.6060414 2
-23.80669 20.536781 -5.015675 4.2216353 2
-42.092365 25.680704 -5.0092897 5.665794 2
-35.29639 21.709473 -4.160352 5.578346 2
-37.075096 22.347767 -3.860426 5.6953945 2
-42.8849 28.03802 -7.8572545 3.3361 2
-32.3057 26.568039 -9.47018 3.4532788 2
-24.469942 27.005375 -9.301921 4.3995037 2
-97.89892 -0.38156664 6.4163384 7.234347 1
-81.96325 0.1821717 -1.2870358 4.703838 1
-78.41986 -6.766374 0.8001185 0.83444935 1
-100.68544 -4.5810957 1.6977689 1.8801615 1
-87.05412 -2.9231584 6.817379 5.4460077 1
-64.121056 -3.7892206 -0.283514 6.3084154 1
-94.504845 -0.9999217 3.2884297 6.881124 1
-61.951996 -8.960198 -1.5915259 5.6160254 1
-108.19452 13.909201 0.6966458 -1.956591 0
-97.4037 22.897585 -2.8488266 1.4105041 0
-92.641335 22.10624 -3.5110545 2.467166 0
-199.18787 3.3090565 -2.5994794 4.0802555 0
-137.5976 6.795896 1.6793671 2.2256763 0
-208.0035 -1.33229 -3.2078092 1.5177402 0
-108.225975 14.341716 1.02891 -1.8651972 0
-121.29299 18.274035 2.2891548 2.3360753 0
I wanted to sort the rows based on different column values in the "Labels" column.
I am able to sort in ascending such that the labels appear as [0 1 2] via the command
df2 = df1.sort_values(by = 'Labels', ascending = True)
Then ascending = False, where the labels appear [2 1 0].
How then do I go about sorting the labels as [1 0 2]?
Any help will be greatly appreciated!
Here's a way using Categorical:
df['Labels'] = pd.Categorical(df['Labels'],
categories = [1, 0, 2],
ordered=True)
df.sort_values('Labels')
Output:
Feat1 Feat2 Feat3 Feat4 Labels
11 -100.685440 -4.581096 1.697769 1.880162 1
15 -61.951996 -8.960198 -1.591526 5.616025 1
8 -97.898920 -0.381567 6.416338 7.234347 1
9 -81.963250 0.182172 -1.287036 4.703838 1
10 -78.419860 -6.766374 0.800118 0.834449 1
14 -94.504845 -0.999922 3.288430 6.881124 1
12 -87.054120 -2.923158 6.817379 5.446008 1
13 -64.121056 -3.789221 -0.283514 6.308415 1
21 -208.003500 -1.332290 -3.207809 1.517740 0
20 -137.597600 6.795896 1.679367 2.225676 0
19 -199.187870 3.309057 -2.599479 4.080255 0
18 -92.641335 22.106240 -3.511055 2.467166 0
17 -97.403700 22.897585 -2.848827 1.410504 0
16 -108.194520 13.909201 0.696646 -1.956591 0
23 -121.292990 18.274035 2.289155 2.336075 0
22 -108.225975 14.341716 1.028910 -1.865197 0
7 -24.469942 27.005375 -9.301921 4.399504 2
6 -32.305700 26.568039 -9.470180 3.453279 2
5 -42.884900 28.038020 -7.857254 3.336100 2
4 -37.075096 22.347767 -3.860426 5.695394 2
3 -35.296390 21.709473 -4.160352 5.578346 2
2 -42.092365 25.680704 -5.009290 5.665794 2
1 -23.806690 20.536781 -5.015675 4.221635 2
0 -46.220314 22.862856 -6.157307 5.606041 2
You can use an ordered Categorical, or if you don't want to change the DataFrame, the poor-man's variant, a mapping Series:
order = [1, 0, 2]
key = pd.Series({k:v for v,k in enumerate(order)}).get
# or
# pd.Series(range(len(order)), index=order).get
df1.sort_values(by='Labels', key=key)
Example:
df1 = pd.DataFrame({'Labels': [1,0,1,2,0,2,1]})
order = [1, 0, 2]
key = pd.Series({k:v for v,k in enumerate(order)}).get
print(df1.sort_values(by='Labels', key=key))
Labels
0 1
2 1
6 1
1 0
4 0
3 2
5 2
here is another way to do it
create a new column using map and map the new order sequence and then sort as usual
df['sort_label'] = df['Labels'].map({1:0, 0:1, 2:2 }) #).sort_values('sort_label', ascending=False)
df.sort_values('sort_label')
Feat1 Feat2 Feat3 Feat4 Labels sort_label
11 -100.685440 -4.581096 1.697769 1.880162 1 0
15 -61.951996 -8.960198 -1.591526 5.616025 1 0
8 -97.898920 -0.381567 6.416338 7.234347 1 0
9 -81.963250 0.182172 -1.287036 4.703838 1 0
10 -78.419860 -6.766374 0.800119 0.834449 1 0
14 -94.504845 -0.999922 3.288430 6.881124 1 0
12 -87.054120 -2.923158 6.817379 5.446008 1 0
13 -64.121056 -3.789221 -0.283514 6.308415 1 0
21 -208.003500 -1.332290 -3.207809 1.517740 0 1
20 -137.597600 6.795896 1.679367 2.225676 0 1
19 -199.187870 3.309057 -2.599479 4.080255 0 1
18 -92.641335 22.106240 -3.511054 2.467166 0 1
17 -97.403700 22.897585 -2.848827 1.410504 0 1
16 -108.194520 13.909201 0.696646 -1.956591 0 1
23 -121.292990 18.274035 2.289155 2.336075 0 1
22 -108.225975 14.341716 1.028910 -1.865197 0 1
7 -24.469942 27.005375 -9.301921 4.399504 2 2
6 -32.305700 26.568039 -9.470180 3.453279 2 2
5 -42.884900 28.038020 -7.857254 3.336100 2 2
4 -37.075096 22.347767 -3.860426 5.695394 2 2
3 -35.296390 21.709473 -4.160352 5.578346 2 2
2 -42.092365 25.680704 -5.009290 5.665794 2 2
1 -23.806690 20.536781 -5.015675 4.221635 2 2
0 -46.220314 22.862856 -6.157307 5.606041 2 2
I've got 4 columns with numeric values between 1 and 4, and I'm trying to see which rows change from a value of 1 to a value of 4 progressing from column a to column d within those 4 columns. Currently I'm pulling the difference between each of the columns and looking for a value of 3. Is there a better way to do this?
Here's what I'm looking for (with 0's in place of nan):
ID a b c d check
1 1 0 1 4 True
2 1 0 1 1 False
3 1 1 1 4 True
4 1 3 3 4 True
5 0 0 1 4 True
6 1 2 3 3 False
7 1 0 0 4 True
8 1 4 4 4 True
9 1 4 3 4 True
10 1 4 1 1 True
You can just do cummax
col = ['a','b','c','d']
s = df[col].cummax(1)
df['new'] = s[col[:3]].eq(1).any(1) & s[col[-1]].eq(4)
Out[523]:
0 True
1 False
2 True
3 True
4 True
5 False
6 True
7 True
8 True
dtype: bool
You can try compare the index of 4 and 1 in apply
cols = ['a', 'b', 'c', 'd']
def get_index(lst, num):
return lst.index(num) if num in lst else -1
df['Check'] = df[cols].apply(lambda row: get_index(row.tolist(), 4) > get_index(row.tolist(), 1), axis=1)
print(df)
ID a b c d check Check
0 1 1 0 1 4 True True
1 2 1 0 1 1 False False
2 3 1 1 1 4 True True
3 4 1 3 3 4 True True
4 5 0 0 1 4 True True
5 6 1 2 3 3 False False
6 7 1 0 0 4 True True
7 8 1 4 4 4 True True
8 9 1 4 3 4 True True
I want to add 1 in column values if column value is greater than 2
here is my dataframe
df=pd.DataFrame({'A':[1,1,1,1,1,1,3,2,2,2,2,2,2],'flag':[1,1,0,1,1,1,5,1,1,0,1,1,1]})
df_out
df=pd.DataFrame({'A':[1,1,1,1,1,1,3,2,2,2,2,2,2],'flag':[1,1,0,1,1,1,6,1,1,0,1,1,1]})
Use DataFrame.loc with add 1:
df.loc[df.A.gt(2), 'flag'] += 1
print (df)
A flag
0 1 1
1 1 1
2 1 0
3 1 1
4 1 1
5 1 1
6 3 6
7 2 1
8 2 1
9 2 0
10 2 1
11 2 1
12 2 1
Or:
df['flag'] = np.where(df.A.gt(2), df['flag'] + 1, df['flag'])
EDIT:
mean = df.groupby(pd.cut(df['x'], bins))['y'].transform('mean')
df['flag'] = np.where(mean.gt(2), df['y'] + 1, df['y'])
And then:
x= df.groupby(pd.cut(df['x'], bins))['y'].apply(lambda x:abs(x-np.mean(x)))
I need some help with comparing two pandas dataframe
I have two dataframes
The first dataframe is
df1 =
a b c d
0 1 1 1 1
1 0 1 0 1
2 0 0 0 1
3 1 1 1 1
4 1 0 1 0
5 1 1 1 0
6 0 0 1 0
7 0 1 0 1
and the second dataframe is
df2 =
a b c d
0 1 1 1 1
1 1 0 1 0
2 0 0 1 0
I want to find the row index of dataframe 1 (df1) which the entire row is the same as the rows in dataframe 2 (df2). My expect result would be
0
3
4
6
The order of the above index does not need to be in order, all I want is the index of dataframe 1 (df1)
Is there a way without using for loop?
Thanks
Tommy
You can using merge
df1.merge(df2,indicator=True,how='left').loc[lambda x : x['_merge']=='both'].index
Out[459]: Int64Index([0, 3, 4, 6], dtype='int64')
In the below example I only want to retain the row 1 and 2
I want to delete all the rows which has 0 anywhere across the column:
kt b tt mky depth
1 1 1 1 1 4
2 2 2 2 2 2
3 3 3 0 3 3
4 0 4 0 0 0
5 5 5 5 5 0
the output should read like below:
kt b tt mky depth
1 1 1 1 1 4
2 2 2 2 2 2
I have tried:
df.loc[(df!=0).any(axis=1)]
But it deletes the row only if all of its corresponding columns are 0
You are really close, need DataFrame.all for check all Trues per row:
df = df.loc[(df!=0).all(axis=1)]
print (df)
kt b tt mky depth
1 1 1 1 1 4
2 2 2 2 2 2
Details:
print (df!=0)
kt b tt mky depth
1 True True True True True
2 True True True True True
3 True True False True True
4 False True False False False
5 True True True True False
print ((df!=0).all(axis=1))
1 True
2 True
3 False
4 False
5 False
dtype: bool
Alternative solution with any for check at least one True for row with changed mask df == 0 and inversing by ~:
df = df.loc[~(df==0).any(axis=1)]