How to avoid a chained apply when selecting from multiindex?

Question

I have a dataframe like this:

import numpy as np
import pandas as pd


df = pd.DataFrame({
    'ind1': list('AAABBBCCC'),
    'ind2': list(map(int, list('123123123'))),
    'val1': [0, 1, 2, -1, -4, -5, 10, 11, 4],
    'val2': [0.1, 0.2, -0.2, 0.1, 0.2, 0.2, -0.1, 2, 0.1]
})

df = df.set_index(['ind1', 'ind2'])

           val1  val2
ind1 ind2            
A    1        0   0.1
     2        1   0.2
     3        2  -0.2
B    1       -1   0.1
     2       -4   0.2
     3       -5   0.2
C    1       10  -0.1
     2       11   2.0
     3        4   0.1

I want to select all entries for which the absolute value of differences between the values in val1 are increasing.

I currently do it as follows:

m_incr = (
    df.groupby('ind1')['val1']
      .apply(lambda x: np.diff(abs(x)))
      .apply(lambda x: all(eli > 0 for eli in x))
)

df_incr = df[m_incr[df.index.get_level_values('ind1')].values]

which gives me the desired outcome:

           val1  val2
ind1 ind2            
A    1        0   0.1
     2        1   0.2
     3        2  -0.2
B    1       -1   0.1
     2       -4   0.2
     3       -5   0.2

My question is whether there is a more straightforward/efficient way that avoids the chained applys.

Answer 1

Use GroupBy.transform for return Series with same size like original DataFrame:

mask = df.groupby('ind1')['val1'].transform(lambda x: (np.diff(abs(x)) > 0).all())

And then filter by mask with boolean indexing:

print (df[mask])

All together:

print (df[df.groupby('ind1')['val1'].transform(lambda x: (np.diff(abs(x)) > 0).all())])

---

           val1  val2
ind1 ind2            
A    1        0   0.1
     2        1   0.2
     3        2  -0.2
B    1       -1   0.1
     2       -4   0.2
     3       -5   0.2

Detail:

print (mask)
ind1  ind2
A     1        True
      2        True
      3        True
B     1        True
      2        True
      3        True
C     1       False
      2       False
      3       False
Name: val1, dtype: bool