pandas - take last N rows from one subgroup - python

Let's suppose we have a dataframe that be generated using this code:
import pandas as pd
d = {'p1': np.random.rand(32),
'a1': np.random.rand(32),
'phase': [0,0,0,0, 1,1,1,1, 2,2,2,2, 3,3,3,3, 0,0,0,0, 1,1,1,1, 2,2,2,2, 3,3,3,3],
'file_number': [1,1,1,1, 1,1,1,1, 1,1,1,1, 1,1,1,1, 2,2,2,2, 2,2,2,2, 2,2,2,2, 2,2,2,2]
}
df = pd.DataFrame(d)
For each file number i want to take only last N rows of phase number 3. So that the result for N==2 looks like this:
Currently I'm doing it in this way:
def phase_3_last_n_observations(df, n):
result = []
for fn in df['file_number'].unique():
file_df = df[df['file_number']==fn]
for phase in [0,1,2,3]:
phase_df = file_df[file_df['phase']==phase]
if phase == 3:
phase_df = phase_df[-n:]
result.append(phase_df)
df = pd.concat(result, axis=0)
return df
phase_3_last_n_observations(df, 2)
However, it is very slow and I have terabytes of data, so I need to worry about performance. Does anyone have any idea how to speed my solution up? Thanks!


Filter the rows where phase is 3 then groupby and use tail to select the last two rows per file_number, finally append to get the result
m = df['phase'].eq(3)
df[~m].append(df[m].groupby('file_number').tail(2)).sort_index()
p1 a1 phase file_number
0 0.223906 0.164288 0 1
1 0.214081 0.748598 0 1
2 0.567702 0.226143 0 1
3 0.695458 0.567288 0 1
4 0.760710 0.127880 1 1
5 0.592913 0.397473 1 1
6 0.721191 0.572320 1 1
7 0.047981 0.153484 1 1
8 0.598202 0.203754 2 1
9 0.296797 0.614071 2 1
10 0.961616 0.105837 2 1
11 0.237614 0.640263 2 1
14 0.500415 0.220355 3 1
15 0.968630 0.351404 3 1
16 0.065283 0.595144 0 2
17 0.308802 0.164214 0 2
18 0.668811 0.826478 0 2
19 0.888497 0.186267 0 2
20 0.199129 0.241900 1 2
21 0.345185 0.220940 1 2
22 0.389895 0.761068 1 2
23 0.343100 0.582458 1 2
24 0.182792 0.245551 2 2
25 0.503181 0.894517 2 2
26 0.144294 0.351350 2 2
27 0.157116 0.847499 2 2
30 0.194274 0.143037 3 2
31 0.542183 0.060485 3 2


I use idea from deleted answer - get indices by previous rows for rows matching 3 by GroupBy.cumcount and remove them by DataFrame.drop:
def phase_3_last_n_observations(df, N):
df1 = df[df['phase'].eq(3)]
idx = df1[df1.groupby('file_number').cumcount(ascending=False).ge(N)].index
return df.drop(idx)
#index is reseted for default, because used for remove rows
df = phase_3_last_n_observations(df.reset_index(drop=True), 2)


As an alternative solution to what already exists: You can calculate the last elements for all phase groups and afterwards just use .loc to get the needed group result. I have written the code for N==2, if you want for N==3, then use [-1, -2, -3]
result = df.groupby(['phase']).nth([-1, -2])
PHASE = 3
result.loc[PHASE]

Related

Allocate lowest value over n rows to n rows in DataFrame

I need to take the lowest value over n rows and add it to these n rows in a new colomn of the dataframe. For example:
n=3
Column 1 Column 2
5 3
3 3
4 3
7 2
8 2
2 2
5 4
4 4
9 4
8 2
2 2
3 2
5 2
Please take note that if the number of rows is not dividable by n, the last values are incorporated in the last group. So in this example n=4 for the end of the dataframe.
Thanking you in advance!

I do not know any straight forward way to do this, but here is a working example (not elegant, but working...).
If you do not worry about the number of rows being dividable by n, you could use .groupby():
import pandas as pd
d = {'col1': [1, 2,1,5,3,2,5,6,4,1,2] }
df = pd.DataFrame(data=d)
n=3
df['new_col']=df.groupby(df.index // n).transform('min')
which yields:
col1 new_col
0 1 1
1 2 1
2 1 1
3 5 2
4 3 2
5 2 2
6 5 4
7 6 4
8 4 4
9 1 1
10 2 1
However, we can see that the last 2 rows are grouped together, instead of them being grouped with the 3 previous values in this case.
A way around would be to look at the .count() of elements in each group generated by grouby, and check the last one:
import pandas as pd
d = {'col1': [1, 2,1,5,3,2,5,6,4,1,2] }
df = pd.DataFrame(data=d)
n=3
# Temporary dataframe
A = df.groupby(df.index // n).transform('min')
# The min value of each group in a second dataframe
min_df = df.groupby(df.index // n).min()
# The size of the last group
last_batch = df.groupby(df.index // n).count()[-1:]
# if the last size is not equal to n
if last_batch.values[0][0] !=n:
last_group = last_batch+n
A[-last_group.values[0][0]:]=min_df[-2:].min()
# Assign the temporary modified dataframe to df
df['new_col'] = A
which yields the expected result:
col1 new_col
0 1 1
1 2 1
2 1 1
3 5 2
4 3 2
5 2 2
6 5 1
7 6 1
8 4 1
9 1 1
10 2 1

Count number of consecutive rows that are greater than current row value but less than the value from other column

Say I have the following sample dataframe (there are about 25k rows in the real dataframe)
df = pd.DataFrame({'A' : [0,3,2,9,1,0,4,7,3,2], 'B': [9,8,3,5,5,5,5,8,0,4]})
df
A B
0 0 9
1 3 8
2 2 3
3 9 5
4 1 5
5 0 5
6 4 5
7 7 8
8 3 0
9 2 4
For the column A I need to know how many next and previous rows are greater than current row value but less than value in column B.
So my expected output is :
A B next count previous count
0 9 2 0
3 8 0 0
2 3 0 1
9 5 0 0
1 5 0 0
0 5 2 1
4 5 1 0
7 8 0 0
3 0 0 2
2 4 0 0
Explanation :
First row is calculated as : since 3 and 2 are greater than 0 but less than corresponding B value 8 and 3
Second row is calculated as : since next value 2 is not greater than 3
Third row is calculated as : since 9 is greater than 2 but not greater than its corresponding B value
Similarly, previous count is calculated
Note : I know how to solve this problem by looping using list comprehension or using the pandas apply method but still I won't mind a clear and concise apply approach. I was looking for a more pandaic approach.
My Solution
Here is the apply solution, which I think is inefficient. Also, as people said that there might be no vector solution for the question. So as mentioned, a more efficient apply solution will be accepted for this question.
This is what I have tried.
This function gets the number of previous/next rows that satisfy the condition.
def get_prev_next_count(row):
next_nrow = df.loc[row['index']+1:,['A', 'B']]
prev_nrow = df.loc[:row['index']-1,['A', 'B']][::-1]
if (next_nrow.size == 0):
return 0, ((prev_nrow.A > row.A) & (prev_nrow.A < prev_nrow.B)).argmin()
if (prev_nrow.size == 0):
return ((next_nrow.A > row.A) & (next_nrow.A < next_nrow.B)).argmin(), 0
return (((next_nrow.A > row.A) & (next_nrow.A < next_nrow.B)).argmin(), ((prev_nrow.A > row.A) & (prev_nrow.A < prev_nrow.B)).argmin())
Generating output :
df[['next count', 'previous count']] = df.reset_index().apply(get_prev_next_count, axis=1, result_type="expand")
Output :
This gives us the expected output
df
A B next count previous count
0 0 9 2 0
1 3 8 0 0
2 2 3 0 1
3 9 5 0 0
4 1 5 0 0
5 0 5 2 1
6 4 5 1 0
7 7 8 0 0
8 3 0 0 2
9 2 4 0 0

I made some optimizations:
You don't need to reset_index() you can access the index with .name
If you only pass df[['A']] instead of the whole frame, that may help.
prev_nrow.empty is the same as (prev_nrow.size == 0)
Applied different logic to get the desired value via first_false, this speeds things up significantly.
def first_false(val1, val2, A):
i = 0
for x, y in zip(val1, val2):
if A < x < y:
i += 1
else:
break
return i
def get_prev_next_count(row):
A = row['A']
next_nrow = df.loc[row.name+1:,['A', 'B']]
prev_nrow = df2.loc[row.name-1:,['A', 'B']]
if next_nrow.empty:
return 0, first_false(prev_nrow.A, prev_nrow.B, A)
if prev_nrow.empty:
return first_false(next_nrow.A, next_nrow.B, A), 0
return (first_false(next_nrow.A, next_nrow.B, A),
first_false(prev_nrow.A, prev_nrow.B, A))
df2 = df[::-1].copy() # Shave a tiny bit of time by only reversing it once~
df[['next count', 'previous count']] = df[['A']].apply(get_prev_next_count, axis=1, result_type='expand')
print(df)
Output:
A B next count previous count
0 0 9 2 0
1 3 8 0 0
2 2 3 0 1
3 9 5 0 0
4 1 5 0 0
5 0 5 2 1
6 4 5 1 0
7 7 8 0 0
8 3 0 0 2
9 2 4 0 0
Timing
Expanding the data:
df = pd.concat([df]*(10000//4), ignore_index=True)
# df.shape == (25000, 2)
Original Method:
Gave up at 15 minutes.
New Method:
1m 20sec
Throw pandarallel at it:
from pandarallel import pandarallel
pandarallel.initialize()
df[['A']].parallel_apply(get_prev_next_count, axis=1, result_type='expand')
26sec

Multiply two columns by following a pattern using pandas

I would like to multiply two columns of a df by following a specific pattern without using a loop. I have this df :
num m d
0 8 5
1 2 3
2 17 8
The idea is to multply for each row in 'm' every row in 'd' except the one with the same 'num'. The resulting df would be :
num1 num2 mult
0 1 8x3 = 24
0 2 8x8 = 64
1 2 2x8 = 16
Is there a way to do that ?
Thank for your help

You can try:
df = df.set_index('num')
((df[['m']].rename(columns={'m':'d'}) # df[['d']].T)
.rename_axis('num2', axis=1)
.stack().reset_index(name='mult')
)
Or use broadcasting:
(pd.DataFrame(df['m'].values * df['d'].values[:,None],
index=df['num'],
columns=df['num'].rename('num2'))
.stack().reset_index(name='mult')
)
num num2 mult
0 0 0 40
1 0 1 24
2 0 2 64
3 1 0 10
4 1 1 6
5 1 2 16
6 2 0 85
7 2 1 51
8 2 2 136

You could use the following:
product = df1['m'][df2['num1']].values*df1['d'][df2['num2']].values
df2['mult'] = pd.Series(product, index=df2.index)

I'd recommend first creating a frame with all possible permutations of the 2 columns, then filtering out the rows which don't correspond to the required pattern.
Something like this
df = df.set_index('num')
((df[['m']].rename(columns={'m':'d'}) # df[['d']].T)
.rename_axis('num2', axis=1)
.stack().reset_index(name='mult')
)
df[df['num']!=df['num2']]

Iterate over a groupby dataframe to operate in each row

I have a DataFrame like this:
subject trial attended
0 1 1 1
1 1 3 0
2 1 4 1
3 1 7 0
4 1 8 1
5 2 1 1
6 2 2 1
7 2 6 1
8 2 8 0
9 2 9 1
10 2 11 1
11 2 12 1
12 2 13 1
13 2 14 1
14 2 15 1
I would like to GroupBy subject.
Then iterate in each row of the GroupBy dataframe.
If for a row 'attended' == 1, then to increase a variable sum_reactive by 1.
If the sum_reactive variable reaches == 4, then to add in a dictionary the 'subject' and 'trial' in which the variable sum_reactive reached a value of 4.
I as trying to define a function for this, but it doesn't work:
def count_attended():
sum_reactive = 0
dict_attended = {}
for i, g in reactive.groupby(['subject']):
for row in g:
if g['attended'][row] == 1:
sum_reactive += 1
if sum_reactive == 4:
dict_attended.update({g['subject'] : g['trial'][row]})
return dict_attended
return dict_attended
I think that I don't have clear how to iterate inside each GroupBy dataframe. I'm quite new using pandas.

IIUC try,
df = df.query('attended == 1')
df.loc[df.groupby('subject')['attended'].cumsum() == 4, ['subject', 'trial']].to_dict(orient='record')
Output:
[{'subject': 2, 'trial': 9}]
Using groupby with cumsum will do the counting attended, then check to see when this value equals to 4 to create a boolean series. You can use this boolean series to do boolean indexing to filter your dataframe to certain rows. Lastly, with lock and column filtering select subject and trial.

Keeping subset of row labels in pandas DataFrame based on second index

Given a DataFrame with a hierarchical index containing three levels (experiment, trial, slot) and a second DataFrame with a hierarchical index containing two levels (experiment, trial), how do I drop all the rows in the first DataFrame whose (experiment, trial) are not contained in the second dataframe?
Example data:
from io import StringIO
import pandas as pd
df1_data = StringIO(u',experiment,trial,slot,token\n0,btn144a10_p_RDT,0,0,4.0\n1,btn144a10_p_RDT,0,1,14.0\n2,btn144a10_p_RDT,1,0,12.0\n3,btn144a10_p_RDT,1,1,14.0\n4,btn145a07_p_RDT,0,0,6.0\n5,btn145a07_p_RDT,0,1,19.0\n6,btn145a07_p_RDT,1,0,17.0\n7,btn145a07_p_RDT,1,1,13.0\n8,chn004b06_p_RDT,0,0,6.0\n9,chn004b06_p_RDT,0,1,8.0\n10,chn004b06_p_RDT,1,0,2.0\n11,chn004b06_p_RDT,1,1,5.0\n12,chn008a06_p_RDT,0,0,12.0\n13,chn008a06_p_RDT,0,1,14.0\n14,chn008a06_p_RDT,1,0,6.0\n15,chn008a06_p_RDT,1,1,4.0\n16,chn008b06_p_RDT,0,0,3.0\n17,chn008b06_p_RDT,0,1,13.0\n18,chn008b06_p_RDT,1,0,12.0\n19,chn008b06_p_RDT,1,1,19.0\n20,chn008c04_p_RDT,0,0,17.0\n21,chn008c04_p_RDT,0,1,2.0\n22,chn008c04_p_RDT,1,0,1.0\n23,chn008c04_p_RDT,1,1,6.0\n')
df1 = pd.DataFrame.from_csv(df1_data).set_index(['experiment', 'trial', 'slot'])
df2_data = StringIO(u',experiment,trial,target\n0,btn145a07_p_RDT,1,13\n1,chn004b06_p_RDT,1,9\n2,chn008a06_p_RDT,0,15\n3,chn008a06_p_RDT,1,15\n4,chn008b06_p_RDT,1,1\n5,chn008c04_p_RDT,1,12\n')
df2 = pd.DataFrame.from_csv(df2_data).set_index(['experiment', 'trial'])
The first dataframe looks like:
token
experiment trial slot
btn144a10_p_RDT 0 0 4
1 14
1 0 12
1 14
btn145a07_p_RDT 0 0 6
1 19
1 0 17
1 13
chn004b06_p_RDT 0 0 6
1 8
1 0 2
1 5
chn008a06_p_RDT 0 0 12
1 14
1 0 6
1 4
chn008b06_p_RDT 0 0 3
1 13
1 0 12
1 19
chn008c04_p_RDT 0 0 17
1 2
1 0 1
1 6
The second dataframe looks like:
target
experiment trial
btn145a07_p_RDT 1 13
chn004b06_p_RDT 1 9
chn008a06_p_RDT 0 15
1 15
chn008b06_p_RDT 1 1
chn008c04_p_RDT 1 12
The result I want:
token
experiment trial slot
btn145a07_p_RDT 1 0 17
1 13
chn004b06_p_RDT 1 0 2
1 5
chn008a06_p_RDT 0 0 12
1 14
1 0 6
1 4
chn008b06_p_RDT 1 0 12
1 19
chn008c04_p_RDT 1 0 1
1 6

One way to do it would by using merge
merged = pd.merge(
df2.reset_index(),
df1.reset_index(),
left_on=['experiment', 'trial'],
right_on=['experiment', 'trial'],
how='left')
You just need to reindex merged to whatever you like (I could not tell exactly from the question).

What should work is
df1.loc[df2.index]
but multi indexing still has some problems. What does work is
df1.reset_index(2).loc[df2.index].set_index('slot', append=True)
which is a bit of a hack around this problem. Note that
df1.loc[df2.index[:1]]
gives garbage while
df.loc[df2.index[0]]
gives what you would expect. So passing multiple values from a m-level index to an n-level index where n > m > 2 doesn't work, though it should.

Categories