I need to filter outliers in a dataset. Replacing the outlier with the previous value in the column makes the most sense in my application.
I was having considerable difficulty doing this with the pandas tools available (mostly to do with copies on slices, or type conversions occurring when setting to NaN).
Is there a fast and/or memory efficient way to do this? (Please see my answer below for the solution I am currently using, which also has limitations.)
A simple example:
>>> import pandas as pd
>>> df = pd.DataFrame({'A':[1,2,3,4,1000,6,7,8],'B':list('abcdefgh')})
>>> df
A B
0 1 a
1 2 b
2 3 c
3 4 d
4 1000 e # '1000 e' --> '4 e'
5 6 f
6 7 g
7 8 h
You can simply mask values over your threshold and use ffill:
df.assign(A=df.A.mask(df.A.gt(10)).ffill())
A B
0 1.0 a
1 2.0 b
2 3.0 c
3 4.0 d
4 4.0 e
5 6.0 f
6 7.0 g
7 8.0 h
Using mask is necessary rather than something like shift, because it guarantees non-outlier output in the case that the previous value is also above a threshold.
I circumvented some of the issues with pandas copies and slices by converting to a numpy array first, doing the operations there, and then re-inserting the column. I'm not certain, but as far as I can tell, the datatype is the same once it is put back into the pandas.DataFrame.
def df_replace_with_previous(df,col,maskfunc,inplace=False):
arr = np.array(df[col])
mask = maskfunc(arr)
arr[ mask ] = arr[ list(mask)[1:]+[False] ]
if inplace:
df[col] = arr
return
else:
df2 = df.copy()
df2[col] = arr
return df2
This creates a mask, shifts it down by one so that the True values point at the previous entry, and updates the array. Of course, this will need to run recursively if there are multiple adjacent outliers (N times if there are N consecutive outliers), which is not ideal.
Usage in the case given in OP:
df_replace_with_previous(df,'A',lambda x:x>10,False)
Related
I am doing some computing on a dataset using loops. Then, based on random event, I am going to compute some float number(This means that I don't know in advance how many floats I am going to retrieve). I want to save these numbers(results) in a some kind of a list and then save them to a dataframe column ( I want to have these results for each iteration in my loop and save them in a column so I can compare them, meaning, each iteration will produce a "list" of results that will be registred in a df column)
example:
for y in range(1,10):
for x in range(1,100):
if(x>random number and x<y):
result=2*x
I want to save all the results in a dataframe columns by combination x,y. For example, the results for x=1,y=2 in a column then x=2,y=2 in column ...etc and the results are not of the same size, so I guess that I'll use fillna.
Now I know that I can create an empty dataframe with max index and then fill it result by result, but I think there's a better way to do it!
Thanks in advance.
You want to take advantage of the efficiency that numpy and pandas give you. If you use numpy.where, you can set the value to nan when the if statement is False, and otherwise you can execute your formula:
import numpy as np
import pandas as pd
np.random.seed(0) # so you can reproduce my result, you can remove this in practice
x = list(range(10))
y = list(range(1, 11))
random_nums = 10 * np.random.random(10)
df = pd.DataFrame({'x' : x, 'y': y})
# the first argument is your if condition
df['new_col'] = np.where((df['x'] > random_nums) & (df['x'] < df['y']), 2*df['x'], np.nan)
print(df)
Here, random_nums generates an entire np.ndarray of random numbers to compare with. This gives
x y new_col
0 0 1 NaN
1 1 2 NaN
2 2 3 NaN
3 3 4 NaN
4 4 5 NaN
5 5 6 NaN
6 6 7 12.0
7 7 8 NaN
8 8 9 NaN
9 9 10 18.0
This is especially faster if your formula (here, 2*x) is relatively quick to compute.
I have a DataFrame with many missing values in columns which I wish to groupby:
import pandas as pd
import numpy as np
df = pd.DataFrame({'a': ['1', '2', '3'], 'b': ['4', np.NaN, '6']})
In [4]: df.groupby('b').groups
Out[4]: {'4': [0], '6': [2]}
see that Pandas has dropped the rows with NaN target values. (I want to include these rows!)
Since I need many such operations (many cols have missing values), and use more complicated functions than just medians (typically random forests), I want to avoid writing too complicated pieces of code.
Any suggestions? Should I write a function for this or is there a simple solution?
pandas >= 1.1
From pandas 1.1 you have better control over this behavior, NA values are now allowed in the grouper using dropna=False:
pd.__version__
# '1.1.0.dev0+2004.g8d10bfb6f'
# Example from the docs
df
a b c
0 1 2.0 3
1 1 NaN 4
2 2 1.0 3
3 1 2.0 2
# without NA (the default)
df.groupby('b').sum()
a c
b
1.0 2 3
2.0 2 5
# with NA
df.groupby('b', dropna=False).sum()
a c
b
1.0 2 3
2.0 2 5
NaN 1 4
This is mentioned in the Missing Data section of the docs:
NA groups in GroupBy are automatically excluded. This behavior is consistent with R
One workaround is to use a placeholder before doing the groupby (e.g. -1):
In [11]: df.fillna(-1)
Out[11]:
a b
0 1 4
1 2 -1
2 3 6
In [12]: df.fillna(-1).groupby('b').sum()
Out[12]:
a
b
-1 2
4 1
6 3
That said, this feels pretty awful hack... perhaps there should be an option to include NaN in groupby (see this github issue - which uses the same placeholder hack).
However, as described in another answer, "from pandas 1.1 you have better control over this behavior, NA values are now allowed in the grouper using dropna=False"
Ancient topic, if someone still stumbles over this--another workaround is to convert via .astype(str) to string before grouping. That will conserve the NaN's.
df = pd.DataFrame({'a': ['1', '2', '3'], 'b': ['4', np.NaN, '6']})
df['b'] = df['b'].astype(str)
df.groupby(['b']).sum()
a
b
4 1
6 3
nan 2
I am not able to add a comment to M. Kiewisch since I do not have enough reputation points (only have 41 but need more than 50 to comment).
Anyway, just want to point out that M. Kiewisch solution does not work as is and may need more tweaking. Consider for example
>>> df = pd.DataFrame({'a': [1, 2, 3, 5], 'b': [4, np.NaN, 6, 4]})
>>> df
a b
0 1 4.0
1 2 NaN
2 3 6.0
3 5 4.0
>>> df.groupby(['b']).sum()
a
b
4.0 6
6.0 3
>>> df.astype(str).groupby(['b']).sum()
a
b
4.0 15
6.0 3
nan 2
which shows that for group b=4.0, the corresponding value is 15 instead of 6. Here it is just concatenating 1 and 5 as strings instead of adding it as numbers.
All answers provided thus far result in potentially dangerous behavior as it is quite possible you select a dummy value that is actually part of the dataset. This is increasingly likely as you create groups with many attributes. Simply put, the approach doesn't always generalize well.
A less hacky solve is to use pd.drop_duplicates() to create a unique index of value combinations each with their own ID, and then group on that id. It is more verbose but does get the job done:
def safe_groupby(df, group_cols, agg_dict):
# set name of group col to unique value
group_id = 'group_id'
while group_id in df.columns:
group_id += 'x'
# get final order of columns
agg_col_order = (group_cols + list(agg_dict.keys()))
# create unique index of grouped values
group_idx = df[group_cols].drop_duplicates()
group_idx[group_id] = np.arange(group_idx.shape[0])
# merge unique index on dataframe
df = df.merge(group_idx, on=group_cols)
# group dataframe on group id and aggregate values
df_agg = df.groupby(group_id, as_index=True)\
.agg(agg_dict)
# merge grouped value index to results of aggregation
df_agg = group_idx.set_index(group_id).join(df_agg)
# rename index
df_agg.index.name = None
# return reordered columns
return df_agg[agg_col_order]
Note that you can now simply do the following:
data_block = [np.tile([None, 'A'], 3),
np.repeat(['B', 'C'], 3),
[1] * (2 * 3)]
col_names = ['col_a', 'col_b', 'value']
test_df = pd.DataFrame(data_block, index=col_names).T
grouped_df = safe_groupby(test_df, ['col_a', 'col_b'],
OrderedDict([('value', 'sum')]))
This will return the successful result without having to worry about overwriting real data that is mistaken as a dummy value.
One small point to Andy Hayden's solution – it doesn't work (anymore?) because np.nan == np.nan yields False, so the replace function doesn't actually do anything.
What worked for me was this:
df['b'] = df['b'].apply(lambda x: x if not np.isnan(x) else -1)
(At least that's the behavior for Pandas 0.19.2. Sorry to add it as a different answer, I do not have enough reputation to comment.)
I answered this already, but some reason the answer was converted to a comment. Nevertheless, this is the most efficient solution:
Not being able to include (and propagate) NaNs in groups is quite aggravating. Citing R is not convincing, as this behavior is not consistent with a lot of other things. Anyway, the dummy hack is also pretty bad. However, the size (includes NaNs) and the count (ignores NaNs) of a group will differ if there are NaNs.
dfgrouped = df.groupby(['b']).a.agg(['sum','size','count'])
dfgrouped['sum'][dfgrouped['size']!=dfgrouped['count']] = None
When these differ, you can set the value back to None for the result of the aggregation function for that group.
Sorry, I know there are so many questions relating to indexing, and it's probably starring me in the face, but I'm having a little trouble with this. I am familiar with .loc, .iloc, and .index methods and slicing in general. The method .reset_index may not have been (and may not be able to be) called on our dataframe and therefore index lables may not be in order. The dataframe and numpy array(s) are actually different length subsets of the dataframe, but for this example I'll keep them the same size (I can handle offsetting once I have an example).
Here is a picture that show's what I'm looking for:
I can pull cols of rows from the dataframe based on some search criteria.
idxlbls = df.index[df['timestamp'] == dt]
stuff = df.loc[idxlbls, 'col3':'col5']
But how do I map that to row number (array indices, not label indices) to be used as an array index in numpy (assuming same row length)?
stuffprime = array[?, ?]
The reason I need it is because the dataframe is much larger and more complete and contains the column searching criteria, but the numpy arrays are subsets that have been extracted and modified prior in the pipeline (and do not have the same searching criteria in them). I need to search the dataframe and pull the equivalent data from the numpy arrays. Basically I need to correlate specific rows from a dataframe to the corresponding rows of a numpy array.
I would map pandas indices to numpy indicies:
keys_dict = dict(zip(idxlbls, range(len(idxlbls))))
Then you may use the dictionary keys_dict to address the array elements by a pandas index: array[keys_dict[some_df_index], :]
I believe need get_indexer for positions by filtered columns names, for index is possible use same way or numpy.where for positions by boolean mask:
df = pd.DataFrame({'timestamp':list('abadef'),
'B':[4,5,4,5,5,4],
'C':[7,8,9,4,2,3],
'D':[1,3,5,7,1,0],
'E':[5,3,6,9,2,4]}, index=list('ABCDEF'))
print (df)
timestamp B C D E
A a 4 7 1 5
B b 5 8 3 3
C a 4 9 5 6
D d 5 4 7 9
E e 5 2 1 2
F f 4 3 0 4
idxlbls = df.index[df['timestamp'] == 'a']
stuff = df.loc[idxlbls, 'C':'E']
print (stuff)
C D E
A 7 1 5
C 9 5 6
a = df.index.get_indexer(stuff.index)
Or get positions by boolean mask:
a = np.where(df['timestamp'] == 'a')[0]
print (a)
[0 2]
b = df.columns.get_indexer(stuff.columns)
print (b)
[2 3 4]
I have a DataFrame with many missing values in columns which I wish to groupby:
import pandas as pd
import numpy as np
df = pd.DataFrame({'a': ['1', '2', '3'], 'b': ['4', np.NaN, '6']})
In [4]: df.groupby('b').groups
Out[4]: {'4': [0], '6': [2]}
see that Pandas has dropped the rows with NaN target values. (I want to include these rows!)
Since I need many such operations (many cols have missing values), and use more complicated functions than just medians (typically random forests), I want to avoid writing too complicated pieces of code.
Any suggestions? Should I write a function for this or is there a simple solution?
pandas >= 1.1
From pandas 1.1 you have better control over this behavior, NA values are now allowed in the grouper using dropna=False:
pd.__version__
# '1.1.0.dev0+2004.g8d10bfb6f'
# Example from the docs
df
a b c
0 1 2.0 3
1 1 NaN 4
2 2 1.0 3
3 1 2.0 2
# without NA (the default)
df.groupby('b').sum()
a c
b
1.0 2 3
2.0 2 5
# with NA
df.groupby('b', dropna=False).sum()
a c
b
1.0 2 3
2.0 2 5
NaN 1 4
This is mentioned in the Missing Data section of the docs:
NA groups in GroupBy are automatically excluded. This behavior is consistent with R
One workaround is to use a placeholder before doing the groupby (e.g. -1):
In [11]: df.fillna(-1)
Out[11]:
a b
0 1 4
1 2 -1
2 3 6
In [12]: df.fillna(-1).groupby('b').sum()
Out[12]:
a
b
-1 2
4 1
6 3
That said, this feels pretty awful hack... perhaps there should be an option to include NaN in groupby (see this github issue - which uses the same placeholder hack).
However, as described in another answer, "from pandas 1.1 you have better control over this behavior, NA values are now allowed in the grouper using dropna=False"
Ancient topic, if someone still stumbles over this--another workaround is to convert via .astype(str) to string before grouping. That will conserve the NaN's.
df = pd.DataFrame({'a': ['1', '2', '3'], 'b': ['4', np.NaN, '6']})
df['b'] = df['b'].astype(str)
df.groupby(['b']).sum()
a
b
4 1
6 3
nan 2
I am not able to add a comment to M. Kiewisch since I do not have enough reputation points (only have 41 but need more than 50 to comment).
Anyway, just want to point out that M. Kiewisch solution does not work as is and may need more tweaking. Consider for example
>>> df = pd.DataFrame({'a': [1, 2, 3, 5], 'b': [4, np.NaN, 6, 4]})
>>> df
a b
0 1 4.0
1 2 NaN
2 3 6.0
3 5 4.0
>>> df.groupby(['b']).sum()
a
b
4.0 6
6.0 3
>>> df.astype(str).groupby(['b']).sum()
a
b
4.0 15
6.0 3
nan 2
which shows that for group b=4.0, the corresponding value is 15 instead of 6. Here it is just concatenating 1 and 5 as strings instead of adding it as numbers.
All answers provided thus far result in potentially dangerous behavior as it is quite possible you select a dummy value that is actually part of the dataset. This is increasingly likely as you create groups with many attributes. Simply put, the approach doesn't always generalize well.
A less hacky solve is to use pd.drop_duplicates() to create a unique index of value combinations each with their own ID, and then group on that id. It is more verbose but does get the job done:
def safe_groupby(df, group_cols, agg_dict):
# set name of group col to unique value
group_id = 'group_id'
while group_id in df.columns:
group_id += 'x'
# get final order of columns
agg_col_order = (group_cols + list(agg_dict.keys()))
# create unique index of grouped values
group_idx = df[group_cols].drop_duplicates()
group_idx[group_id] = np.arange(group_idx.shape[0])
# merge unique index on dataframe
df = df.merge(group_idx, on=group_cols)
# group dataframe on group id and aggregate values
df_agg = df.groupby(group_id, as_index=True)\
.agg(agg_dict)
# merge grouped value index to results of aggregation
df_agg = group_idx.set_index(group_id).join(df_agg)
# rename index
df_agg.index.name = None
# return reordered columns
return df_agg[agg_col_order]
Note that you can now simply do the following:
data_block = [np.tile([None, 'A'], 3),
np.repeat(['B', 'C'], 3),
[1] * (2 * 3)]
col_names = ['col_a', 'col_b', 'value']
test_df = pd.DataFrame(data_block, index=col_names).T
grouped_df = safe_groupby(test_df, ['col_a', 'col_b'],
OrderedDict([('value', 'sum')]))
This will return the successful result without having to worry about overwriting real data that is mistaken as a dummy value.
One small point to Andy Hayden's solution – it doesn't work (anymore?) because np.nan == np.nan yields False, so the replace function doesn't actually do anything.
What worked for me was this:
df['b'] = df['b'].apply(lambda x: x if not np.isnan(x) else -1)
(At least that's the behavior for Pandas 0.19.2. Sorry to add it as a different answer, I do not have enough reputation to comment.)
I answered this already, but some reason the answer was converted to a comment. Nevertheless, this is the most efficient solution:
Not being able to include (and propagate) NaNs in groups is quite aggravating. Citing R is not convincing, as this behavior is not consistent with a lot of other things. Anyway, the dummy hack is also pretty bad. However, the size (includes NaNs) and the count (ignores NaNs) of a group will differ if there are NaNs.
dfgrouped = df.groupby(['b']).a.agg(['sum','size','count'])
dfgrouped['sum'][dfgrouped['size']!=dfgrouped['count']] = None
When these differ, you can set the value back to None for the result of the aggregation function for that group.
I have a DataFrame with many missing values in columns which I wish to groupby:
import pandas as pd
import numpy as np
df = pd.DataFrame({'a': ['1', '2', '3'], 'b': ['4', np.NaN, '6']})
In [4]: df.groupby('b').groups
Out[4]: {'4': [0], '6': [2]}
see that Pandas has dropped the rows with NaN target values. (I want to include these rows!)
Since I need many such operations (many cols have missing values), and use more complicated functions than just medians (typically random forests), I want to avoid writing too complicated pieces of code.
Any suggestions? Should I write a function for this or is there a simple solution?
pandas >= 1.1
From pandas 1.1 you have better control over this behavior, NA values are now allowed in the grouper using dropna=False:
pd.__version__
# '1.1.0.dev0+2004.g8d10bfb6f'
# Example from the docs
df
a b c
0 1 2.0 3
1 1 NaN 4
2 2 1.0 3
3 1 2.0 2
# without NA (the default)
df.groupby('b').sum()
a c
b
1.0 2 3
2.0 2 5
# with NA
df.groupby('b', dropna=False).sum()
a c
b
1.0 2 3
2.0 2 5
NaN 1 4
This is mentioned in the Missing Data section of the docs:
NA groups in GroupBy are automatically excluded. This behavior is consistent with R
One workaround is to use a placeholder before doing the groupby (e.g. -1):
In [11]: df.fillna(-1)
Out[11]:
a b
0 1 4
1 2 -1
2 3 6
In [12]: df.fillna(-1).groupby('b').sum()
Out[12]:
a
b
-1 2
4 1
6 3
That said, this feels pretty awful hack... perhaps there should be an option to include NaN in groupby (see this github issue - which uses the same placeholder hack).
However, as described in another answer, "from pandas 1.1 you have better control over this behavior, NA values are now allowed in the grouper using dropna=False"
Ancient topic, if someone still stumbles over this--another workaround is to convert via .astype(str) to string before grouping. That will conserve the NaN's.
df = pd.DataFrame({'a': ['1', '2', '3'], 'b': ['4', np.NaN, '6']})
df['b'] = df['b'].astype(str)
df.groupby(['b']).sum()
a
b
4 1
6 3
nan 2
I am not able to add a comment to M. Kiewisch since I do not have enough reputation points (only have 41 but need more than 50 to comment).
Anyway, just want to point out that M. Kiewisch solution does not work as is and may need more tweaking. Consider for example
>>> df = pd.DataFrame({'a': [1, 2, 3, 5], 'b': [4, np.NaN, 6, 4]})
>>> df
a b
0 1 4.0
1 2 NaN
2 3 6.0
3 5 4.0
>>> df.groupby(['b']).sum()
a
b
4.0 6
6.0 3
>>> df.astype(str).groupby(['b']).sum()
a
b
4.0 15
6.0 3
nan 2
which shows that for group b=4.0, the corresponding value is 15 instead of 6. Here it is just concatenating 1 and 5 as strings instead of adding it as numbers.
All answers provided thus far result in potentially dangerous behavior as it is quite possible you select a dummy value that is actually part of the dataset. This is increasingly likely as you create groups with many attributes. Simply put, the approach doesn't always generalize well.
A less hacky solve is to use pd.drop_duplicates() to create a unique index of value combinations each with their own ID, and then group on that id. It is more verbose but does get the job done:
def safe_groupby(df, group_cols, agg_dict):
# set name of group col to unique value
group_id = 'group_id'
while group_id in df.columns:
group_id += 'x'
# get final order of columns
agg_col_order = (group_cols + list(agg_dict.keys()))
# create unique index of grouped values
group_idx = df[group_cols].drop_duplicates()
group_idx[group_id] = np.arange(group_idx.shape[0])
# merge unique index on dataframe
df = df.merge(group_idx, on=group_cols)
# group dataframe on group id and aggregate values
df_agg = df.groupby(group_id, as_index=True)\
.agg(agg_dict)
# merge grouped value index to results of aggregation
df_agg = group_idx.set_index(group_id).join(df_agg)
# rename index
df_agg.index.name = None
# return reordered columns
return df_agg[agg_col_order]
Note that you can now simply do the following:
data_block = [np.tile([None, 'A'], 3),
np.repeat(['B', 'C'], 3),
[1] * (2 * 3)]
col_names = ['col_a', 'col_b', 'value']
test_df = pd.DataFrame(data_block, index=col_names).T
grouped_df = safe_groupby(test_df, ['col_a', 'col_b'],
OrderedDict([('value', 'sum')]))
This will return the successful result without having to worry about overwriting real data that is mistaken as a dummy value.
One small point to Andy Hayden's solution – it doesn't work (anymore?) because np.nan == np.nan yields False, so the replace function doesn't actually do anything.
What worked for me was this:
df['b'] = df['b'].apply(lambda x: x if not np.isnan(x) else -1)
(At least that's the behavior for Pandas 0.19.2. Sorry to add it as a different answer, I do not have enough reputation to comment.)
I answered this already, but some reason the answer was converted to a comment. Nevertheless, this is the most efficient solution:
Not being able to include (and propagate) NaNs in groups is quite aggravating. Citing R is not convincing, as this behavior is not consistent with a lot of other things. Anyway, the dummy hack is also pretty bad. However, the size (includes NaNs) and the count (ignores NaNs) of a group will differ if there are NaNs.
dfgrouped = df.groupby(['b']).a.agg(['sum','size','count'])
dfgrouped['sum'][dfgrouped['size']!=dfgrouped['count']] = None
When these differ, you can set the value back to None for the result of the aggregation function for that group.