Mean of values in a column for unique values in another column

Question

I am using Python 2.7 (Anaconda) for processing tabular data. I have loaded a textfile with two columns, e.g.

[[ 1.  8.]
 [ 2.  4.]
 [ 3.  1.]
 [ 4.  5.]
 [ 5.  6.]
 [ 1.  9.]
 [ 2.  0.]
 [ 3.  7.]
 [ 4.  3.]
 [ 5.  2.]]

my goal is to calculate the mean over all values in the second column which match the unique values in the first one, e.g. the mean value for 1 would be 8.5, for 2 it would be two, for 3 it would be 4. First, I filtered out the unique values in the first column by extracting the column and applying np.unique() resulting in the array "unique". I created a loop that works when I define the unique value:

mean= 0
values=[]
for i in range(0,len(first),1):
    if first[i]==1:
        values.append(second[i])
print(np.mean(values))

where first and second are the specific columns. Now I want to make this not so specific. I tried

mean = 0
values = []
means=[]

for i in unique:
    for k in range(0,len(first),1):
        if first[k]==i:
            values.append(second[k])
            mean = np.mean(values)
            means.append(mean)
    mean=0
    values=[]
print(means)

but it only returns the original second column. Does anybody have an idea on how to make this code non-specific? In reality, I have about 70k rows, so I cannot do it manually.

Answer 1

In pandas, you can achieve this by using groupby:

In [97]: data
Out[97]: 
array([[ 1.,  8.],
       [ 2.,  4.],
       [ 3.,  1.],
       [ 4.,  5.],
       [ 5.,  6.],
       [ 1.,  9.],
       [ 2.,  0.],
       [ 3.,  7.],
       [ 4.,  3.],
       [ 5.,  2.]])

In [98]: import pandas as pd

In [99]: df = pd.DataFrame(data, columns=['first', 'second'])

In [100]: df.groupby('first').mean().reset_index()
Out[100]: 
   first  second
0    1.0     8.5
1    2.0     2.0
2    3.0     4.0
3    4.0     4.0
4    5.0     4.0

Answer 2

Write a comparison statement checking the first column for your unique value, use that statement as a boolean index,

>>> mask = a[:,0] == 1
>>> a[mask]
array([[ 1.,  8.],
       [ 1.,  9.]])

for n in np.unique(a[:,0]):
    mask = a[:,0] == n
    print(np.mean(a[mask], axis = 0))

>>> 
[ 1.   8.5]
[ 2.  2.]
[ 3.  4.]
[ 4.  4.]
[ 5.  4.]

---

If your data file looks something like this

'''
1.,  8.
2.,  4.
3.,  1.
4.,  5.
'''

and you don't really need a numpy array, just use a dictionary:

import collections
d = collections.defaultdict(list)
with open('file.txt') as f:
    for line in f:
        line = line.strip()
        first, second = map(float, line.split(','))
        d[first].append(second)

for first, second in d.iteritems():
    print(first, sum(second) / len(second))