How do I efficiently obtain the frequency count for each unique value in a NumPy array?
>>> x = np.array([1,1,1,2,2,2,5,25,1,1])
>>> freq_count(x)
[(1, 5), (2, 3), (5, 1), (25, 1)]
Use numpy.unique with return_counts=True (for NumPy 1.9+):
import numpy as np
x = np.array([1,1,1,2,2,2,5,25,1,1])
unique, counts = np.unique(x, return_counts=True)
>>> print(np.asarray((unique, counts)).T)
[[ 1 5]
[ 2 3]
[ 5 1]
[25 1]]
In comparison with scipy.stats.itemfreq:
In [4]: x = np.random.random_integers(0,100,1e6)
In [5]: %timeit unique, counts = np.unique(x, return_counts=True)
10 loops, best of 3: 31.5 ms per loop
In [6]: %timeit scipy.stats.itemfreq(x)
10 loops, best of 3: 170 ms per loop
Take a look at np.bincount:
http://docs.scipy.org/doc/numpy/reference/generated/numpy.bincount.html
import numpy as np
x = np.array([1,1,1,2,2,2,5,25,1,1])
y = np.bincount(x)
ii = np.nonzero(y)[0]
And then:
zip(ii,y[ii])
# [(1, 5), (2, 3), (5, 1), (25, 1)]
or:
np.vstack((ii,y[ii])).T
# array([[ 1, 5],
[ 2, 3],
[ 5, 1],
[25, 1]])
or however you want to combine the counts and the unique values.
Use this:
>>> import numpy as np
>>> x = [1,1,1,2,2,2,5,25,1,1]
>>> np.array(np.unique(x, return_counts=True)).T
array([[ 1, 5],
[ 2, 3],
[ 5, 1],
[25, 1]])
Original answer:
Use scipy.stats.itemfreq (warning: deprecated):
>>> from scipy.stats import itemfreq
>>> x = [1,1,1,2,2,2,5,25,1,1]
>>> itemfreq(x)
/usr/local/bin/python:1: DeprecationWarning: `itemfreq` is deprecated! `itemfreq` is deprecated and will be removed in a future version. Use instead `np.unique(..., return_counts=True)`
array([[ 1., 5.],
[ 2., 3.],
[ 5., 1.],
[ 25., 1.]])
I was also interested in this, so I did a little performance comparison (using perfplot, a pet project of mine). Result:
y = np.bincount(a)
ii = np.nonzero(y)[0]
out = np.vstack((ii, y[ii])).T
is by far the fastest. (Note the log-scaling.)
Code to generate the plot:
import numpy as np
import pandas as pd
import perfplot
from scipy.stats import itemfreq
def bincount(a):
y = np.bincount(a)
ii = np.nonzero(y)[0]
return np.vstack((ii, y[ii])).T
def unique(a):
unique, counts = np.unique(a, return_counts=True)
return np.asarray((unique, counts)).T
def unique_count(a):
unique, inverse = np.unique(a, return_inverse=True)
count = np.zeros(len(unique), dtype=int)
np.add.at(count, inverse, 1)
return np.vstack((unique, count)).T
def pandas_value_counts(a):
out = pd.value_counts(pd.Series(a))
out.sort_index(inplace=True)
out = np.stack([out.keys().values, out.values]).T
return out
b = perfplot.bench(
setup=lambda n: np.random.randint(0, 1000, n),
kernels=[bincount, unique, itemfreq, unique_count, pandas_value_counts],
n_range=[2 ** k for k in range(26)],
xlabel="len(a)",
)
b.save("out.png")
b.show()
Using pandas module:
>>> import pandas as pd
>>> import numpy as np
>>> x = np.array([1,1,1,2,2,2,5,25,1,1])
>>> pd.value_counts(x)
1 5
2 3
25 1
5 1
dtype: int64
This is by far the most general and performant solution; surprised it hasn't been posted yet.
import numpy as np
def unique_count(a):
unique, inverse = np.unique(a, return_inverse=True)
count = np.zeros(len(unique), np.int)
np.add.at(count, inverse, 1)
return np.vstack(( unique, count)).T
print unique_count(np.random.randint(-10,10,100))
Unlike the currently accepted answer, it works on any datatype that is sortable (not just positive ints), and it has optimal performance; the only significant expense is in the sorting done by np.unique.
numpy.bincount is the probably the best choice. If your array contains anything besides small dense integers it might be useful to wrap it something like this:
def count_unique(keys):
uniq_keys = np.unique(keys)
bins = uniq_keys.searchsorted(keys)
return uniq_keys, np.bincount(bins)
For example:
>>> x = array([1,1,1,2,2,2,5,25,1,1])
>>> count_unique(x)
(array([ 1, 2, 5, 25]), array([5, 3, 1, 1]))
Even though it has already been answered, I suggest a different approach that makes use of numpy.histogram. Such function given a sequence it returns the frequency of its elements grouped in bins.
Beware though: it works in this example because numbers are integers. If they where real numbers, then this solution would not apply as nicely.
>>> from numpy import histogram
>>> y = histogram (x, bins=x.max()-1)
>>> y
(array([5, 3, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
1]),
array([ 1., 2., 3., 4., 5., 6., 7., 8., 9., 10., 11.,
12., 13., 14., 15., 16., 17., 18., 19., 20., 21., 22.,
23., 24., 25.]))
Old question, but I'd like to provide my own solution which turn out to be the fastest, use normal list instead of np.array as input (or transfer to list firstly), based on my bench test.
Check it out if you encounter it as well.
def count(a):
results = {}
for x in a:
if x not in results:
results[x] = 1
else:
results[x] += 1
return results
For example,
>>>timeit count([1,1,1,2,2,2,5,25,1,1]) would return:
100000 loops, best of 3: 2.26 µs per loop
>>>timeit count(np.array([1,1,1,2,2,2,5,25,1,1]))
100000 loops, best of 3: 8.8 µs per loop
>>>timeit count(np.array([1,1,1,2,2,2,5,25,1,1]).tolist())
100000 loops, best of 3: 5.85 µs per loop
While the accepted answer would be slower, and the scipy.stats.itemfreq solution is even worse.
A more indepth testing did not confirm the formulated expectation.
from zmq import Stopwatch
aZmqSTOPWATCH = Stopwatch()
aDataSETasARRAY = ( 100 * abs( np.random.randn( 150000 ) ) ).astype( np.int )
aDataSETasLIST = aDataSETasARRAY.tolist()
import numba
#numba.jit
def numba_bincount( anObject ):
np.bincount( anObject )
return
aZmqSTOPWATCH.start();np.bincount( aDataSETasARRAY );aZmqSTOPWATCH.stop()
14328L
aZmqSTOPWATCH.start();numba_bincount( aDataSETasARRAY );aZmqSTOPWATCH.stop()
592L
aZmqSTOPWATCH.start();count( aDataSETasLIST );aZmqSTOPWATCH.stop()
148609L
Ref. comments below on cache and other in-RAM side-effects that influence a small dataset massively repetitive testing results.
import pandas as pd
import numpy as np
x = np.array( [1,1,1,2,2,2,5,25,1,1] )
print(dict(pd.Series(x).value_counts()))
This gives you:
{1: 5, 2: 3, 5: 1, 25: 1}
To count unique non-integers - similar to Eelco Hoogendoorn's answer but considerably faster (factor of 5 on my machine), I used weave.inline to combine numpy.unique with a bit of c-code;
import numpy as np
from scipy import weave
def count_unique(datain):
"""
Similar to numpy.unique function for returning unique members of
data, but also returns their counts
"""
data = np.sort(datain)
uniq = np.unique(data)
nums = np.zeros(uniq.shape, dtype='int')
code="""
int i,count,j;
j=0;
count=0;
for(i=1; i<Ndata[0]; i++){
count++;
if(data(i) > data(i-1)){
nums(j) = count;
count = 0;
j++;
}
}
// Handle last value
nums(j) = count+1;
"""
weave.inline(code,
['data', 'nums'],
extra_compile_args=['-O2'],
type_converters=weave.converters.blitz)
return uniq, nums
Profile info
> %timeit count_unique(data)
> 10000 loops, best of 3: 55.1 µs per loop
Eelco's pure numpy version:
> %timeit unique_count(data)
> 1000 loops, best of 3: 284 µs per loop
Note
There's redundancy here (unique performs a sort also), meaning that the code could probably be further optimized by putting the unique functionality inside the c-code loop.
multi-dimentional frequency count, i.e. counting arrays.
>>> print(color_array )
array([[255, 128, 128],
[255, 128, 128],
[255, 128, 128],
...,
[255, 128, 128],
[255, 128, 128],
[255, 128, 128]], dtype=uint8)
>>> np.unique(color_array,return_counts=True,axis=0)
(array([[ 60, 151, 161],
[ 60, 155, 162],
[ 60, 159, 163],
[ 61, 143, 162],
[ 61, 147, 162],
[ 61, 162, 163],
[ 62, 166, 164],
[ 63, 137, 162],
[ 63, 169, 164],
array([ 1, 2, 2, 1, 4, 1, 1, 2,
3, 1, 1, 1, 2, 5, 2, 2,
898, 1, 1,
import pandas as pd
import numpy as np
print(pd.Series(name_of_array).value_counts())
from collections import Counter
x = array( [1,1,1,2,2,2,5,25,1,1] )
mode = counter.most_common(1)[0][0]
Most of simple problems get complicated because simple functionality like order() in R that gives a statistical result in both and descending order is missing in various python libraries. But if we devise our thinking that all such statistical ordering and parameters in python are easily found in pandas, we can can result sooner than looking in 100 different places. Also, development of R and pandas go hand-in-hand because they were created for same purpose. To solve this problem I use following code that gets me by anywhere:
unique, counts = np.unique(x, return_counts=True)
d = {'unique':unique, 'counts':count} # pass the list to a dictionary
df = pd.DataFrame(d) #dictionary object can be easily passed to make a dataframe
df.sort_values(by = 'count', ascending=False, inplace = True)
df = df.reset_index(drop=True) #optional only if you want to use it further
some thing like this should do it:
#create 100 random numbers
arr = numpy.random.random_integers(0,50,100)
#create a dictionary of the unique values
d = dict([(i,0) for i in numpy.unique(arr)])
for number in arr:
d[j]+=1 #increment when that value is found
Also, this previous post on Efficiently counting unique elements seems pretty similar to your question, unless I'm missing something.
You can write freq_count like this:
def freq_count(data):
mp = dict();
for i in data:
if i in mp:
mp[i] = mp[i]+1
else:
mp[i] = 1
return mp
In numpy / PyTorch, I have two matrices, e.g. X=[[1,2],[3,4],[5,6]], Y=[[1,1],[2,2]]. I would like to dot product every row of X with every row of Y, and have the results
[[3, 6],[7, 14], [11,22]]
How do I achieve this?, Thanks!
I think this is what you are looking for:
import numpy as np
x= [[1,2],[3,4],[5,6]]
y= [[1,1],[2,2]]
x = np.asarray(x) #convert list to numpy array
y = np.asarray(y) #convert list to numpy array
product = np.dot(x, y.T)
.T transposes the matrix, which is neccessary in this case for the multiplication (because of the way dot products are defined). print(product) will output:
[[ 3 6]
[ 7 14]
[11 22]]
Using einsum
np.einsum('ij,kj->ik', X, Y)
array([[ 3, 6],
[ 7, 14],
[11, 22]])
In PyTorch, you can achieve this using torch.mm(a, b) or torch.matmul(a, b), as shown below:
x = np.array([[1,2],[3,4],[5,6]])
y = np.array([[1,1],[2,2]])
x = torch.from_numpy(x)
y = torch.from_numpy(y)
# print(torch.matmul(x, torch.t(y)))
print(torch.mm(x, torch.t(y)))
output:
tensor([[ 3, 6],
[ 7, 14],
[11, 22]], dtype=torch.int32)
I have a numpy array A of size ((s1,...sm)) with integer entries and a dictionary D with integers as keys and numpy arrays of size ((t)) as values. I would like to evaluate the dictionary on every entry of the array A to get a new array B of size ((s1,...sm,t)).
For example
D={1:[0,1],2:[1,0]}
A=np.array([1,2,1])
The output shout be
array([[0,1],[1,0],[0,1]])
Motivation: I have an array with indexes of unit vectors as entries and I need to transform it into an array with the vectors as entries.
If you can rename your keys to be 0-indexed, you might use direct array querying on your unit vectors:
>>> units = np.array([D[1], D[2]])
>>> B = units[A - 1] # -1 because 0 indexed: 1 -> 0, 2 -> 1
>>> B
array([[0, 1],
[1, 0],
[0, 1]])
And similarly for any shape:
>>> A = np.random.random_integers(0, 1, (10, 11, 12))
>>> A.shape
(10, 11, 12)
>>> B = units[A]
>>> B.shape
(10, 11, 12, 2)
You can learn more about advanced indexing on the numpy doc
>>> np.asarray([D[key] for key in A])
array([[0, 1],
[1, 0],
[0, 1]])
Here's an approach using np.searchsorted to locate those row indices to index into the values of the dictionary and then simply indexing it to get the desired output, like so -
idx = np.searchsorted(D.keys(),A)
out = np.asarray(D.values())[idx]
Sample run -
In [45]: A
Out[45]: array([1, 2, 1])
In [46]: D
Out[46]: {1: [0, 1], 2: [1, 0]}
In [47]: idx = np.searchsorted(D.keys(),A)
...: out = np.asarray(D.values())[idx]
...:
In [48]: out
Out[48]:
array([[0, 1],
[1, 0],
[0, 1]])
I am having an issue with Ipython - Numpy. I want to do the following operation:
x^T.x
with and x^T the transpose operation on vector x. x is extracted from a txt file with the instruction:
x = np.loadtxt('myfile.txt')
The problem is that if i use the transpose function
np.transpose(x)
and uses the shape function to know the size of x, I get the same dimensions for x and x^T. Numpy gives the size with a L uppercase indice after each dimensions. e.g.
print x.shape
print np.transpose(x).shape
(3L, 5L)
(3L, 5L)
Does anybody know how to solve this, and compute x^T.x as a matrix product?
Thank you!
What np.transpose does is reverse the shape tuple, i.e. you feed it an array of shape (m, n), it returns an array of shape (n, m), you feed it an array of shape (n,)... and it returns you the same array with shape(n,).
What you are implicitly expecting is for numpy to take your 1D vector as a 2D array of shape (1, n), that will get transposed into a (n, 1) vector. Numpy will not do that on its own, but you can tell it that's what you want, e.g.:
>>> a = np.arange(4)
>>> a
array([0, 1, 2, 3])
>>> a.T
array([0, 1, 2, 3])
>>> a[np.newaxis, :].T
array([[0],
[1],
[2],
[3]])
As explained by others, transposition won't "work" like you want it to for 1D arrays.
You might want to use np.atleast_2d to have a consistent scalar product definition:
def vprod(x):
y = np.atleast_2d(x)
return np.dot(y.T, y)
I had the same problem, I used numpy matrix to solve it:
# assuming x is a list or a numpy 1d-array
>>> x = [1,2,3,4,5]
# convert it to a numpy matrix
>>> x = np.matrix(x)
>>> x
matrix([[1, 2, 3, 4, 5]])
# take the transpose of x
>>> x.T
matrix([[1],
[2],
[3],
[4],
[5]])
# use * for the matrix product
>>> x*x.T
matrix([[55]])
>>> (x*x.T)[0,0]
55
>>> x.T*x
matrix([[ 1, 2, 3, 4, 5],
[ 2, 4, 6, 8, 10],
[ 3, 6, 9, 12, 15],
[ 4, 8, 12, 16, 20],
[ 5, 10, 15, 20, 25]])
While using numpy matrices may not be the best way to represent your data from a coding perspective, it's pretty good if you are going to do a lot of matrix operations!
For starters L just means that the type is a long int. This shouldn't be an issue. You'll have to give additional information about your problem though since I cannot reproduce it with a simple test case:
In [1]: import numpy as np
In [2]: a = np.arange(12).reshape((4,3))
In [3]: a
Out[3]:
array([[ 0, 1, 2],
[ 3, 4, 5],
[ 6, 7, 8],
[ 9, 10, 11]])
In [4]: a.T #same as np.transpose(a)
Out[4]:
array([[ 0, 3, 6, 9],
[ 1, 4, 7, 10],
[ 2, 5, 8, 11]])
In [5]: a.shape
Out[5]: (4, 3)
In [6]: np.transpose(a).shape
Out[6]: (3, 4)
There is likely something subtle going on with your particular case which is causing problems. Can you post the contents of the file that you're reading into x?
This is either the inner or outer product of the two vectors, depending on the orientation you assign to them. Here is how to calculate either without changing x.
import numpy
x = numpy.array([1, 2, 3])
inner = x.dot(x)
outer = numpy.outer(x, x)
The file 'myfile.txt' contain lines such as
5.100000 3.500000 1.400000 0.200000 1
4.900000 3.000000 1.400000 0.200000 1
Here is the code I run:
import numpy as np
data = np.loadtxt('iris.txt')
x = data[1,:]
print x.shape
print np.transpose(x).shape
print x*np.transpose(x)
print np.transpose(x)*x
And I get as a result
(5L,)
(5L,)
[ 24.01 9. 1.96 0.04 1. ]
[ 24.01 9. 1.96 0.04 1. ]
I would be expecting one of the two last result to be a scalar instead of a vector, because x^T.x (or x.x^T) should give a scalar.
b = np.array([1, 2, 2])
print(b)
print(np.transpose([b]))
print("rows, cols: ", b.shape)
print("rows, cols: ", np.transpose([b]).shape)
Results in
[1 2 2]
[[1]
[2]
[2]]
rows, cols: (3,)
rows, cols: (3, 1)
Here (3,) can be thought as "(3, 0)".
However if you want the transpose of a matrix A, np.transpose(A) is the solution. Shortly, [] converts a vector to a matrix, a matrix to a higher dimension tensor.