We Keep Coding

Python: how to speed up this function and make it more scalable?

I have the following function which accepts an indicator matrix of shape (20,000 x 20,000). And I have to run the function 20,000 x 20,000 = 400,000,000 times. Note that the indicator_Matrix has to be in the form of a pandas dataframe when passed as parameter into the function, as my actual problem's dataframe has timeIndex and integer columns but I have simplified this a bit for the sake of understanding the problem.
Pandas Implementation
indicator_Matrix = pd.DataFrame(np.random.randint(0,2,[20000,20000]))
def operations(indicator_Matrix):
s = indicator_Matrix.sum(axis=1)
d = indicator_Matrix.div(s,axis=0)
res = d[d>0].mean(axis=0)
return res.iloc[-1]
I tried to improve it by using numpy but it is still taking ages to run. I also tried concurrent.future.ThreadPoolExecutor but it still take a long time to run and not much improvement from list comprehension.
Numpy Implementation
indicator_Matrix = pd.DataFrame(np.random.randint(0,2,[20000,20000]))
def operations(indicator_Matrix):
s = indicator_Matrix.to_numpy().sum(axis=1)
d = (indicator_Matrix.to_numpy().T / s).T
d = pd.DataFrame(d, index = indicator_Matrix.index, columns = indicator_Matrix.columns)
res = d[d>0].mean(axis=0)
return res.iloc[-1]
output = [operations(indicator_Matrix) for i in range(0,20000**2)]
Note that the reason I convert d to a dataframe again is because I need to obtain the column means and retain only the last column mean using .iloc[-1]. d[d>0].mean(axis=0) return column means, i.e.
2478 1.0
0 1.0
Update: I am still stuck in this problem. I wonder if using gpu packages like cudf and CuPy on my local desktop would make any difference.

Assuming the answer of #CrazyChucky is correct, one can implement a faster parallel Numba implementation. The idea is to use plain loops and care about reading data the contiguous way. Reading data contiguously is important so to make the computation cache-friendly/memory-efficient. Here is an implementation:
import numba as nb
#nb.njit(['(int_[:,:],)', '(int_[:,::1],)', '(int_[::1,:],)'], parallel=True)
def compute_fastest(matrix):
n, m = matrix.shape
sum_by_row = np.zeros(n, matrix.dtype)
is_row_major = matrix.strides[0] >= matrix.strides[1]
if is_row_major:
for i in nb.prange(n):
s = 0
for j in range(m):
s += matrix[i, j]
sum_by_row[i] = s
else:
for chunk_id in nb.prange(0, (n+63)//64):
start = chunk_id * 64
end = min(start+64, n)
for j in range(m):
for i2 in range(start, end):
sum_by_row[i2] += matrix[i2, j]
count = 0
s = 0.0
for i in range(n):
value = matrix[i, -1] / sum_by_row[i]
if value > 0:
s += value
count += 1
return s / count
# output = [compute_fastest(indicator_Matrix.to_numpy()) for i in range(0,20000**2)]
Pandas dataframes can contain both row-major and column-major arrays. Regarding the memory layout, it is better to iterate over the rows or the column. This is why there is two implementations of the sum based on is_row_major. There is also 3 Numba signatures: one for row-major contiguous arrays, one for columns-major contiguous arrays and one for non-contiguous arrays. Numba will compile the 3 function variants and automatically pick the best one at runtime. The JIT-compiler of Numba can generate a faster implementation (eg. using SIMD instructions) when the input 2D array is known to be contiguous.
Experimental Results
This computation is about 14.5 times faster than operations_simpler on my i5-9600KF processor (6 cores). It still takes a lot of time but the computation is memory-bound and nearly optimal on my machine: it is bounded by the main-memory which has to be read:
On a 2000x2000 dataframe with 32-bit integers:
- operations: 86.310 ms/iter
- operations_simpler: 5.450 ms/iter
- compute_fastest: 0.375 ms/iter
- optimal: 0.345-0.370 ms/iter
If you want to get a faster code, then you need to use more compact data types. For example, a uint8 data type is large enough to contain the values 0 and 1, and it is 4 times smaller in memory on Windows. This means the code can be up to 4 time faster in this case. The smaller the data type, the faster the program. One could even try to compact 8 columns in 1 using bit tweaks though it is generally significantly slower using Numba unless you have a lot of available cores.
Notes & Discussion
The above code works only with uniformly-typed columns. If this is not the case, you can split the dataframe in multiple groups and convert each column group to Numpy array so to then call the Numba function (modified to support groups). Note the #CrazyChucky code has a similar issue: a dataframe column with mixed datatypes converted to a Numpy array results in an object-based Numpy array which is very inefficient (especially a row-major Numpy array).
Note that using a GPU will not make the computation faster unless the input dataframe is already stored in the GPU memory. Indeed, CPU-GPU data transfers are more expensive than just reading the RAM (due to the interconnect overhead which is generally a quite slow PCI one). Note that the GPU memory is quite limited compared to the CPU. If the target dataframe(s) do not need to be transferred, then using cudf is relatively simple and should give a small speed up. For a faster code, one need to implement a fast CUDA code but this is clearly far from being easy for dataframes with mixed dataype. In the end, the resulting speed up should be main_ram_throughput / gpu_ram_througput assuming there is no data transfer. Note that this factor is generally 5-12. Note also that CUDA and cudf require a Nvidia GPU.
Finally, reducing the input data size or just the amount of computation is certainly the best solution (as indicated in the comment by #zvone) since it is very computationally intensive.

You're doing some extra math you don't have to. In plain English, what you're doing is:
Summing each column
Turning the list of sums "sideways" and dividing each column by it
Taking the mean of each column, ignoring values ≤ 0
Returning only the rightmost mean
After step one, you no longer need anything but the rightmost column; you can ignore the other columns, only dividing and averaging the one whose result you care about. Changing your code accordingly:
def operations_simpler(indicator_matrix):
sums = indicator_matrix.sum(axis=1)
last_column = indicator_matrix.iloc[:, -1]
divided = last_column / sums
return divided[divided > 0].mean()
...yields the same result, and takes about a hundredth of the time. Extrapolating from shorter test runs, this cuts the time for 400,000,000 runs on my machine from about 114 years down to... about 324 days. Still not great. So far I've not managed to get it to run any faster by converting to NumPy, compiling with Numba, or employing multiprocessing, but I'll go ahead and post this for now in case it's helpful.
Note: You're unlikely to see any improvements with compute-heavy work like this from threading; if anything, you'd want to use multiprocessing. concurrent.futures offers executors for both. Threads are mostly useful to avoid waiting around for I/O.

As per the previous answer you can use Numba or you can you two other alternatives such as Dask which is a distributed computing package, to parallelize your function's execution it can divide your data into smaller bits and distribute computing across many CPU cores or even numerous machines.
import dask.array as da
def operations(indicator_matrix):
s = indicator_matrix.sum(axis=1)
d = indicator_matrix.div(s, axis=0)
res = d[d > 0].mean(axis=0)
return res.iloc[-1]
indicator_matrix_dask = da.from_array(indicator_matrix, chunks=(1000, 1000))
output_dask = indicator_matrix_dask.map_blocks(operations, dtype=float)
output = output_dask.compute()
or you can use CuPy which uses GPU to increase your function excution
import cupy as cp
def operations(indicator_matrix):
s = cp.sum(indicator_matrix, axis=1)
d = cp.divide(indicator_matrix.T, s).T
d = pd.DataFrame(d, index = indicator_matrix.index, columns = indicator_matrix.columns)
res = d[d > 0].mean(axis=0)
return res.iloc[-1]
indicator_matrix_cupy = cp.asarray(indicator_matrix)
output_cupy = operations(indicator_matrix_cupy)
output = cp.asnumpy(output_cupy)

Efficient Way to Repeatedly Split Large NumPy Array and Record Middle

I have a large NumPy array nodes = np.arange(100_000_000) and I need to rearrange this array by:
Recording and then removing the middle value in the array
Split the array into the left half and right half
Repeat Steps 1-2 for each half
Stop when all values are exhausted
So, for a smaller input example nodes = np.arange(10), the output would be:
[5 2 8 1 4 7 9 0 3 6]
This was accomplished by naively doing:
import numpy as np
def split(node, out):
mid = len(node) // 2
out.append(node[mid])
return node[:mid], node[mid+1:]
def reorder(a):
nodes = [a.tolist()]
out = []
while nodes:
tmp = []
for node in nodes:
for n in split(node, out):
if n:
tmp.append(n)
nodes = tmp
return np.array(out)
if __name__ == "__main__":
nodes = np.arange(10)
print(reorder(nodes))
However, this is way too slow for nodes = np.arange(100_000_000) and so I am looking for a much faster solution.

You can vectorize your function with Numpy by working on groups of slices.
Here is an implementation:
# Similar to [e for tmp in zip(a, b) for e in tmp] ,
# but on Numpy arrays and much faster
def interleave(a, b):
assert len(a) == len(b)
return np.column_stack((a, b)).reshape(len(a) * 2)
# n is the length of the input range (len(a) in your example)
def fast_reorder(n):
if n == 0:
return np.empty(0, dtype=np.int32)
startSlices = np.array([0], dtype=np.int32)
endSlices = np.array([n], dtype=np.int32)
allMidSlices = np.empty(n, dtype=np.int32) # Similar to "out" in your implementation
midInsertCount = 0 # Actual size of allMidSlices
# Generate a bunch of middle values as long as there is valid slices to split
while midInsertCount < n:
# Generate the new mid/left/right slices
midSlices = (endSlices + startSlices) // 2
# Computing the next slices is not needed for the last step
if midInsertCount + len(midSlices) < n:
# Generate the nexts slices (possibly with invalid ones)
newStartSlices = interleave(startSlices, midSlices+1)
newEndSlices = interleave(midSlices, endSlices)
# Discard invalid slices
isValidSlices = newStartSlices < newEndSlices
startSlices = newStartSlices[isValidSlices]
endSlices = newEndSlices[isValidSlices]
# Fast appending
allMidSlices[midInsertCount:midInsertCount+len(midSlices)] = midSlices
midInsertCount += len(midSlices)
return allMidSlices[0:midInsertCount]
On my machine, this is 89 times faster than your scalar implementation with the input np.arange(100_000_000) dropping from 2min35 to 1.75s. It also consume far less memory (rougthly 3~4 times less). Note that if you want a faster code, then you probably need to use a native language like C or C++.

Edit:
The question has been updated to have a much smaller input array so I leave the below for historical reasons. Basically it was likely a typo but we often get accustomed to computers working with insanely large numbers and when memory is involved they can be a real problem.
There is already a numpy based solution submitted by someone else that I think fits the bill.
Your code requires an insane amount of RAM just to hold 100 billion 64 bit integers. Do you have 800GB of RAM? Then you convert the numpy array to a list which will be substantially larger than the array (each packed 64 bit int in the numpy array will become a much less memory efficient python int object and the list will have a pointer to that object). Then you make a lot of slices of the list which will not duplicate the data but will duplicate the pointers to the data and use even more RAM. You also append all the result values to a list a single value at a time. Lists are very fast for adding items generally but with such an extreme size this will not only be slow but the way the list is allocated is likely to be extremely wasteful RAM wise and contribute to major problems (I believe they double in size when they get to a certain level of fullness so you will end up allocating more RAM than you need and doing many allocations and likely copies). What kind of machine are you running this on? There are ways to improve your code but unless you're running it on a super computer I don't know that you're going to ever finish that calculation. I only..only? have 32GB of RAM and I'm not going to even try to create a 100B int_64 numpy array as I don't want to use up ssd write life for a mass of virtual memory.
As for improving your code stick to numpy arrays don't change to a python list it will greatly increase the RAM you need. Preallocate a numpy array to put the answer in. Then you need a new algorithm. Anything recursive or recursive like (ie a loop splitting the input,) will require tracking a lot of state, your nodes list is going to be extraordinarily gigantic and again use a lot of RAM. You could use len(a) to indicate values that are removed from your list and scan through the entire array each time to figure out what to do next but that will save RAM in favour of a tremendous amount of searching a gigantic array. I feel like there is an algorithm to cut numbers from each end and place them in the output and just track the beginning and end but I haven't figured it out at least not yet.
I also think there is a simpler algorithm where you just track the number of splits you've done instead of making a giant list of slices and keeping it all in memory. Take the middle of the left half and then the middle of the right then count up one and when you take the middle of the left half's left half you know you have to jump to the right half then the count is one so you jump over to the original right half's left half and on and on... Based on the depth into the halves and the length of the input you should be able to jump around without scanning or tracking all of those slices though I haven't been able to dedicate much time to thinking this through in my head.
With a problem of this nature if you really need to push the limits you should consider using C/C++ so you can be as efficient as possible with RAM usage and because you're doing an insane number of tiny things which doesn't map well to python performance.

Performance of sorting structured arrays (numpy)

I have an array with several fields, which I want to be sorted with respect to 2 of them. One of these fields is binary, e.g.:
size = 100000
data = np.empty(
shape=2 * size,
dtype=[('class', int),
('value', int),]
)
data['class'][:size] = 0
data['value'][:size] = (np.random.normal(size=size) * 10).astype(int)
data['class'][size:] = 1
data['value'][size:] = (np.random.normal(size=size, loc=0.5) * 10).astype(int)
np.random.shuffle(data)
I need the result to be sorted with respect to value, and for same values class=0 should go first. Doing it like so (a):
idx = np.argsort(data, order=['value', 'class'])
data_sorted = data[idx]
seems to be an order of magnitude slower compared to sorting just data['value']. Is there a way to improve the speed, given that there are only two classes?
By experimenting randomly I noticed that an approach like this (b):
idx = np.argsort(data['value'])
data_sorted = data[idx]
idx = np.argsort(data_sorted, order=['value', 'class'], kind='mergesort')
data_sorted = data_sorted[idx]
takes ~20% less time than (a). Changing field datatypes seem to also have some effect - floats instead of ints seem to be slightly faster.

The simplest way to do this is using the order parameter of sort
sort(data, order=['value', 'class'])
However, this takes 121 ms to run on my computer, while data['class'] and data['value'] take only 2.44 and 5.06 ms respectively. Interestingly, sort(data, order='class') takes 135 ms again, suggesting the problem is with sorting structured arrays.
So, the approach you've taken of sorting each field using argsort then indexing the final array seems to be on the right track. However, you need to sort each field individually,
idx=argsort(data['class'])
data_sorted = data[idx][argsort(data['value'][idx], kind='stable')]
This runs in 43.9 ms.
You can get a very slight speedup by removing one temporary array from indexing
idx = argsort(data['class'])
tmp = data[idx]
data_sorted = tmp[argsort(tmp['value'], kind='stable')]
Which runs in 40.8 ms. Not great, but it is a workaround if performance is critical.
This seems to be a known problem:
sorting numpy structured and record arrays is very slow
Edit
The sourcecode for the comparisons used in sort can be seen at https://github.com/numpy/numpy/blob/dea85807c258ded3f75528cce2a444468de93bc1/numpy/core/src/multiarray/arraytypes.c.src .
The numeric types are much, much simpler. Still, that large of a difference in performance is surprising.

In addition to the good (general-purpose) answer of #user2699, in your specific case, you can cheat because the two fields of the structured array is of the same integer type and values are relatively small (they fit in 32-bits). The cheat consists in the following steps:
subtract the minimum values of each fields to all items the field (to make them positive) using arr - np.min(arr)
transform each field to a np.uint64 with np.astype
pack bits the two fields in one binary array using: (class_arr << 32) | value_arr
sort the resulting array using np.sort
unpack the array using: class_arr = sorted_arr >> 32 and value_arr = sorted_arr & ((1<<32)-1)
This strategy is significantly faster than using two np.argsort that are pretty expensive. This is especially true for bigger array since sorting big array is even more expensive and np.sort is cheaper than np.argsort. Not to mention indirect indexing is relatively slow on big array because of the unpredictable pseudo-random memory access pattern and the high latency of the RAM. The downside of this approach is that it is a bit more tricky to implement and it does not apply in all cases.

Equation calculations with 4D arrays

Basically I have over 1000 3D arrays with the shape (100,100,1000). So some pretty large arrays, which I need to use in some calculations. The great thing about python and Numpy is that instead of interations, calculations on each element and such can be done very quickly. For example, I can make a sum of each index for each 3D array almost instant. The result is one large array with the sum of each index for each array. In principle, that is ALMOST what I want to do, however, there is a bit of a problem.
What I need to do is use an equation that looks like this:
So as stated, I have around 1000 3D arrays. In total, the shape of this total array is (1000, 100, 100, 1000). For each of the 1000 I also have a list going from 1 to 1000 that corresponds to the 1000 3D arrays, and each index of that list contains either a 1 or a 0. If it has a 1 that entire 3D array of that index should go in the first term of the equation, and if 0, it goes into the other.
I am however very much in doubt about how I am going to do this without turning to some kind of looping that might destroy the speed of the calculations by a great deal.

You could sort it by locating the 1's and 0's.
Something like:
list_ones = np.where(Array[0] == 1)
list_zeros = np.where(Array[0] == 0)
Then Array[list_ones,:,:,:] will contain all elements corresponding to a one and Array[list_zeros,:,:,:] will correspond to all elements corresponding to a zero.
Then you can just put
first_term = Array[list_ones,:,:,:]
second_term = Array[list_zeros,:,:,:]
And sum as appropriate.
Would this work for your purpose?

Fast way to construct a matrix in Python

I have been browsing through the questions, and could find some help, but I prefer having confirmation by asking it directly. So here is my problem.
I have an (numpy) array u of dimension N, from which I want to build a square matrix k of dimension N^2. Basically, each matrix element k(i,j) is defined as k(i,j)=exp(-|u_i-u_j|^2).
My first naive way to do it was like this, which is, I believe, Fortran-like:
for i in range(N):
for j in range(N):
k[i][j]=np.exp(np.sum(-(u[i]-u[j])**2))
However, this is extremely slow. For N=1000, for example, it is taking around 15 seconds.
My other way to proceed is the following (inspired by other questions/answers):
i, j = np.ogrid[:N,:N]
k = np.exp(np.sum(-(u[i]-u[j])**2,axis=2))
This is way faster, as for N=1000, the result is almost instantaneous.
So I have two questions.
1) Why is the first method so slow, and why is the second one so fast ?
2) Is there a faster way to do it ? For N=10000, it is starting to take quite some time already, so I really don't know if this was the "right" way to do it.
Thank you in advance !
P.S: the matrix is symmetric, so there must also be a way to make the process faster by calculating only the upper half of the matrix, but my question was more related to the way to manipulate arrays, etc.

First, a small remark, there is no need to use np.sum if u can be re-written as u = np.arange(N). Which seems to be the case since you wrote that it is of dimension N.
1) First question:
Accessing indices in Python is slow, so best is to not use [] if there is a way to not use it. Plus you call multiple times np.exp and np.sum, whereas they can be called for vectors and matrices. So, your second proposal is better since you compute your k all in once, instead of elements by elements.
2) Second question:
Yes there is. You should consider using only numpy functions and not using indices (around 3 times faster):
k = np.exp(-np.power(np.subtract.outer(u,u),2))
(NB: You can keep **2 instead of np.power, which is a bit faster but has smaller precision)
edit (Take into account that u is an array of tuples)
With tuple data, it's a bit more complicated:
ma = np.subtract.outer(u[:,0],u[:,0])**2
mb = np.subtract.outer(u[:,1],u[:,1])**2
k = np.exp(-np.add(ma, mb))
You'll have to use twice np.substract.outer since it will return a 4 dimensions array if you do it in one time (and compute lots of useless data), whereas u[i]-u[j] returns a 3 dimensions array.
I used np.add instead of np.sum since it keep the array dimensions.
NB: I checked with
N = 10000
u = np.random.random_sample((N,2))
I returns the same as your proposals. (But 1.7 times faster)