I have a vector and wish to make another vector of the same length whose k-th component is
The question is: how can we vectorize this for speed? NumPy vectorize() is actually a for loop, so it doesn't count.
Veedrac pointed out that "There is no way to apply a pure Python function to every element of a NumPy array without calling it that many times". Since I'm using NumPy functions rather than "pure Python" ones, I suppose it's possible to vectorize, but I don't know how.
import numpy as np
from scipy.integrate import quad
ws = 2 * np.random.random(10) - 1
n = len(ws)
integrals = np.empty(n)
def f(x, w):
if w < 0: return np.abs(x * w)
else: return np.exp(x) * w
def temp(x): return np.array([f(x, w) for w in ws]).sum()
def integrand(x, w): return f(x, w) * np.log(temp(x))
## Python for loop
for k in range(n):
integrals[k] = quad(integrand, -1, 1, args = ws[k])[0]
## NumPy vectorize
integrals = np.vectorize(quad)(integrand, -1, 1, args = ws)[0]
On a side note, is a Cython for loop always faster than NumPy vectorization?
The function quad executes an adaptive algorithm, which means the computations it performs depend on the specific thing being integrated. This cannot be vectorized in principle.
In your case, a for loop of length 10 is a non-issue. If the program takes long, it's because integration takes long, not because you have a for loop.
When you absolutely need to vectorize integration (not in the example above), use a non-adaptive method, with the understanding that precision may suffer. These can be directly applied to a 2D NumPy array obtained by evaluating all of your functions on some regularly spaced 1D array (a linspace). You'll have to choose the linspace yourself since the methods aren't adaptive.
numpy.trapz is the simplest and least precise
scipy.integrate.simps is equally easy to use and more precise (Simpson's rule requires an odd number of samples, but the method works around having an even number, too).
scipy.integrate.romb is in principle of higher accuracy than Simpson (for smooth data) but it requires the number of samples to be 2**n+1 for some integer n.
#zaq's answer focusing on quad is spot on. So I'll look at some other aspects of the problem.
In recent https://stackoverflow.com/a/41205930/901925 I argue that vectorize is of most value when you need to apply the full broadcasting mechanism to a function that only takes scalar values. Your quad qualifies as taking scalar inputs. But you are only iterating on one array, ws. The x that is passed on to your functions is generated by quad itself. quad and integrand are still Python functions, even if they use numpy operations.
cython improves low level iteration, stuff that it can convert to C code. Your primary iteration is at a high level, calling an imported function, quad. Cython can't touch or rewrite that.
You might be able to speed up integrand (and on down) with cython, but first focus on getting the most speed from that with regular numpy code.
def f(x, w):
if w < 0: return np.abs(x * w)
else: return np.exp(x) * w
With if w<0 w must be scalar. Can it be written so it works with an array w? If so, then
np.array([f(x, w) for w in ws]).sum()
could be rewritten as
fn(x, ws).sum()
Alternatively, since both x and w are scalar, you might get a bit of speed improvement by using math.exp etc instead of np.exp. Same for log and abs.
I'd try to write f(x,w) so it takes arrays for both x and w, returning a 2d result. If so, then temp and integrand would also work with arrays. Since quad feeds a scalar x, that may not help here, but with other integrators it could make a big difference.
If f(x,w) can be evaluated on a regular nx10 grid of x=np.linspace(-1,1,n) and ws, then an integral (of sorts) just requires a couple of summations over that space.
You can use quadpy for fully vectorized computation. You'll have to adapt your function to allow for vector inputs first, but that is done rather easily:
import numpy as np
import quadpy
np.random.seed(0)
ws = 2 * np.random.random(10) - 1
def f(x):
out = np.empty((len(ws), *x.shape))
out0 = np.abs(np.multiply.outer(ws, x))
out1 = np.multiply.outer(ws, np.exp(x))
out[ws < 0] = out0[ws < 0]
out[ws >= 0] = out1[ws >= 0]
return out
def integrand(x):
return f(x) * np.log(np.sum(f(x), axis=0))
val, err = quadpy.quad(integrand, -1, +1, epsabs=1.0e-10)
print(val)
[0.3266534 1.44001826 0.68767868 0.30035222 0.18011948 0.97630376
0.14724906 2.62169217 3.10276876 0.27499376]
Related
For time-series analysis, it's useful to have rolling PCA functions to analyse how the dynamics of the time-series changes over time to avoid look-ahead bias.
We may want to answer the question: 'how many principle components are needed to keep 90% of the variance?'. The number of principle components to explain 90% variance may change over time, depending on the dynamics of the time-series.
In addition, we may want to reduce the number of components p in a given dataset to k < p on a rolling basis to more easily visualise the data.
While scikit has a PCA module, it does not support rolling calculations. Similarly with numpy SVD. We could use these packages in a manual for loop, but for large arrays (>10,000 rows) it would become very slow.
Is there a fast rolling implementation of PCA in python to address some of the questions above?
While I didn't manage to find a rolling implementation of PCA, it is a relatively straightforward matter to use the packages and tools mentioned in the question to code a manual rolling PCA function. In addition, we will use numba to gain a small speed-up, as it supports numpy.linalg.svd and numpy.linalg.eig.
The code in this answer is inspired by the excellent explanations of PCA here and here
import numpy as np
from numpy.linalg import eig
from numba import njit
import numpy.typing as npt
#njit
def rolling_pca(
arr: npt.NDArray[np.float64],
n_components: int,
window: int,
min_periods: int
) -> npt.NDArray[np.float64]:
"""Perform PCA on the covariance matrix of arr.
Return the lower dimensional array.
Data is assumed to have non-zero mean, so will be demeaned
in the process.
Args:
arr: Input data. Shape (n_samples, n_variables).
n_components: Number of components to reduce data matrix to.
Must be less than arr.shape[1].
window: Sliding window size.
min_periods: Minimum number of observations required to perform calculation.
Returns:
Reduced data matrix. Shape (n_samples, n_components)
"""
# create a copy to ensure we don't change data in place
arr_copy = arr.copy()
n = arr_copy.shape[0]
# create an empty array which will be populated with the output
reduced_out = np.full((n, n_components), np.nan)
# iterate over each row (timestamp in a timeseries)
for i in range(min_periods, n + 1):
if i < window:
lookback = i
else:
lookback = window
start_idx = i - lookback
curr_arr = arr_copy[start_idx: i, :]
# demean returns
curr_arr = curr_arr - (np.sum(curr_arr, axis=0) / lookback)
# calculate the covariance matrix
cov = (curr_arr.T # curr_arr) / (lookback - 1)
# get the eigenvectors
# sort eigvals to get top largest corresponding eigenvectors
evals, evecs = eig(cov)
idx = np.argsort(evals)[:n_components]
evecs = evecs[:, idx]
# multiply the top eigenvectors by the current array to get a reduced matrix
reduced = (evecs.T # curr_arr.T).T
reduced_out[start_idx: i, :] = reduced
return reduced_out
After profiling the code, the two slowest parts are, as expected, the calls to eig() and the matrix multiplication curr_arr.T # curr_arr. As the array curr_arr is limited by the window size, a pure numpy (no numba) implementation of matrix multiplcation is faster than using numba. This is because the arrays used in the matrix multiplication are small, and are not contiguous (see this post for more details). I didn't get around to resolving this issue, but if anyone has any suggestions, it would speed up this function quite a bit more.
I've compared the average timings between 3 implementations to see the effect of the speedup that numba offers. The 3 implementations are:
Manual for loop using numba, exactly as above
Manual for loop without numba, but otherwise same as the code above
Manual for loop using Sklearn instead of numpy eig, no numba (as numba does not support Sklearn)
Note that the following parameters are fixed so we get as fair a comparison as possible between implementations
Window size = 120
Minimum number of periods = 22
Input data number of variables = 20
Number of components to reduce to via PCA = 10
Number of iterations to time function over so as to get an average timing per implementation = 10
Only the row size (number of samples) is allowed to vary so we can visualise how execution time varies with array length.
We can see for large arrays (100k rows), we can decrease the time from about 14.19s using Sklearn, to 5.36s using numba, about a 2.6X speedup.
PCA Reconstruction
We implement some code similar to what we used above to reconstruct the original data matrix, using only the top principle components. Namely, we use SVD to decompose the matrix X into 3 matrices, U, S, and V^T. With these matrices, we can calculate how much variance is kept cumulatively by the components, and only keep the top k components that explain a desired amount of variance.
import numpy as np
from numpy.linalg import svd
from numba import njit
import numpy.typing as npt
#njit
def __get_num_top_components(
singular_values: npt.NDArray[np.float64],
threshold: float,
n: int,
) -> int:
"""Get the number of top eigen-components required by threshold.
Args:
singular_values: Singular values from SVD.
threshold: Minimum amount of explained variance to be kept.
n: Number of samples in data matrix.
Returns:
Required number of components to keep.
"""
evals = singular_values ** 2 / (n - 1)
evals = evals / np.sum(evals)
cumsum_evals = np.cumsum(evals)
top_k = np.argwhere(cumsum_evals > threshold).min()
return top_k
#njit
def rolling_pca_reconstruction(
arr: npt.NDArray[np.float64],
threshold: float,
window: int,
min_periods: int
) -> npt.NDArray[np.float64]:
"""Perform PCA on arr and return reconstructed matrix.
This method follows the logic succinctly outlined here:
https://stats.stackexchange.com/a/134283/178320
Args:
arr: Input data. Shape (n_samples, n_variables).
threshold: Minimum amount of explained variance to be kept.
Must be a number in (0., 1.).
window: Sliding window size.
min_periods: Minimum number of observations required to perform calculation.
Returns:
Reconstructed data matrix. Shape (n_samples, n_variables)
"""
arr_copy = arr.copy()
n = arr_copy.shape[0]
p = arr_copy.shape[1]
recon_out = np.full((n, p), np.nan)
for i in range(min_periods, n + 1):
if i < window:
lookback = i
else:
lookback = window
start_idx = i - lookback
curr_arr = arr_copy[start_idx: i, :]
# demean data
curr_arr = curr_arr - (np.sum(curr_arr, axis=0) / lookback)
# perform SVD on data, no need for full matrices, this is faster
u, s, vh = svd(curr_arr, full_matrices=False)
# calculate the number of components that explains threshold variance
top_k = __get_num_top_components(
singular_values=s,
threshold=threshold,
n=lookback,
)
# reconstruct the data matrix using the top_k components
tmp_recon = u[:, :top_k] # np.diag(s)[:top_k, :top_k] # vh[:, :top_k].T
recon_out[start_idx: i, :] = tmp_recon
return recon_out
The output of rolling_pca_reconstruction() is the reconstructed data, of same dimension as the input data arr. One useful modification that could be made to this code is to record top_k at each iteration, to understand how many components are needed to explain treshold variance over time.
I created a typed memoryview in cython and would like to multiply it by a scalar:
import numpy as np
import math
cimport numpy as np
def foo():
N = 10
cdef np.double_t [:, :] A = np.ones(shape=(N,N),dtype=np.double_)
cdef int i,j
cdef double pi = math.pi
for i in range(N):
for j in range(N):
A[i,j] *= pi
return A
def bar():
N = 10
cdef np.double_t [:, :] A = np.ones(shape=(N,N),dtype=np.double_)
cdef double pi = math.pi
A *= pi
return A
Function foo() does this task but is not very convenient/readable.
The line A *= pi in function bar() does however not compile: Invalid operand types for '*' (double_t[:, :]; double).
Is there a way to perform such a broadcasting operation on a cython memoryview?
No, memoryviews don't do this. A memoryview is literally just a way to access individual elements of an array quickly. It has no really concept of the mathematical operations that can be performed on the array.
In the case of your bar function, any attempt to type it is probably actually going to make it worse (i.e. it'll spend extra time checking the type, but ultimately the work is done in ordinary calls to Numpy function).
There's a number of (not 100% satisfactory) ways of getting a Numpy array from a memoryview:
np.asarray(memview) - this should be done without copying (provided you aren't using the esoteric indirect memory layout). It might be worth adding an assertion to check that no copy was made though.
memview.base - be slightly careful with this. If the memoryview is a result of slicing then .base will be the original unsliced object.
Keep a parallel numpy array and memoryview variable:
Anp = np.array(...)
cdef double[:] Amview = Anp
because the memoryview is a view of some memory, modifications to the array will be reflected in the memoryview and vice-versa. (Reassigning the array variable, e.g. Anp = something_else, won't be reflected though).
In summary, memoryviews are designed for one main job: being able to access individual elements quickly. If that's not what you're doing then you probably don't want to use a memoryview.
I've been struggling to find a way to get this calc that works for a dask workflow.
I have code that uses np.random.mulivariate_normal function and while many of these types are available to us on dask array it seems this one it not. Sooo.... I attempted to create my own based on an example provided in the dask documentation.
Here is my attempt which is giving errors that I am having difficulty understanding. I also provided random input variables to make it easy to replicate:
import numpy as np
from dask.distributed import Client
import dask.array as da
def mvn(mu, sigma, n, blocksize):
chunks = ((blocksize,) * (n // blocksize),
(blocksize,) * (n // blocksize))
name = 'mvn' # unique identifier
dsk = {(name, i, j): (np.random.multivariate_normal(mu,sigma, blocksize))
if i == j else
(np.zeros, (blocksize, blocksize))
for i in range(n // blocksize)
for j in range(n // blocksize)}
dtype = np.random.multivariate_normal(0).dtype # take dtype default from numpy
return da.Array(dsk, name, chunks, dtype)
n = 10000
A = da.random.normal(0, 1, size=(n,n), chunks=(1000, 1000))
sigma = da.dot(A,A.transpose())
mu = 4.0*da.ones(n, chunks = 1000)
R = da.numpy.random.mvn(mu, sigma, n, chunks=(100))
Any suggestions or am I so far off the mark here that I should abandon all hope? Thanks!
If you have a cluster to run this on, you can use my answer from this post, copied here for refrence:
An work arround for now, is to use a cholesky decomposition. Note that any covariance matrix C can be expressed as C=G*G'. It then follows that x = G'*y is correlated as specified in C if y is standard normal (see this excellent post on StackExchange Mathematic). In code:
Numpy
n_dim =4
size = 100000
A = np.random.randn(n_dim, n_dim)
covm = A.dot(A.T)
x= np.random.multivariate_normal(size=size, mean=np.zeros(len(covm)),cov=covm)
## verify numpys covariance is correct
np.cov(x, rowvar=False)
covm
Dask
## create covariance matrix
A = da.random.standard_normal(size=(n_dim, n_dim),chunks=(2,2))
covm = A.dot(A.T)
## get cholesky decomp
L = da.linalg.cholesky(covm, lower=True)
## drawn standard normal
sn= da.random.standard_normal(size=(size, n_dim),chunks=(100,100))
## correct for correlation
x =L.dot(sn.T)
x.shape
## verify
covm.compute()
da.cov(x, rowvar=True).compute()
This answer can be fleshed out, but I imagine you would have an easier time using dask's delayed, da.from_delayed and da.*stack.
One immediate problem I see with what you have: with np.random.multivariate_normal(mu,sigma, blocksize) you are directly calling the function, instead of making the spec. You probably wanted (np.random.multivariate_normal, mu,sigma, blocksize). This shows that working with raw dask dictionaries can be tricky!
I'm having trouble solving a discrepancy between something breaking at runtime, but using the exact same data and operations in the python console, having it work fine.
# f_err - currently has value 1.11819388872025
# l_scales - currently a numpy array [1.17840183376334 1.13456764589809]
sq_euc_dists = self.se_term(x1, x2, l_scales) # this is fine. It calls cdists on x1/l_scales, x2/l_scales vectors
return (f_err**2) * np.exp(-0.5 * sq_euc_dists) # <-- errors on this line
The error that I get is
AttributeError: 'Zero' object has no attribute 'exp'
However, calling those exact same lines, with the same f_err, l_scales, and x1, x2 in the console right after it errors out, somehow does not produce errors.
I was not able to find a post referring to the 'Zero' object error specifically, and the non-'Zero' ones I found didn't seem to apply to my case here.
EDIT: It was a bit lacking in info, so here's an actual (extracted) runnable example with sample data I took straight out of a failed run, which when run in isolation works fine/I can't reproduce the error except in runtime.
Note that the sqeucld_dist function below is quite bad and I should be using scipy's cdist instead. However, because I'm using sympy's symbols for matrix elementwise gradients with over 15 partial derivatives in my real data, cdist is not an option as it doesn't deal with arbitrary objects.
import numpy as np
def se_term(x1, x2, l):
return sqeucl_dist(x1/l, x2/l)
def sqeucl_dist(x, xs):
return np.sum([(i-j)**2 for i in x for j in xs], axis=1).reshape(x.shape[0], xs.shape[0])
x = np.array([[-0.29932052, 0.40997373], [0.40203481, 2.19895326], [-0.37679417, -1.11028267], [-2.53012051, 1.09819485], [0.59390005, 0.9735], [0.78276777, -1.18787904], [-0.9300892, 1.18802775], [0.44852545, -1.57954101], [1.33285028, -0.58594779], [0.7401607, 2.69842268], [-2.04258086, 0.43581565], [0.17353396, -1.34430191], [0.97214259, -1.29342284], [-0.11103534, -0.15112815], [0.41541759, -1.51803154], [-0.59852383, 0.78442389], [2.01323359, -0.85283772], [-0.14074266, -0.63457529], [-0.49504797, -1.06690869], [-0.18028754, -0.70835799], [-1.3794126, 0.20592016], [-0.49685373, -1.46109525], [-1.41276934, -0.66472598], [-1.44173868, 0.42678815], [0.64623684, 1.19927771], [-0.5945761, -0.10417961]])
f_err = 1.11466725760716
l = [1.18388412685279, 1.02290811104357]
result = (f_err**2) * np.exp(-0.5 * se_term(x, x, l)) # This runs fine, but fails with the exact same calls and data during runtime
Any help greatly appreciated!
Here is how to reproduce the error you are seeing:
import sympy
import numpy
zero = sympy.sympify('0')
numpy.exp(zero)
You will see the same exception you are seeing.
You can fix this (inefficiently) by changing your code to the following to make things floating point.
def sqeucl_dist(x, xs):
return np.sum([np.vectorize(float)(i-j)**2 for i in x for j in xs],
axis=1).reshape(x.shape[0], xs.shape[0])
It will be better to fix your gradient function using lambdify.
Here's an example of how lambdify can be used on partial d
from sympy.abc import x, y, z
expression = x**2 + sympy.sin(y) + z
derivatives = [expression.diff(var, 1) for var in [x, y, z]]
derivatives is now [2*x, cos(y), 1], a list of Sympy expressions. To create a function which will evaluate this numerically at a particular set of values, we use lambdify as follows (passing 'numpy' as an argument like that means to use numpy.cos rather than sympy.cos):
derivative_calc = sympy.lambdify((x, y, z), derivatives, 'numpy')
Now derivative_calc(1, 2, 3) will return [2, -0.41614683654714241, 1]. These are ints and numpy.float64s.
A side note: np.exp(M) will calculate the element-wise exponent of each of the elements of M. If you are trying to do a matrix exponential, you need np.linalg.exmp.
I'm having a bit of trouble with fitting a curve to some data, but can't work out where I am going wrong.
In the past I have done this with numpy.linalg.lstsq for exponential functions and scipy.optimize.curve_fit for sigmoid functions. This time I wished to create a script that would let me specify various functions, determine parameters and test their fit against the data. While doing this I noticed that Scipy leastsq and Numpy lstsq seem to provide different answers for the same set of data and the same function. The function is simply y = e^(l*x) and is constrained such that y=1 at x=0.
Excel trend line agrees with the Numpy lstsq result, but as Scipy leastsq is able to take any function, it would be good to work out what the problem is.
import scipy.optimize as optimize
import numpy as np
import matplotlib.pyplot as plt
## Sampled data
x = np.array([0, 14, 37, 975, 2013, 2095, 2147])
y = np.array([1.0, 0.764317544, 0.647136491, 0.070803763, 0.003630962, 0.001485394, 0.000495131])
# function
fp = lambda p, x: np.exp(p*x)
# error function
e = lambda p, x, y: (fp(p, x) - y)
# using scipy least squares
l1, s = optimize.leastsq(e, -0.004, args=(x,y))
print l1
# [-0.0132281]
# using numpy least squares
l2 = np.linalg.lstsq(np.vstack([x, np.zeros(len(x))]).T,np.log(y))[0][0]
print l2
# -0.00313461628963 (same answer as Excel trend line)
# smooth x for plotting
x_ = np.arange(0, x[-1], 0.2)
plt.figure()
plt.plot(x, y, 'rx', x_, fp(l1, x_), 'b-', x_, fp(l2, x_), 'g-')
plt.show()
Edit - additional information
The MWE above includes a small sample of the dataset. When fitting the actual data the scipy.optimize.curve_fit curve presents an R^2 of 0.82, while the numpy.linalg.lstsq curve, which is the same as that calculated by Excel, has an R^2 of 0.41.
You are minimizing different error functions.
When you use numpy.linalg.lstsq, the error function being minimized is
np.sum((np.log(y) - p * x)**2)
while scipy.optimize.leastsq minimizes the function
np.sum((y - np.exp(p * x))**2)
The first case requires a linear dependency between the dependent and independent variables, but the solution is known analitically, while the second can handle any dependency, but relies on an iterative method.
On a separate note, I cannot test it right now, but when using numpy.linalg.lstsq, I you don't need to vstack a row of zeros, the following works as well:
l2 = np.linalg.lstsq(x[:, None], np.log(y))[0][0]
To expound a bit on Jaime's point, any non-linear transformation of the data will lead to a different error function and hence to different solutions. These will lead to different confidence intervals for the fitting parameters. So you have three possible criteria to use to make a decision: which error you want to minimize, which parameters you want more confidence in, and finally, if you are using the fitting to predict some value, which method yields less error in the interesting predicted value. Playing around a bit analytically and in Excel suggests that different kinds of noise in the data (e.g. if the noise function scales the amplitude, affects the time-constant or is additive) leads to different choices of solution.
I'll also add that while this trick "works" for exponential decay to 0, it can't be used in the more general (and common) case of damped exponentials (rising or falling) to values that cannot be assumed to be 0.