when to use numpy vs statistics modules - python

While working on some statistical analysis tools, I discovered there are at least 3 Python methods to calculate mean and standard deviation (not counting the "roll your own" techniques):
np.mean(), np.std() (with ddof=0 or 1)
statistics.mean(), statistics.pstdev() (and/or statistics.stdev)
scipy.statistics package
That has me scratching my head. There should be one obvious way to do it, right? :-) I've found some older SO posts. One compares the performance advantages of np.mean() vs statistics.mean(). It also highlights differences in the sum operator. That post is here:
why-is-statistics-mean-so-slow
I am working with numpy array data, and my values fall in a small range (-1.0 to 1.0, or 0.0 to 10.0), so the numpy functions seem the obvious answer for my application. They have a good balance of speed, accuracy, and ease of implementation for the data I will be processing.
It appears the statistics module is primarily for those that have data in lists (or other forms), or for widely varying ranges [1e+5, 1.0, 1e-5]. Is that still a fair statement? Are there any numpy enhancements that address the differences in the sum operator? Do recent developments bring any other advantages?
Numerical algorithms generally have positive and negative aspects: some are faster, or more accurate, or require a smaller memory footprint. When faced with a choice of 3-4 ways to do a calculation, a developer's responsibility is to select the "best" method for his/her application. Generally this is a balancing act between competing priorities and resources.
My intent is to solicit replies from programmers experienced in statistical analysis to provide insights into the strengths and weaknesses of the methods above (or other/better methods). [I'm not interested in speculation or opinions without supporting facts.] I will make my own decision based on my design requirements.

Why does NumPy duplicate features of SciPy?
From the SciPy FAQ What is the difference between NumPy and SciPy?:
In an ideal world, NumPy would contain nothing but the array data type and the most basic operations: indexing, sorting, reshaping, basic elementwise functions, etc. All numerical code would reside in SciPy. However, one of NumPy’s important goals is compatibility, so NumPy tries to retain all features supported by either of its predecessors.
It recommends using SciPy over NumPy:
In any case, SciPy contains more fully-featured versions of the linear algebra modules, as well as many other numerical algorithms. If you are doing scientific computing with Python, you should probably install both NumPy and SciPy. Most new features belong in SciPy rather than NumPy.
When should I use the statistics library?
From the statistics library documentation:
The module is not intended to be a competitor to third-party libraries such as NumPy, SciPy, or proprietary full-featured statistics packages aimed at professional statisticians such as Minitab, SAS and Matlab. It is aimed at the level of graphing and scientific calculators.
Thus I would not use it for serious (i.e. resource intensive) computation.
What is the difference between statsmodels and SciPy?
From the statsmodels about page:
The models module of scipy.stats was originally written by Jonathan Taylor. For some time it was part of scipy but was later removed. During the Google Summer of Code 2009, statsmodels was corrected, tested, improved and released as a new package. Since then, the statsmodels development team has continued to add new models, plotting tools, and statistical methods.
Thus you may have a requirement that SciPy is not able to fulfill, or is better fulfilled by a dedicated library.
For example the SciPy documentation for scipy.stats.probplot notes that
Statsmodels has more extensive functionality of this type, see statsmodels.api.ProbPlot.
Thus in cases like these you will need to turn to statistical libraries beyond SciPy.

Just a simple benchmark.
statistics vs NumPy
import statistics as stat
import numpy as np
from timeit import default_timer as timer
l = list(range(1_000_000))
start = timer()
m, std = stat.mean(l), stat.stdev(l)
end = timer()
print(end-start, "sec elapsed.")
print("mean, std:", m, std)
start = timer()
m, std = np.mean(l), np.std(l)
end = timer()
print(end-start, "sec elapsed.")
print("mean, std:", m, std)
(no iteration for brevity.)
Result
NumPy is way faster.
# statistics
3.603845 sec elapsed.
mean, std: 499999.5 288675.2789323441
# numpy
0.2315261999999998 sec elapsed.
mean, std: 499999.5 288675.1345946685

Related

Calculate IRR in Python

I am running into a roadblock and would appreciate some help on this.
Problem Statement:
I am trying to calculate XIRR for a cash flow over 30 years in Python.
What have I tried so far:
However, none of the established libraries(like numpy and pandas) seem to have support for this. After doing some research, I learned through this source (https://vindeep.com/Corporate/XIRRCalculation.aspx) that with some simple manipulation, XIRR can be calculated from IRR.
So, all I need is an IRR function that is implemented well. The functionality used to exist in numpy but has moved to this other package (https://github.com/numpy/numpy-financial). While, this package works, it is very very slow. Here is a small test:
import pandas as pd
import numpy as np
import numpy_financial as npf
from time import time
# Generate some example data
t = pd.date_range('2022-01-01', '2037-01-01', freq='D')
cash_flows = np.random.randint(10000, size=len(t)-1)
cash_flows = np.insert(cash_flows, 0, -10000)
# Calculate IRR
start_timer = time()
npf.irr(cash_flows, guess)
stop_timer = time()
print(f"""Time taken to calculate IRR over 30 years of daily data: {round((stop_timer-start_timer)/60, 2)}""")
One other alternative seems to be https://github.com/better/irr - however, this has an edge case bug that has not been addressed in over 4 years.
Can anyone kindly offer to a more stable implementation. It feels like such simple and very commonly used functionality and the lack of a good stable implementation surprises me. Can someone point to any good resources.
Thanks
Uday
Try using pyxirr package. Implemented in Rust, it is blazing fast. For 30 years time period it took about .001 sec.
pyxirr creator here. The library has been used in a financial project for over a year, but I only recently found the time to publish it. We had the task of quickly calculating XIRR for various portfolios and existing implementations quickly became a bottleneck. pyxirr also mimics some numpy-financial functions and works much faster.
The XIRR implementation in Excel is not always correct. In edge cases the algorithm does not converge and shows incorrect result instead of error or NA. The result can be checked with the xnpv function: xnpv(xirr_rate, dates, values) and should be close to zero. Similarly, you can check irr using the npv function: npv(irr_rate, values), but note the difference in npv calculation between Excel and numpy-financial.
Taking a look at the implementation on their GitHub, it is pretty evident to me that the npf.irr() function is implemented pretty well. Your alternative seems to be to implement the function yourself using NumPy operations but I am doubtful that a) that is easy to accomplish or b) possible to accomplish in pure Python.
NumPy Financial seems to be doing their implementation using eigenvalues which means they are performing complex mathematic operations. Perhaps, if you are not bounded to Python, consider Microsoft's C# implementation of IRR and see if that works faster. I suspect that they are using regression to calculate the IRR. Therefore, based on your guess, it may indeed be quicker than NumPy Financial.
Your final alternative is to continue with what you have at the moment and just run on a more powerful machine. On my machine, this operation took about 71 seconds and it is does not even have a GPU. I am sure more powerful computers, with parallelization, should be able to compute this much much faster than that.
Look at the answer I provided here: https://stackoverflow.com/a/66069439/4045275.
I didn't benchmark it against pyxirr

Maximum Likelihood Estimation with statsmodels overcomplicates things? Hoping for Recommendations

After taking a couple advanced statistics courses, I decided to code some functions/classes to just automate estimating parameters for different distributions via MLE. In Matlab, the below is something I easily coded once:
function [ params, max, confidence_interval ] = newroutine( fun, data, guesses )
lh = #(x,data) -sum(log(fun(x,data))); %Gets log-likelihood from user-defined fun.
options = optimset('Display', 'off', 'MaxIter', 1000000, 'TolX', 10^-20, 'TolFun', 10^-20);
[theta, max1] = fminunc(#(x) lh(x,data), guesses,options);
params = theta
max = max1
end
Where I just have to correctly specify the underlying pdf equation as fun, and with more code I can calculate p-values, confidence-intervals, etc.
With Python, however, all the sources I've found on MLE automation (for ex., here and here) insist that the easiest way to do this is to delve into OOP using a subclass of statsmodel's, GenericLikelihoodModel, which seems way too complicated for me. My reasoning is that, since the log-likelihood can be automatically created from the pdf (at least for the vast majority of functions), and scipy.stats."random_dist".fit() already easily returns MLE estimates, it seems ridiculous to have to write ~30 lines of class code each time you have a new dist. to fit.
I realize that doing it the way the two links suggests allows you to automatically tap into statsmodel's functions, but it honestly does not seem simpler than tapping into scipy oneself and writing much simpler functions.
Am I missing an easier way to perform basic MLE, or is there a real good reason for the way statsmodels does this?
I wrote the first post outlining the various methods, and I think it is fair to say that while I recommend the statsmodels approach, I did so to leverage the postestimation tools it provides and to get standard errors every time a model is estimated.
When using minimize, the python equivalent of fminunc (as you outline in your example), oftentimes I am forced to use "Nelder-Meade" or some other gradiant-free method to get convergence . Since I need standard errors for statistical inference, this entails an additional step using numdifftools to recover the hessian. So in the end, the method you propose has its complications too (for my work). If all you care about is the maximum likelihood estimate and not inference, then the approach you outline is probably best and you are correct that you don't need the machinery of statsmodel.
FYI: in a later post, I use your approach combined with autograd for significant speedups of big maximum likelihood models. I haven't successfully gotten this to work with statsmodels.

convert a sparse matrix to dense and get the full eigenvalues

recently I'm working on a problem which requires
diagonalizing a huge hermitian matrix to get all the eigenvalues.
Currently I'm using Mathematica to do the job.
However it is not applicable due to the limitation of memory
when the matrix size approaches (2^15,2^15), where the diagonalization costs approximately 32 GBs memory.
I've tried using python by importing the matrix from mathematica,
import numpy as np
from scipy.io import mmread
from scipy.sparse import csc_matrix
#importing sparse matrix to save space
h = mmread("h.mtx")
h = csc_matrix(h)
#diagonlizing the dense one
ev = np.linalg.eigvalsh(h.todense())
It works but unfortunately an order of magnitude slower than Mathematica.
So, is there any other possible solutions, say, C++?
I know nothing about C++ so I guess the simplest way may be importing the
matrix to C++ and diagonalizing.
Thanks!
Running some preliminary test using this matrix:
http://math.nist.gov/MatrixMarket/data/NEP/h2plus/qc2534.html
I determined that the conversion to dense does not take up much of the time. The eigenvalue calculation does.
Numpy uses highly-optimized Lapack routines to calculate. These are the same you'd use in C++. Therefore C++ won't give you much of a speedup. If you want a speedup use the sparseness as a property, go to a better computer or switch to a distributed matrix storage(lot's of labor here).
P.S: if you do this for a university project you might want to look around if your university has a cluster of some sort. A cluster node typically has lots of memory. If not, check amazons AWS EC2 or googles compute engine for instances with lot's of ram.
Edit:
Here Wolfram says what Mathematica does behind the scenes: http://reference.wolfram.com/language/tutorial/LinearAlgebraAppendix.html#83486633
Arpack is a (arnoldi)subspace solver, giving you only the highest or lowest k-eigenvalues, ATLAS is just a Lapack implementation and the rest seems to be for solving linear systems.
All methods giving you the full eigenspectrum will require the matrix decomposition of a NxN matrix. If you only want k vectors there are methods which reduce it to a decomposition of a k x k-matrix.
There are modern alternatives to Arpack(http://slepc.upv.es/ or the one that comes with MKL), but they all give you a subspace.
c++ won't help much.
In python you can delegate easily to C++ and a lot of scipy routines will do just that (for performance). I also expect that if you only time the eigen value line you will get similar performance to Matematica and the difference in performance comes from reading the data.
The best solution is to look for a more appropriate algorithm, maybe something that operates on the sparse matrix directly, or decompose the original into smaller matrices and combine them.
To make the original solution more tractable you could try increasing the amount of swap space. In linux it's a dedicated partition, in windows it's a setting. This should allow Matematica/python to use more memory, but it's going to be much slower due to memory trashing. Get an SSD to speed this setup up, but note that it's going to be destroyed faster due to often writes. Or even better buy more RAM.

python module for trace/chromatogram analysis

Is there a python module that integrates simple chromatogram/trace analysis algorithms? I am looking for baseline correction, peak detection and peak integration functionality for simple time-courses (with data stored in numpy arrays).
I spent quite some time searching now and there doesn't seem to be any which really surprises me.
I'm not sure what analysis you are conducting but have you looked at PyLSS ?
It can (and I quote from the documentation):
PyLSS is able to compute:
LSS parameters (log kw and S)
Build and plot chromatograms from experimental/predicted retention times
Regarding peak detection and peak integration, I have used the functionality of the ptp() method in the Numpy module for this and I find that it is pretty powerful. Would this satisfy your requirement?

Numpy for R user?

long-time R and Python user here. I use R for my daily data analysis and Python for tasks heavier on text processing and shell-scripting. I am working with increasingly large data sets, and these files are often in binary or text files when I get them. The type of things I do normally is to apply statistical/machine learning algorithms and create statistical graphics in most cases. I use R with SQLite sometimes and write C for iteration-intensive tasks; before looking into Hadoop, I am considering investing some time in NumPy/Scipy because I've heard it has better memory management [and the transition to Numpy/Scipy for one with my background seems not that big] - I wonder if anyone has experience using the two and could comment on the improvements in this area, and if there are idioms in Numpy that deal with this issue. (I'm also aware of Rpy2 but wondering if Numpy/Scipy can handle most of my needs). Thanks -
R's strength when looking for an environment to do machine learning and statistics is most certainly the diversity of its libraries. To my knowledge, SciPy + SciKits cannot be a replacement for CRAN.
Regarding memory usage, R is using a pass-by-value paradigm while Python is using pass-by-reference. Pass-by-value can lead to more "intuitive" code, pass-by-reference can help optimize memory usage. Numpy also allows to have "views" on arrays (kind of subarrays without a copy being made).
Regarding speed, pure Python is faster than pure R for accessing individual elements in an array, but this advantage disappears when dealing with numpy arrays (benchmark). Fortunately, Cython lets one get serious speed improvements easily.
If working with Big Data, I find the support for storage-based arrays better with Python (HDF5).
I am not sure you should ditch one for the other but rpy2 can help you explore your options about a possible transition (arrays can be shuttled between R and Numpy without a copy being made).
I use NumPy daily and R nearly so.
For heavy number crunching, i prefer NumPy to R by a large margin (including R packages, like 'Matrix') I find the syntax cleaner, the function set larger, and computation is quicker (although i don't find R slow by any means). NumPy's Broadcasting functionality for instance, i do not think has an analog in R.
For instance, to read in a data set from a csv file and 'normalize' it for input to an ML algorithm (e.g., mean center then re-scale each dimension) requires just this:
data = NP.loadtxt(data1, delimiter=",") # 'data' is a NumPy array
data -= NP.mean(data, axis=0)
data /= NP.max(data, axis=0)
Also, i find that when coding ML algorithms, i need data structures that i can operate on element-wise and that also understand linear algebra (e.g., matrix multiplication, transpose, etc.). NumPy gets this and allows you to create these hybrid structures easily (no operator overloading or subclassing, etc.).
You won't be disappointed by NumPy/SciPy, more likely you'll be amazed.
So, a few recommendations--in general and in particular, given the facts in your question:
install both NumPy and Scipy. As a rough guide, NumPy provides the
core data structures (in particular
the ndarray) and SciPy (which is
actually several times larger than
NumPy) provides the domain-specific
functions (e.g., statistics, signal
processing, integration).
install the repository versions, particularly w/r/t NumPy because the
dev version is 2.0. Matplotlib and NumPy are tightly integrated, you can use one without the other of course, but both are the best in their respective class among python libraries. You can get all three via easy_install, which i assume you already.
NumPy/SciPy have several modules
specifically directed to Machine
Learning/Statistics, including the Clustering package and the Statistics package.
As well as packages directed to
general computation, but which are
make coding ML algorithms a lot
faster, in particular,
Optimization and Linear Algebra.
There are also the SciKits, not included in the base NumPy or
SciPy libraries; you need to install them separately.
Generally speaking, each SciKit is a
set of convenience wrappers to
streamline coding in a given domain. The SciKits you are likely to find most relevant are: ann (approximate Nearest Neighbor), and learn (a set of ML/Statistics regression and classification algorithms, e.g., Logistic Regression, Multi-Layer Perceptron, Support Vector Machine).

Categories