linregmc 0.1.1__py3-none-any.whl

linregmc/.bashlog

@@ -0,0 +1 @@
+  521  cd linregmc_python2025/linregmc/src/linregmc/

linregmc/__init__.py

@@ -0,0 +1 @@
+from .linregmc import *

linregmc/linregmc.py

@@ -0,0 +1,345 @@
+import numpy as np
+from numpy.matlib import repmat,randn
+import logging
+
+def addnoise(yinp,ysiginp,nmc=10000,distrib='normal'):	
+	"""Adds noise to an array of data points (or a single value).
+	
+	Creates a matrix of nmc vectors with the mean values of y but with
+	added random noise of standard deviation ysig. 
+	You might want to initialize the random number generator in forehand.
+	
+	
+	Parameters
+	----------
+	y : array
+	    data vector
+	ysig : array
+	    standard deviation vector (same length as y, or a single value)
+	nmc : int, default 10000
+	    number of Monte Carlo copies
+	distrib : str, default 'normal'
+	    'norm'/'normal' gives normal distribution
+	    'lognorm'/'lognormal' give lognormal distribution (useful for example if negative results are unphysical)
+	
+	Returns
+	-------
+	array
+		the data vector with added noise
+		(always two-dimensional, but only one column if y was a single value)
+	
+	Examples
+	--------
+	>>> y = np.array([1.2, 2.3, 3.7])
+	>>> ysig = np.array([0.1, 0.3, 0.2])
+	>>> y1 = addnoise(y, ysig)  # different stdev for the three points, normal distribution
+	>>> y1.shape  # (10000, 3)
+	>>> y2 = addnoise(y, 0.2, 1000, distrib='lognorm')   # same stdev for the three points, lognormal distribution
+	>>> y2.shape  # (1000, 3)
+	>>> y3 = addnoise(1.5, 0.2) # only one point
+	>>> y3.shape  # (10000, 1)   
+	"""
+	
+	
+	yinp = np.asarray(yinp)
+	ysiginp = np.asarray(ysiginp)
+	if np.ndim(yinp)>1 or np.ndim(ysiginp)>1:
+		raise Exception('y and ysig must not have higher dimension than 1.')
+	if np.size(ysiginp) == 1:
+		ysiginp = ysiginp*np.ones(np.size(yinp))  #If ysiginp is a scalar, turn it into a vector with identical elements
+	if np.size(yinp) != np.size(ysiginp):
+		raise Exception('y and ysig must have the same length.')
+
+	n=np.size(yinp)
+	y=yinp.reshape((1,n))
+	ysig=ysiginp.reshape((1,n))
+	if distrib.lower() in ('norm' ,'normal'):
+		
+		return np.array(repmat(y,nmc,1)) + np.array(repmat(ysig,nmc,1))*np.array(randn(nmc,n))
+	elif  distrib.lower() in ('lognorm','lognormal'):
+			mu = np.log(y**2/np.sqrt(ysig**2+y**2))  # mu of lognormal dist
+			sigma = np.sqrt(np.log(ysig**2/y**2+1))  # sigma of lognormal dist
+			return np.exp(np.array(randn(nmc,n))*np.array(repmat(sigma,nmc,1)) + np.array(repmat(mu,nmc,1)))
+	else:
+		raise Exception('Distribution named "' + distrib + '" is not recognized.')
+
+
+
+def linreg(xinp, yinp, ndeg=1, fitfunc = None, weighted=True, plot = False):
+	"""Performs linear fitting ax+b=y with error analysis using a Monte Carlo approach.
+	
+	Parameters
+	----------
+	xinp : array
+	    an NM x N matrix: the NX data sets of x values (N data points)
+	yinp : array
+	    an NY x N matrix: the NY data sets of y values (N data points)
+		NX and NY need not be the same. In particular one may use a
+		single data set (without added noise) for one of them.
+		The number of fits equals NM = max(NX,NY) and if there are less data
+		sets for one of x or y, they are just cyclically reused.
+	ndeg : int, default 1
+		the degree of the polynomial used for fitting
+		(the ordering of the obtained coefficients is from 0 to ndeg, i.e.
+		corresponds to numpy.polynomial.polynomial.polyfit)
+	fitfunc : list, optional
+		a list of functions of x that are used as basis functions instead of a polonymial
+		(ndeg and fitfunc cannot both be specified)
+	weighted: boolean, default True
+		weight the importance of each data point by 1/stdev (recommended)
+	plot : boolean, default False
+	    an optional argument that specifies whether to plot the chi2 distribution
+	    to visualize the "goodness-of-fit".
+	
+	Returns
+	-------
+	pp  : array (ndeg+1 elements)
+		single-fit value of each parameter (can be used as the result)
+	psig  : array (ndeg+1 elements)
+		standard deviation of each parameter
+	pchi2 : float
+		goodness-of-fit, i.e. probability of chi>chi0
+		Note: not well-defined
+	pmc  : array
+		a (NM x (ndeg+1) matrix, the fitted parameters for all data sets
+		
+	Examples
+	--------
+	>>> x = np.array([0.1, 0.2, 0.3])
+	>>> y = np.array([1.2, 2.3, 3.7])
+	>>> y_mc=addnoise(y, 0.1)
+	>>> pp,psig,pchi2,pmc = linreg(x, y_mc)	
+	>>> print(pp)   #[-0.102 12.517]
+	>>> print(pmc.shape)   # (10000, 2)	
+	>>> pp,psig,pchi2,pmc = linreg(x, y_mc, fitfunc=[lambda x: np.exp(x), lambda x: 1]) #Fit to y=ae^x+b instead	
+	"""
+
+	if fitfunc != None and ndeg==1:  #Assume that ndeg was not specified, there is no way to know...
+		ndeg = len(fitfunc)-1
+	elif fitfunc != None and ndeg!=1: 
+		raise Exception('Only one of ndeg and fitfunc can be specified.')
+	if np.ndim(xinp) == 1:
+		x=xinp.reshape((1,np.size(xinp)))
+	else:
+		x= xinp
+	if np.ndim(yinp) == 1:
+		y=yinp.reshape((1,np.size(yinp)))
+	else:
+		y=yinp
+	if np.size(x,1) != np.size(y,1):
+		raise Exception('Number of columns in x and y must be equal')
+	N=np.size(x,1)
+	xn=np.size(x,0)
+	yn=np.size(y,0)
+
+
+	def buildmat(xx,ss=1.0):
+		if fitfunc==None:
+			return np.stack([np.ones(N)/ss]+[xx**k/ss for k in range(1,ndeg+1)], axis=1)
+		else:
+			return np.stack([(np.zeros(N)+func(xx))/ss for func in fitfunc], axis=1)
+
+    
+	xs=np.median(x, axis=0)
+	ys=np.median(y, axis=0)   #Reproduces original data points independent of distribution
+	if weighted:
+		deltax=np.std(xs)/(N*100)   #gives deltax with correct order of magnitude
+		pnow=np.linalg.lstsq(buildmat(xs), ys, rcond=None)[0]   #parameters if no weights are used
+		fprime=(np.matmul(buildmat(xs+deltax),pnow)-np.matmul(buildmat(xs-deltax),pnow))/(2*deltax)  #numerical derivative
+		sig=np.sqrt(np.var(y, axis=0)+fprime**2*np.var(x, axis=0))  #Standard error propagation
+		#TODO: If weighting is important, the estimation of sig should be done iteratively because fprime depends on the fit and thus on sig
+		if xn==1 and yn==1:
+			sig=1.0
+			logging.warning('Single data sets, using unweighted fit instead')
+		elif np.any(sig==0):
+			sig=1.0
+			logging.warning('Points with no variation encountered, using unweighted fit instead')
+	else:
+		sig=1.0
+		
+	#Perform single fit to get the base chi2 value
+	Xt=buildmat(xs)
+	X=buildmat(xs,sig)
+	YS=ys/sig
+	pp=np.linalg.lstsq(X,YS, rcond=None)[0]
+	yfit=np.matmul(Xt,pp)  # y(xs) value according to model   
+	chi2 = sum((YS - np.matmul(X,pp))**2)
+
+	nmc = max(xn,yn)
+	pmc = np.zeros((nmc,ndeg+1)) 
+	chi2mc = np.zeros(nmc)
+	for i in range(nmc):
+		X=buildmat(x[i%xn,:],sig)
+		Y=(yfit+y[i%yn,:]-ys)/sig   
+		p=np.linalg.lstsq(X,Y, rcond=None)[0] 
+		pmc[i,:]=p
+		chi2mc[i] = sum((Y - np.matmul(X,p))**2)
+
+	psig = np.std(pmc,0)
+	pmean = np.mean(pmc,0) #Not used
+	pchi2=sum(chi2mc>chi2)/nmc     #Percentage of MC samples having greater chi2 than the observation
+	
+	if plot:
+		import matplotlib.pyplot as plt
+		fig, ax = plt.subplots(1, 1, figsize=(4, 2))
+		counts,*_=ax.hist(chi2mc,bins=50)
+		ycent=0.5*max(counts)
+		ax.plot([chi2,chi2],[0,ycent],'r-')
+		ax.set_yticks([])
+		ax.set_xlabel(r"$\chi^2$")
+		plt.show()
+		
+	return (pp,psig,pchi2,pmc)
+
+
+
+def confidence(X, level=0.683, plot=False):
+	"""Statistical analysis of the data in matrix X.
+	
+	It is assumed that the number of data points are large; all properties
+	are calculated from the data itself.
+	
+	Parameters
+	----------
+	X : array
+	    data matrix. Data in columns. For example, if X contains data
+	    from two measurements, data for measurement 1 is in column 1
+	    and measurement 2 in columns 2.
+	    If only one column, a 1d-array is also acceptable
+	level : float, default 0.683
+	    desired confidence level
+	plot : boolean, default False
+	    an optional boolean specifying whether to plot histograms for each column
+	    where a general statistic is shown as a red errorbar (median +/- stdev)
+	    and the confidence intervals are shown with black lines.
+	    The red markers at the bottom show the simpler (median +/- err) 
+	    interval which should normally coincide with the confidence interval
+	    unless the distribution is skew (in which case the confidence interval is more reliable).
+	    If X has exactly two columns, a scatter plot showing possible correlation between
+	    the two columns is also produced.
+	
+	Returns
+	-------
+	err : float
+	    estimated error in the columns based on selected confidence level.
+	confint : list
+	    a list of tuples (low, high), the confidence interval for each input column
+	    (pconf*100% of values are found within this interval around median) 
+	    If the input x was a 1d-array, a single tuple is returned instead of a list
+
+	Examples
+	--------
+	>>> x = np.array([0.1, 0.2, 0.3])
+	>>> y = np.array([1.2, 2.3, 3.7])
+	>>> y_mc=addnoise(y, 0.1)
+	>>> pp,psig,pchi2,pmc = linreg(x, y_mc)
+	>>> err,confint = confidence(pmc, 0.95)
+	>>> print(err)    # gives [1.364 0.296]
+	>>> print(confint)    # gives [(11.16, 13.88), (-0.401, 0.190)]
+	
+	"""
+
+	onedim = (np.ndim(X) == 1)
+	
+	if onedim:  #convert to matrix, then convert back to onedim at the end
+		X=X.reshape((np.size(X),1))
+	
+	if level <= 0 or level >= 1:
+		raise Exception("levvel must be 0 < level < 1.")
+	
+	if np.size(X,1) > np.size(X,0):
+		print("Warning. It appears that your data is not placed column-wise.")
+	
+	N = np.size(X,0) #number of data points
+	n = np.size(X,1)  #number of dimensions (columns)
+	
+	# GUM recommendation. ncut is the complement to pconf, ie the 1-pconf
+	# fraction of points.
+	#ncut = floor((N - floor(level*N+0.5) + 1)/2);   
+	
+	median = np.median(X,0)
+	sig = np.std(X,0)
+	absdiff = abs(X-np.mean(X,0)) #Absolute difference to mean value
+	plow = np.zeros(n)
+	phigh = np.zeros(n)
+	err = np.zeros(n)
+	for j in range(n):
+		tmp=np.sort(X[:,j])
+		plow[j]=tmp[round(max(1,0.5*(1-level)*N))-1]
+		phigh[j]=tmp[round(min(N,1-0.5*(1-level)*N))-1]
+		tmp=np.sort(absdiff[:,j])
+		err[j]=tmp[round(min(N,level*N))-1]
+	
+	if plot:
+		import matplotlib.pyplot as plt
+		import matplotlib.gridspec as gridspec
+		nvar=np.size(X,1)
+		if nvar==2: #Exactly two parameters so produce a scatter plot and histograms
+			fig = plt.figure(figsize=(8, 4.8))
+			gs = gridspec.GridSpec(2, 2, width_ratios=[1.5, 1], height_ratios=[1, 1])
+			# Left square spans both rows
+			ax_left = fig.add_subplot(gs[:, 0])
+			axes = [fig.add_subplot(gs[0, 1]), fig.add_subplot(gs[1, 1])]
+			ax_left.set_aspect('equal')
+			ax_left.scatter(X[:,0],X[:,1],s=0.1)
+			ax_left.set_xlabel(r'$c_0$')
+			ax_left.set_ylabel(r'$c_1$')
+			ax_left.plot([plow[0],plow[0]],[np.min(X[:,1]),np.max(X[:,1])],'k--')
+			ax_left.plot([phigh[0],phigh[0]],[np.min(X[:,1]),np.max(X[:,1])],'k--')
+			ax_left.plot([np.min(X[:,0]),np.max(X[:,0])],[plow[1],plow[1]], 'k--')
+			ax_left.plot([np.min(X[:,0]),np.max(X[:,0])],[phigh[1],phigh[1]], 'k--')
+			
+			ax_left.set_aspect(1.0/ax_left.get_data_ratio(), adjustable='box')
+		else:  #only produce histograms
+			fig, axes = plt.subplots(nrows=nvar, ncols=1, figsize=(4, 2*nvar))
+			if nvar==1: axes=[axes] # fix stupid inconsistency in plt.subplots so that axes is always a list
+		
+		for i,ax in enumerate(axes):
+			counts,*_=ax.hist(X[:,i], bins=50)
+			ycent=0.5*max(counts)
+			ax.errorbar(median[i],ycent,xerr=sig[i],fmt='ro',capsize=5)
+			ax.plot([plow[i],plow[i]]  ,[0,0.8*ycent],'k--')
+			ax.plot([phigh[i],phigh[i]],[0,0.8*ycent],'k--')
+			ax.plot([median[i]-err[i], median[i]-err[i]], [0,0.1*ycent],'r-')
+			ax.plot([median[i]+err[i], median[i]+err[i]], [0,0.1*ycent],'r-')
+			ax.set_xlabel(r'$c_{%d}$'%i)  #Name the variables c0,c1...
+			ax.set_yticks([])
+	
+		plt.tight_layout()
+		plt.show()
+	
+	if onedim:
+		return (err[0], (plow[0], phigh[0])) #simply return scalars
+	else:
+		return (err, list(zip(plow, phigh)))
+
+
+def linconf(xinp, yinp, ysig, nmc=10000, distrib='normal', level=0.683, ytransform=None, restransform=None):
+	"""Performs the full Monte Carlo linear regression with confidence calculation.
+	
+	This is done by applying the following 5 steps in succession:
+	- addnoise to y values
+	- transform y values (skipped if ytransform==None)
+	- linreg (x,y)
+	- calculate a tuple of results from a,b   (skipped if restransform==None)
+	- confidence for each result
+	
+	For detailed description of parameters, see previous functions
+	Returns (reslist, pchi2) where reslist is a list of (result, error, confidenceinterval) for each calculated result
+	"""
+	
+	ymc=addnoise(yinp, ysig, nmc, distrib)
+	if ytransform!=None:
+		ymc = ytransform(ymc)
+	pp,psig,pchi2,pmc=linreg(xinp,ymc)
+	if restransform!=None:
+		results=restransform(pp[0],pp[1])
+		results_mc=restransform(pmc[:,0],pmc[:,1])
+	else:
+		results=(pp[0],pp[1])
+		results_mc=(pmc[:,0],pmc[:,1])
+	rlist=[]
+	for r,rmc in zip(results,results_mc):
+		perr,confint=confidence(rmc, level)
+		rlist.append((r,perr,confint))
+	return (rlist,pchi2)

linregmc-0.1.1.dist-info/LICENSE

@@ -0,0 +1,22 @@
+MIT License
+
+Copyright (c) 2025, Pär Söderhjelm
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+

linregmc-0.1.1.dist-info/METADATA

@@ -0,0 +1,25 @@
+Metadata-Version: 2.3
+Name: linregmc
+Version: 0.1.1
+Summary: Linear regression with Monte Carlo error analysis
+License: MIT
+Author: Pär Söderhjelm
+Requires-Python: >=3.9,<4.0
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Requires-Dist: numpy (>=1.23,<3.0)
+Description-Content-Type: text/markdown
+
+# linregmc package  (Linear regression with Monte Carlo error analysis)
+
+This package is a single module which contains four functions:<br>
+addnoise        - introduce noise in the data based on known experimental precision<br>
+linreg          - perform multiple linear regressions and calculate goodness-of-fit<br>
+confidence      - get confidence intervals, plot distributions etc.<br>
+linconf         - do all the steps in succession, including possible transformation of the data<br>
+

linregmc-0.1.1.dist-info/RECORD

@@ -0,0 +1,7 @@
+linregmc/.bashlog,sha256=Sj9bZeGI2l4Wo30gCiM-ygL7RsrqlWS9FUgcr0dRV6o,53
+linregmc/__init__.py,sha256=QP5Y7zpbieaPw5lQbGZt4kCFwHCLNgr9O5NSt8n2mIg,24
+linregmc/linregmc.py,sha256=Yst53oktPSNTVU-0bzdvwGQ5cSVjuwWVbt7CuvgHdFo,12425
+linregmc-0.1.1.dist-info/LICENSE,sha256=lPKblSErTvcRCRHGHckVUnxeAQ4dPtcVM6zxK8XuU80,1075
+linregmc-0.1.1.dist-info/METADATA,sha256=ZpI2tS_Z3quDZO6SufXQwKvRvKY0caEc23ln3k4kspM,1095
+linregmc-0.1.1.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
+linregmc-0.1.1.dist-info/RECORD,,

linregmc-0.1.1.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: poetry-core 2.1.3
+Root-Is-Purelib: true
+Tag: py3-none-any