def plotBinomials(dataMatrix,nVals,p):
nTrials = nVals.size # rows are the trials
# same the mean and variances...
means = np.zeros(nTrials)
varReal = np.zeros(nTrials)
varDist = np.zeros(nTrials)
for i,n in enumerate(nVals):
means[i],varReal[i],varDist[i] =\
plotSingleHist(n,p,dataMatrix[i,:],outDir)
# plot the means and variances
fig = pPlotUtil.figure()
plt.subplot(1,2,1)
plt.title("Mean of g(xBar)-g(mu) approaches 0",fontsize=fontsize)
expMean = 0
plt.plot(nVals,means,'ko',label="Actual Mean")
plt.axhline(expMean,color='b',linestyle='--',
label="Expected Mean: {:.2g}".format(expMean))
plt.ylim(-min(means),max(means)*1.1)
plt.xlabel("Value of n for binomial",fontsize=fontsize)
plt.ylabel("Value of g(xBar)-g(mu)",fontsize=fontsize)
plt.legend(fontsize=fontsize)
pPlotUtil.tickAxisFont()
plt.subplot(1,2,2)
plt.semilogy(nVals,varReal,'ko',label="Actual Variance")
plt.semilogy(nVals,varDist,'b--',label="Expected Variance")
plt.title("Variance of g(xBar)-g(mu)\n approaches expected",
fontsize=fontsize)
plt.xlabel("Value of n for binomial",fontsize=fontsize)
plt.ylabel("Value of g(xBar) variance",fontsize=fontsize)
pPlotUtil.tickAxisFont()
plt.legend(fontsize=fontsize)
pPlotUtil.savefig(fig,outDir + "MeanVar")
def plotAlignments(outDir,alignments,scoreOptimal,label):
fontSize = 25
scores = [ getAlignScore(a) for a in alignments ]
fig = pPlotUtil.figure(xSize=24,ySize=12)
plt.subplot(1,2,1)
meanScore,stdevScore,bins = plotScoreHistograms(scores,fontSize,'k')
plt.title("Shuffled DNA alignment Histogram",
fontsize=fontSize)
pdfFunc = norm(loc=meanScore,scale=stdevScore).pdf(bins)
plotPDF = lambda : plt.plot(bins,pdfFunc,'g--',linewidth=3.0,
label="Normal(mean,var)")
plotPDF()
plt.legend(fontsize=fontSize)
ax = plt.subplot(1,2,2)
plotScoreHistograms(scores,fontSize)
plotPDF()
zScore = (scoreOptimal-meanScore)/stdevScore
print("Z Score for {:s} is {:.2f}".format(label,zScore))
# ??? is this the real p Value? Dont think so
extrProb = 1-norm().cdf(zScore)
plt.title(("Histogram of optimal alignment score for {:d} trials\n" +
"Optimal score: {:d}*sigma from shuffled mean.\n"
"P(shuffled score>=optimal) ~ {:.5g}").\
format(len(scores),int(zScore),extrProb),fontsize=fontSize)
plt.axvline(scoreOptimal,color='r',linestyle='--',
label="Optimal global alignment score using {:s}: {:d}".\
format(label,int(scoreOptimal)))
plt.legend(loc='best',fontsize=fontSize)
pPlotUtil.savefig(fig,outDir+ "q2Histograms" + label)
def plotSingleHist(n,p,xTrials,outDir):
# coverage is just a plotting artifact
fig = pPlotUtil.figure()
# mu: expected value of Binomial(n,p)
# effectie variance
dist,distMean,distVar,normalStd,normalDist,xVals,nBins = \
getDeltaModelDistr(n,p,xTrials)
normV = normHist(dist,nBins,\
label=("Actual Distr: Mean={:.4f},Stdev={:.4f}").\
format(distMean,np.sqrt(distVar)))
rawPDF = normalDist.pdf(xVals)
plt.plot(xVals,rawPDF,'r--',linewidth=5.0,
label="Theorertical Distr: Stdev={:.4f}".\
format(normalStd))
plt.title("Histogram for g(xBar)-g(mu) for n={:d},p={:.2f}".\
format(int(n),p),fontsize=g_title)
plt.xlabel("(g(Xbar)-g(mu)) ~ Normal(0,[g'(x)*sigma]^2/n)",
fontsize=g_label)
plt.ylabel("Proportion",fontsize=g_label)
plt.legend(frameon=False)
pPlotUtil.tickAxisFont()
catArr = list(rawPDF) + list(normV)
plt.ylim([0,max(catArr)*1.2])
plt.xlim([-max(nBins),max(nBins)])
pPlotUtil.tickAxisFont()
pPlotUtil.savefig(fig,outDir + "trial_n{:d}".format(int(n)))
#return the statistics for plotting
return distMean,distVar,normalStd**2
def plotSingleHist(n, p, xTrials, outDir):
# coverage is just a plotting artifact
fig = pPlotUtil.figure()
# mu: expected value of Binomial(n,p)
# effectie variance
dist,distMean,distVar,normalStd,normalDist,xVals,nBins = \
getDeltaModelDistr(n,p,xTrials)
normV = normHist(dist,nBins,\
label=("Actual Distr: Mean={:.4f},Stdev={:.4f}").\
format(distMean,np.sqrt(distVar)))
rawPDF = normalDist.pdf(xVals)
plt.plot(xVals,rawPDF,'r--',linewidth=5.0,
label="Theorertical Distr: Stdev={:.4f}".\
format(normalStd))
plt.title("Histogram for g(xBar)-g(mu) for n={:d},p={:.2f}".\
format(int(n),p),fontsize=g_title)
plt.xlabel("(g(Xbar)-g(mu)) ~ Normal(0,[g'(x)*sigma]^2/n)",
fontsize=g_label)
plt.ylabel("Proportion", fontsize=g_label)
plt.legend(frameon=False)
pPlotUtil.tickAxisFont()
catArr = list(rawPDF) + list(normV)
plt.ylim([0, max(catArr) * 1.2])
plt.xlim([-max(nBins), max(nBins)])
pPlotUtil.tickAxisFont()
pPlotUtil.savefig(fig, outDir + "trial_n{:d}".format(int(n)))
#return the statistics for plotting
return distMean, distVar, normalStd**2
def plotBinomials(dataMatrix, nVals, p):
nTrials = nVals.size # rows are the trials
# same the mean and variances...
means = np.zeros(nTrials)
varReal = np.zeros(nTrials)
varDist = np.zeros(nTrials)
for i, n in enumerate(nVals):
means[i],varReal[i],varDist[i] =\
plotSingleHist(n,p,dataMatrix[i,:],outDir)
# plot the means and variances
fig = pPlotUtil.figure()
plt.subplot(1, 2, 1)
plt.title("Mean of g(xBar)-g(mu) approaches 0", fontsize=fontsize)
expMean = 0
plt.plot(nVals, means, 'ko', label="Actual Mean")
plt.axhline(expMean,
color='b',
linestyle='--',
label="Expected Mean: {:.2g}".format(expMean))
plt.ylim(-min(means), max(means) * 1.1)
plt.xlabel("Value of n for binomial", fontsize=fontsize)
plt.ylabel("Value of g(xBar)-g(mu)", fontsize=fontsize)
plt.legend(fontsize=fontsize)
pPlotUtil.tickAxisFont()
plt.subplot(1, 2, 2)
plt.semilogy(nVals, varReal, 'ko', label="Actual Variance")
plt.semilogy(nVals, varDist, 'b--', label="Expected Variance")
plt.title("Variance of g(xBar)-g(mu)\n approaches expected",
fontsize=fontsize)
plt.xlabel("Value of n for binomial", fontsize=fontsize)
plt.ylabel("Value of g(xBar) variance", fontsize=fontsize)
pPlotUtil.tickAxisFont()
plt.legend(fontsize=fontsize)
pPlotUtil.savefig(fig, outDir + "MeanVar")
def plotErrorAnalysis(mean,std,params,labels,fullOutput):
rowsPerPlot = min(4,len(mean))
fig = pPlotUtil.figure(xSize=rowsPerPlot*6,ySize=len(mean)*4)
nTrials = len(mean)
colors = pPlotUtil.cmap(nTrials)
minP = min([ min(p) for p in params] )
maxP = max([ max(p) for p in params] )
lowerAcc = min([min(acc.flatten()) for acc in mean])
lowerBounds = [(meanV[:,1]-stdV[:,1]) for meanV,stdV in zip(mean,std) ]
validLowerBound = np.array([np.max(bound) for bound in lowerBounds ])
bestIdx = np.array([np.argmax(bound) for bound in lowerBounds ] )
sortedBestValid = np.argsort(validLowerBound)[::-1]
for idx in sortedBestValid:
print("{:s} has lower accuracy of {:.3f} at condition {:.2g}".\
format(labels[idx],validLowerBound[idx],bestIdx[idx]))
i=0
fontsize=20
for meanV,stdV,pVals,lab in zip(mean,std,params,labels):
ax=plt.subplot(np.ceil(nTrials/rowsPerPlot),rowsPerPlot,i+1)
plt.errorbar(pVals,meanV[:,0],stdV[:,0],fmt='o-',color=colors[i],
label='train')
plt.errorbar(pVals,meanV[:,1],stdV[:,1],fmt='x--',color=colors[i],
label='vld')
ax.set_xscale('log')
plt.axhline(0.8,color='r',linestyle='--')
plt.ylim([lowerAcc*0.9,1])
plt.xlim([minP*0.7,maxP*1.3])
plt.title(lab,fontsize=fontsize)
i+=1
plt.xlabel('Classifier parameter')
plt.ylabel('Accuracy')
pPlotUtil.savefig(fig,fullOutput + 'allAcc')
def plotAccuracies(outDir,label,accMean,accStd,fitParam):
fig = pPlotUtil.figure()
plt.errorbar(fitParam,accMean[:,0],accStd[:,0],fmt='ro-',
label="Training Set")
plt.errorbar(fitParam,accMean[:,1],accStd[:,1],fmt='kx-',
label="Validation Set")
plt.xscale('log', nonposy='clip')
plt.axhline(1,color='b',linestyle='--',label='max')
plt.xlabel("Fit parameter")
plt.ylabel("Accuracy")
plt.xlim([min(fitParam)*0.7,max(fitParam)*1.3])
plt.legend(loc='best')
plt.title('Accuracy vs fit parameter for fitter: {:s}'.format(label))
pPlotUtil.savefig(fig,outDir + "accuracies")
def predict(fitter,x,yReal,rawDat,label,saveDir,colNames,fitterCoeff,objClass,
featureObjects,saveBad=False,saveCoeffs=True,plot=True):
try:
yPred = fitter.predict(x)
except TypeError:
yPred = fitter.predict(x.toarray())
cm = confusion_matrix(yReal,yPred)
acc= accuracy_score(yReal,yPred)
# Show confusion matrix in a separate window
if (saveBad):
# XXX could profile?
profileLosers(saveDir,label,yPred,yReal,rawDat,objClass,x,
featureObjects)
if (plot):
fig = pPlotUtil.figure()
ax = plt.subplot(1,1,1)
numCols = colNames.size
coeffs = fitterCoeff(fitter)
nCoeffs = coeffs.size
xRange = range(nCoeffs)
saveName = saveDir + label + "coeffs"
sortIdx = np.argsort(coeffs)[::-1]
sortedCoeffs = coeffs[sortIdx]
sortedNames = colNames[sortIdx]
sortedFeatures = [featureObjects[s] for s in sortIdx]
stacked = np.vstack((sortedNames,sortedCoeffs)).T
np.savetxt(saveName,stacked,fmt=["%s","%.3g"],delimiter="\t")
print numCols, " Columns"
maxToPlot = min(numCols//2,25) # on each side
if( numCols == nCoeffs):
# then we have a coefficient per feature (column), so use them for ticks
coeffsToPlot = list(sortedCoeffs[:maxToPlot]) + \
list(sortedCoeffs[-maxToPlot:])
labelsToPlot = list(sortedNames[:maxToPlot]) +\
list(sortedNames[-maxToPlot:])
featuresPlotted = list(sortedFeatures[:maxToPlot]) + \
list(sortedFeatures[-maxToPlot:])
xToPlot = range(len(coeffsToPlot))
ax.bar(xToPlot,coeffsToPlot,align='center')
ax.set_xticks(xToPlot)
ax.set_xticklabels(labelsToPlot,rotation='vertical')
plt.xlabel("coefficient name")
plt.ylabel("Predictor strength")
else:
plt.plot(xRange,coeffs,'ro-')
plt.xlabel("Fitter Coefficients")
plt.ylabel("Predictor strength")
pPlotUtil.savefig(fig,saveName)
return acc
def plotFec(expect,algorithm,inFile,saveAs):
time,sep,force = HDF5Util.GetTimeSepForce(inFile)
fig = pPlotUtil.figure()
IgorUtil.PlotFec(sep,force)
# limit the axis to close to the touchoff (10% of range)
# (plotFEC starts at 0)
minV = min(sep)
rangeSepNm = 1e9 * abs(max(sep)-minV)
rangeX = [0,rangeSepNm/10]
plt.xlim(rangeX)
# plot the expected and algorithm locations as nm, normalized to min
norm = lambda x : (x-minV)*1e9
plt.axvline(norm(expect),
label="Expected surface location",lw=3,linestyle="--",
color="g")
plt.axvline(norm(algorithm),label="Algorithm surface location",lw=3,
linestyle="--",color="k")
pPlotUtil.legend()
pPlotUtil.savefig(fig,saveAs)
def plotAll(kArrs,outDir):
maxKeach = [max(k) for k in kArrs ]
maxK = max(maxKeach)
bins = range(maxK+1)
numTrials = len(kArr)
means = np.array([np.mean(k) for k in kArrs])
stdevs = np.array([np.std(k) for k in kArrs])
for i,k in enumerate(kArrs):
fig = pPlotUtil.figure()
plt.hist(k,bins=bins,align='left',label='Data from {:d} sequences'.
format(int(numOligos)),normed=True)
mean = means[i]
plt.axvline(mean,color='r',label="Mean:{:.3f}".format(mean),
linewidth=2.0)
plt.xlim([0,maxK])
plt.xlabel('K, minimum k-mer with at most 1 occurence in DNA sequence')
plt.ylabel('Proportion of occurences')
plt.title('K histogram (normalized) for DNA sequences of length {:d}'.
format(lengths[i]))
plt.legend()
pPlotUtil.savefig(fig,outDir + "k{:d}".format(i))
return means,stdevs
def profileLosers(saveDir,label,yPred,yActual,rawDat,dataClass,featureMat,
featureObjects):
# get what we got wrong
badIdx,predictedDeath,predictedSurv = getIdxMistakes(yPred,yActual)
nSurv = len(predictedSurv)
nDead = len(predictedDeath)
fig = pPlotUtil.figure(xSize=16,ySize=12,dpi=200)
# get the matrix, all features 0 --> 1
toPlot = getNormalizedFeatureMatrix(badIdx,featureMat,
lambda x: sortByPred(x,yPred,yActual))
# get the number of non-zero elements in each column
aspectStr = plotFeatMatr(toPlot,featureObjects,featureMat,saveDir,label,
badIdx)
plt.axhline(len(predictedSurv),linewidth=3,color='c')
plt.title("Line Divides {:d} actual deceased from {:d} actual survived".\
format(nSurv,nDead),y=1.3,fontsize=g_title)
plt.legend(loc="upper right", bbox_to_anchor=(0.4, -0.4))
badVals = rawDat[badIdx,:]
np.savetxt(saveDir + 'debug_{:s}.csv'.format(label),badVals,fmt="%s",
delimiter=',')
pPlotUtil.savefig(fig,saveDir + "mOut" + label,tight=True)
def plotSpotDist(mLabels,spots,outPath,subtractMean):
colors = ['r', 'g', 'b', 'y','k']
nColors = len(colors)
# go to nm
mLabelsNm = mLabels * 1e9
mSetSpots = sorted(set(spots))
labelsBySpot = []
rawBySpot = []
flattenedFromMean = []
# first, get the spot-wise labelling
for i,spot in enumerate(mSetSpots):
# get the indices of the spots
spotIdx = np.where(abs((spots - spot)) < 1e-9)[0]
thisSpotLabels = mLabelsNm[spotIdx]
meanV = np.mean(thisSpotLabels)
if (subtractMean):
thisSpotLabels -= meanV
labelsBySpot.append(thisSpotLabels)
flattenedFromMean.extend(thisSpotLabels)
if (subtractMean):
rawBySpot.append(thisSpotLabels + meanV)
else:
rawBySpot.append(thisSpotLabels)
# get the min and max from the labelsBySpot array
bins = np.linspace(min(flattenedFromMean),max(flattenedFromMean),10)
fig = pPlotUtil.figure(xSize=12,ySize=12)
ax = fig.add_subplot(111, projection='3d',)
for i,thisSpotLabels in enumerate(labelsBySpot):
mColor = colors[i % nColors]
height,left = np.histogram(thisSpotLabels,bins=bins)
ax.bar(left[:-1], height, zs=i,zdir='y', color=mColor, alpha=0.7,
edgecolor="none",linewidth=0)
xStr = r'$\Delta$ from Expected Surface Loc. [nm]'
pPlotUtil.lazyLabel(xStr,
"Surface Position (arb)",
"Dependence of Surface Location Distribution on Position",
zlab="Count")
pPlotUtil.savefig(fig,outPath + "AllSpots.png")
# get a figure showing the mean surface location, assuming
# we reshape into an Nx(whatever) array
N = 5
# -1: infer dimension
meanVals = [np.mean(mList) for mList in rawBySpot]
meanSurf = np.reshape(meanVals,(-1,N))
meanSurf -= np.min(meanSurf)
fig = pPlotUtil.figure(ySize=14,xSize=10)
ax = fig.add_subplot(111, projection='3d')
# convert to nm (XXX assuming grid is 1micron for each)
Nx = N
Ny = meanSurf.shape[0]
x = np.linspace(0, Nx, Nx) * 1e3
y = np.linspace(0, Ny, Ny) * 1e3
xv, yv = np.meshgrid(x, y)
ax.plot_wireframe(xv,yv,meanSurf)
pPlotUtil.lazyLabel("X Location [nm]","Y Location [nm]",
"Surface Position Varies with height")
pPlotUtil.zlabel("Surface height (relative to min)")
pPlotUtil.savefig(fig,outPath + "Surface.png")
fig = pPlotUtil.figure(ySize=14,xSize=10)
plt.subplot(2,1,1)
nPoints = len(flattenedFromMean)
vals,edges,_=plt.hist(flattenedFromMean,bins=bins)
# add a 'fudge' factor to make plotting better,
fudgeX = (max(edges)-min(edges))*0.05
xlim = [min(edges)-fudgeX,max(edges)+fudgeX]
yLim = [0,max(vals)]
pPlotUtil.lazyLabel(xStr,
"Number of counts",
"Algorithm finds surface within 10nm, >98%, N={:d}".\
format(nPoints))
normed = [0,max(vals)/sum(vals)]
plt.xlim(xlim)
propAx = pPlotUtil.secondAxis(plt.gca(),"Proportion",normed,yColor="Red")
propAx.axhline("0.05",color='r',
label="5% of Curves",linestyle='--',linewidth=4.0)
pPlotUtil.legend()
# plot the CDF
plt.subplot(2,1,2)
# add a zero at the start, so the plot matches the PDF
cdf = np.zeros(edges.size)
cdf[1:] = np.cumsum(vals/sum(vals))
mX = edges
plt.plot(mX,cdf,linestyle='--',linewidth=4,color='k')
plt.xlim(xlim)
pPlotUtil.lazyLabel(xStr,
"Cummulative Proportion",
("Cummulative Density Function," +
"Surface Detection Performance"))
plt.gca().fill_between(mX, 0, cdf,alpha=0.3)
pPlotUtil.savefig(fig,outPath + "FlatSpots.png")
lengths = np.array([2,4,8,16,32,64,128,256,512])
# save the K array: minimum k to have at most one k-mer
# initialize to -1, so that we know when we have the minimum
outDir = "./out/"
pGenUtil.ensureDirExists(outDir)
forceRun = False
test = False
# use checkpointing to save data, since it takes forever
kArr = pCheckUtil.getCheckpoint('./tmp/check.pkl',getKSequence,forceRun,
lengths,numOligos,weights,chars)
meanVals,std = pCheckUtil.getCheckpoint('./tmp/meanStd.pkl',plotAll,
forceRun,kArr,outDir)
if (test):
testDnaGeneration(chars,lengths,numOligos,weights)
# plot the mean k vs dna length, l (in theory, k is approx log_1/q(l+1))
fig = pPlotUtil.figure()
ax = plt.subplot(1,3,1)
plt.errorbar(x=lengths,y=meanVals,yerr=std,fmt='ro-',label='Mean K')
tKVals = getTheoryK(lengths,q)
plt.plot(lengths,tKVals,'b--',label='Log_[1/q](l+1)')
xLab = 'DNA Length (l)'
plt.xlabel(xLab)
plt.ylabel('Mean K value')
plt.title('Mean K vs length')
ax.set_xscale('log')
plt.legend(loc='best')
ax = plt.subplot(1,3,2)
plotError(meanVals,tKVals,lengths,xLab,'Absolute Error in Mean K ',
'Absolute error in Mean K',ax,relative=False)
ax = plt.subplot(1,3,3)
plotError(meanVals,tKVals,lengths,xLab,'Relative Error in Mean K [0-->1]',
import PlotUtilities as pPlotUtil
import CheckpointUtilities as pCheckUtil
from scipy.stats import norm
outDir = "./out/"
pGenUtil.ensureDirExists(outDir)
mean = 0
stdev = 1
epsilon = stdev / 100
nPoints = 1000
normDist = norm(loc=mean, scale=stdev)
offsets = np.linspace(mean - 3 * stdev, mean + 3 * stdev, nPoints)
probability = 2 * (normDist.cdf(
(offsets + epsilon - mean) / stdev) - normDist.cdf(
(offsets - epsilon - mean) / stdev))
fig = pPlotUtil.figure()
plt.plot(offsets,probability,'r-',
label="mu = {:.1f}, sigma = {:.1f}, epsilon = {:.2f}".\
format(mean,stdev,epsilon))
plt.xlabel("offset for CDF, c0")
plt.ylabel("Probability (arbitrary units) to land within epsilon of c0")
plt.axvline(0,
color='k',
linestyle='--',
label="Maximum probability when centered near mu")
plt.legend(loc='best')
plt.title("Probability of landing within epsilon of c0 maximized near mu")
pPlotUtil.savefig(fig, outDir + "q1_1")
