def getLatexTable(measures, cvScalings, idealMeasures):
rowNames = getRowNames(cvScalings, True)
table = Latex.array1DToRow(foldsSet) + "\\\\ \n"
for j in range(sampleSizes.shape[0]):
meanMeasures = numpy.mean(measures, 0)
stdMeasures = numpy.std(measures, 0)
table += Latex.array2DToRows(meanMeasures[j, :, :].T, stdMeasures[j, :, :].T) + "\n"
meanIdealMeasures = numpy.mean(idealMeasures, 0)
stdIdealMeasures = numpy.std(idealMeasures, 0)
table += Latex.array2DToRows(numpy.ones((1, len(foldsSet)))*meanIdealMeasures[j], numpy.ones((1, len(foldsSet)))*stdIdealMeasures[j]) + "\n"
table = Latex.addRowNames(rowNames, table)
return table, meanMeasures, stdMeasures
def testAddRowNames(self):
numpy.random.seed(21)
X = numpy.random.rand(2, 2)
Y = numpy.random.rand(2, 2)
latexTable = Latex.array2DToRows(X, Y)
rowNames = ["a", "b"]
latexTable = Latex.addRowNames(rowNames, latexTable)
outputStr = "a & 0.049 (0.206) & 0.289 (0.051)\\\\\n"
outputStr += "b & 0.721 (0.302) & 0.022 (0.664)\\\\\n"
self.assertTrue(latexTable == outputStr)
#Now test error method
rowNames = ["a", "b", "c"]
self.assertRaises(ValueError, Latex.addRowNames, rowNames, latexTable)
#Now test error method
rowNames = ["a"]
self.assertRaises(ValueError, Latex.addRowNames, rowNames, latexTable)
def summary(datasetNames, sampleSizes, foldsSet, cvScalings, sampleMethods, fileNameSuffix, gridResultsSuffix="GridResults"):
"""
Print the errors for all results plus a summary.
"""
numMethods = (1+(cvScalings.shape[0]+1))
numDatasets = len(datasetNames)
overallErrors = numpy.zeros((numDatasets, len(sampleMethods), sampleSizes.shape[0], foldsSet.shape[0], numMethods))
overallStdWins = numpy.zeros((len(sampleMethods), len(sampleSizes), foldsSet.shape[0], numMethods+1, 3), numpy.int)
overallErrorsPerSampMethod = numpy.zeros((numDatasets, len(sampleMethods), len(sampleSizes), numMethods), numpy.float)
table1 = ""
table2 = ""
table3 = ""
for i in range(len(datasetNames)):
table3Error = numpy.zeros((2, len(sampleMethods)))
table3Stds = numpy.zeros((2, len(sampleMethods)))
for j in range(len(sampleMethods)):
print("="*50 + "\n" + datasetNames[i] + "-" + sampleMethods[j] + "\n" + "="*50 )
outfileName = outputDir + datasetNames[i] + sampleMethods[j] + fileNameSuffix + ".npz"
try:
data = numpy.load(outfileName)
errors = data["arr_0"]
params = data["arr_1"]
meanErrorGrids = data["arr_2"]
stdErrorGrids = data["arr_3"]
meanApproxGrids = data["arr_4"]
stdApproxGrids = data["arr_5"]
#Load ideal results
outfileName = outputDir + datasetNames[i] + gridResultsSuffix + ".npz"
data = numpy.load(outfileName)
idealErrors = data["arr_0"]
errorTable, meanErrors, stdErrors = getLatexTable(errors, cvScalings, idealErrors)
wins = getWins(errors)
idealWins = getIdealWins(errors, idealErrors)
excessError = numpy.zeros(errors.shape)
for k in range(errors.shape[1]):
excessError[:, k, :, :] = errors[:, k, :, :] - numpy.tile(errors[:, k, :, 0, numpy.newaxis], (1, 1, numMethods))
meanExcessError = numpy.mean(excessError, 0)
stdExcessError = numpy.std(excessError, 0)
excessErrorTable, meanExcessErrors, stdExcessErrors = getLatexTable(excessError, cvScalings, idealErrors)
overallErrorsPerSampMethod[i, j, :, :] = numpy.mean(meanErrors, 1)
overallErrors[i, j, :, :, :] = meanExcessError
overallStdWins[j, :, :, 0:-1, :] += wins
overallStdWins[j, :, :, -1, :] += idealWins
print(errorTable)
#print("Min error is: " + str(numpy.min(meanErrors)))
#print("Max error is: " + str(numpy.max(meanErrors)))
#print("Mean error is: " + str(numpy.mean(meanErrors)) + "\n")
#This is a table with V=10, alpha=1 and CV sampling
sliceFoldIndex = 0
print(meanErrors[0, 1, 0])
numSliceMethods = 3
table1Error = numpy.zeros(len(sampleSizes)*numSliceMethods)
table1Std = numpy.zeros(len(sampleSizes)*numSliceMethods)
for k in range(len(sampleSizes)):
table1Error[k*numSliceMethods] = meanErrors[k, sliceFoldIndex, 0]
table1Error[k*numSliceMethods+1] = meanErrors[k, sliceFoldIndex, 1]
table1Error[k*numSliceMethods+2] = meanErrors[k, sliceFoldIndex, 4]
table1Std[k*numSliceMethods] = stdErrors[k, sliceFoldIndex, 0]
table1Std[k*numSliceMethods+1] = stdErrors[k, sliceFoldIndex, 1]
table1Std[k*numSliceMethods+2] = stdErrors[k, sliceFoldIndex, 4]
if j == 0:
table1 += datasetNames[i] + " & " + Latex.array2DToRows(numpy.array([table1Error]), numpy.array([table1Std])) + "\n"
#See how alpha varies with V=10, CV sampling
table2Error = numpy.zeros(range(numMethods-2))
table2Std = numpy.zeros(range(numMethods-2))
for s in range(len(sampleSizes)):
table2Error = meanErrors[s, sliceFoldIndex, 2:]
table2Std = stdErrors[s, sliceFoldIndex, 2:]
if j == 0:
table2 += datasetNames[i] + " $m=" + str(sampleSizes[s]) + "$ & " + Latex.array2DToRows(numpy.array([table2Error]), numpy.array([table2Std])) + "\n"
"""
#See how each sample method effects CV and pen alpha=1
fourFoldIndex = 4
hundredMIndex = 1
table3Error[0, j] = meanErrors[hundredMIndex, fourFoldIndex, 0]
table3Error[1, j] = meanErrors[hundredMIndex, fourFoldIndex, 3]
table3Stds[0, j] = stdErrors[hundredMIndex, fourFoldIndex, 0]
table3Stds[1, j] = stdErrors[hundredMIndex, fourFoldIndex, 3]
"""
except IOError:
print("Failed to open file: " + outfileName)
table3 += Latex.addRowNames([datasetNames[i] + " Std ", datasetNames[i] + " PenVF "], Latex.array2DToRows(table3Error, table3Stds))
datasetMeanErrors = Latex.listToRow(sampleMethods) + "\n"
for j in range(len(sampleSizes)):
datasetMeanErrors += Latex.array2DToRows(overallErrorsPerSampMethod[i, :, j, :].T) + "\n"
datasetMeanErrors = Latex.addRowNames(getRowNames(cvScalings), datasetMeanErrors)
print(datasetMeanErrors)
print("="*50 + "\n" + "Sliced Tables" + "\n" + "="*50)
print(table1 + "\n")
print(table2 + "\n")
print(table3)
print("="*50 + "\n" + "Overall" + "\n" + "="*50)
overallMeanErrors = numpy.mean(overallErrors, 0)
overallStdErrors = numpy.std(overallErrors, 0)
for i in range(len(sampleMethods)):
print("-"*20 + sampleMethods[i] + "-"*20)
overallErrorTable = Latex.array1DToRow(foldsSet) + "\\\\ \n"
overallWinsTable = Latex.array1DToRow(foldsSet) + " & Total & " +Latex.array1DToRow(foldsSet) + " & Total \\\\ \n"
rowNames = getRowNames(cvScalings)
for j in range(sampleSizes.shape[0]):
overallErrorTable += Latex.array2DToRows(overallMeanErrors[i, j, :, :].T, overallStdErrors[i, j, :, :].T, bold=overallMeanErrors[i, j, :, :].T<0) + "\n"
tiesWins = numpy.r_[overallStdWins[i, j, :, :, 0], overallStdWins[i, j, :, :, 1], overallStdWins[i, j, :, :, 2]]
overallWinsTable += Latex.array2DToRows(tiesWins.T) + "\n"
overallErrorTable = Latex.addRowNames(rowNames, overallErrorTable)
rowNames = getRowNames(cvScalings, True)
overallWinsTable = Latex.addRowNames(rowNames, overallWinsTable)
print(Latex.latexTable(overallWinsTable, "Wins for " + sampleMethods[i], True))
print(Latex.latexTable(overallErrorTable.replace("0.", "."), "Excess errors for " + sampleMethods[i], True))
#print(overallWinsTable)
#print(overallErrorTable)
#Now print the mean errors for all datasets
datasetMeanErrors = Latex.listToRow(sampleMethods) + "\n"
overallErrorsPerSampMethod = numpy.mean(overallErrorsPerSampMethod[:, :, :, :], 0)
for j in range(len(sampleSizes)):
datasetMeanErrors += Latex.array2DToRows(overallErrorsPerSampMethod[:, j, :].T) + "\n"
datasetMeanErrors = Latex.addRowNames(getRowNames(cvScalings), datasetMeanErrors)
print(datasetMeanErrors)
plt.figure(plotInd)
plt.errorbar(numpy.arange(numDegrees), meanDegreeDists[ind, :], yerr=stdDegreeDists[ind, :], color="k")
plt.plot(numpy.arange(numDegrees), idealDegreeDists[ind, :], "k--")
plt.xlabel("degree")
plt.ylabel("frequency")
plotInd += 1
#Print the table of thetas
thetas = numpy.array(thetas)
meanThetas = numpy.mean(thetas, 1)
stdThetas = numpy.std(thetas, 1)
table = Latex.array2DToRows(meanThetas.T, stdThetas.T, precision=4)
rowNames = ["$|\\mathcal{I}_0 |$", "$\\alpha$", "$\\gamma$", "$\\beta$", "$\\lambda$", "$\\sigma$"]
table = Latex.addRowNames(rowNames, table)
print(table)
#Now print the graph properties
idealTable = []
tableMeanArray = []
tableStdArray = []
for ind in inds:
idealTable.append(idealMeasures[ind, :, timeInds])
tableMeanArray.append(meanMeasures[ind, :, timeInds])
tableStdArray.append(stdMeasures[ind, :, timeInds])
idealTable = numpy.vstack(idealTable).T
tableMeanArray = numpy.vstack(tableMeanArray).T
tableStdArray = numpy.vstack(tableStdArray).T
plt.xlabel("False Positive Rate")
plt.ylabel("True Positive Rate")
plt.savefig(figureDir + labelName.replace(".", "_") + "-ROC.eps")
plotInd += 1
#The last column is the mean value
meanMeanAUCs = numpy.mean(meanAUCs, 1)
meanStdAUCs = numpy.mean(stdAUCs, 1)
meanAUCs = numpy.c_[meanAUCs, meanMeanAUCs]
stdAUCs = numpy.c_[stdAUCs, meanStdAUCs]
print("\n")
print(Latex.listToRow(labelNames))
print(Latex.addRowNames(rowNames, Latex.array2DsToRows(meanAUCs, stdAUCs, 2)))
#----------------- Results for RankBoost and RankSVM ----------------------------------
algorithmNames = ["RankBoost", "RankSVM"]
numMethods = len(dataTypes)*len(algorithmNames)
rowNames = numpy.zeros(numMethods, "a20")
meanAUCs = numpy.zeros((numMethods, len(labelNames)))
stdAUCs = numpy.zeros((numMethods, len(labelNames)))
for m in range(len(algorithmNames)):
algorithmName = algorithmNames[m]
for i in range(len(labelNames)):
labelName = labelNames[i]
logging.debug(resultsDir)
newNumRecordSteps = numRecordSteps + 5
endDate += HIVModelUtils.realTestPeriods[j]
recordStep = (endDate-startDate)/float(newNumRecordSteps)
thetaArray = loadThetaArray(N, resultsDir, t)[0]
print(thetaArray)
meanTable = numpy.array([thetaArray.mean(0)]).T
print(meanTable)
stdTable = numpy.array([thetaArray.std(0)]).T
table = Latex.array2DToRows(meanTable, stdTable, precision=4)
rowNames = ["$\\|\\mathcal{I}_0 \\|$", "$\\rho_B$", "$\\alpha$", "$C$", "$\\gamma$", "$\\beta$", "$\\kappa_{max}$", "$\\lambda_H$", "$\\lambda_B$", "$\\sigma_{WM}$", "$\\sigma_{MW}$","$\\sigma_{MB}$"]
table = Latex.addRowNames(rowNames, table)
print(table)
resultsFileName = outputDir + "IdealStats.pkl"
stats = Util.loadPickle(resultsFileName)
times, vertexArray, removedGraphStats = stats
times = numpy.array(times) - startDate
times2 = numpy.arange(startDate, endDate+1, recordStep)
times2 = times2[1:]
times2 = numpy.array(times2) - startDate
graphStats = GraphStatistics()
#First plot graphs for ideal theta
plotInd = 0
except:
logging.debug("File not found : " + str(fileName))
numMissingFiles += 1
logging.debug("Number of missing files: " + str(numMissingFiles))
for i, dataName in enumerate(dataNames):
print("-"*10 + dataName + "-"*10)
algorithms = [x.ljust(20) for x in algorithmsAbbr]
currentTestAucsMean = testAucsMean[:, i, :].T
maxAUCs = numpy.zeros(currentTestAucsMean.shape, numpy.bool)
maxAUCs[numpy.argmax(currentTestAucsMean, 0), numpy.arange(currentTestAucsMean.shape[1])] = 1
table = Latex.array2DToRows(testAucsMean[:, i, :].T, testAucsStd[:, i, :].T, precision=2, bold=maxAUCs)
print(Latex.listToRow(hormoneNameIndicators))
print(Latex.addRowNames(algorithms, table))
#Now looks at the features for the raw spectra
algorithm = "L1SvmTreeRankForest"
dataName = "raw"
numMissingFiles = 0
numFeatures = 100
numIndicators = 6
featureInds = numpy.zeros((numFeatures, numIndicators))
for i, (hormoneName, hormoneConc) in enumerate(helper.hormoneDict.items()):
try:
fileName = resultsDir + "Weights" + algorithm + "-" + hormoneName + "-0" + "-" + dataName + ".npy"
weights = numpy.load(fileName)