gonum / stat

73%

master: 74%

LAST BUILD ON BRANCH keepsigma

branch: keepsigma

CHANGE BRANCH

x

Reset

• keepsigma
• master
• test-coverage-improvement
• only-coveralls-on-gonum-branch
• simplify-sigs
• covariance-matrix
• incremental-stats
• addrsquared
• descale
• covariancematrix-multrans
• blas64
• fixhisto
• fixmax
• addmvgauss
• fixcovman
• use-multrans
• fix-marshaller-spelling
• add-numparameters
• remove-marshaler
• internalize-parameters
• correlationmatrix
• travis-gimme
• remove-redundant-types
• travis-go-1.4.2
• fixnormal
• featuresample
• addmvnorm
• docfix
• addlognormal
• weighted
• mvinterface
• fixmul
• rc1
• tcopy-fix
• equal
• no-overload
• equalapprox
• fixcol
• covsym
• coveralls-badge
• sudo-false
• scaled-symouterk
• errors
• normalcondition
• addmarg
• addcat
• normcondpan
• fixuniform
• go1.6beta2
• fixunifprob
• batchsample
• addsamplemv
• go1.6rc1
• addcombin
• go1.6rc2
• build-status-only-master
• normprec
• pca
• addexpected
• typo
• trim
• docs
• names
• go-1.5.4-and-1.6.1
• addmargsing
• travis-go-1.6.2
• addtransformgauss
• go1.7beta1
• renamescore
• addbernouilli
• fixdistuv
• go-1.7beta2
• go-1.6.3-and-1.7rc2
• go-1.7rc3
• regression
• add-go.1.7-drop-go-1.4
• covdoc
• allocs
• addgamma
• dirichlet
• fixbetacomment
• addchisquared
• fixgamma
• expandstudents
• addmvt
• addquantiles
• mvtmargsing
• addwishart
• cleanup-wishart
• cheaper-ca
• addmahal
• fixnormchol
• addnormquant
• statdist
• fixlatin
• movequant
• fixcovrace
• setsigma
• cc-example
• binomial-overflow
• suffstat
• unifent
• addquants
• unifbound
• statdist4
• statdistunif
• close
• archive
• deprecate

Build # 920

Build Type

push

travis-ci

Committed by btracey

Commit Message

Store sigma in Normal and StudentsT

Currently, we throw sigma away, and recompute it if necessary. This PR keeps sigma. This fixes an issue with concurrent calling of methods. In addition, however, it removes any possible issues with reconstructing a badly-conditioned sigma from its Cholesky decomposition, and avoids an extra n^3 work if sigma does need to be recomputed. The complexity of the implementation and difficulties listed above is not worth the memory savings in some cases, especially since the memory of the type is already O(n^2)

Run Details

2426 of 3321 relevant lines covered (73.05%)

2.49 hits per line