close Warning: Can't synchronize with repository "(default)" (/var/svn/tolp does not appear to be a Subversion repository.). Look in the Trac log for more information.

What is new in Tol 1.1.7

The main contributions of Tol 1.1.7 are

  1. consolidation of NameBlock
  2. new data type VMatrix and
  3. new package for Bayesian Sparse Regression (BSR)

Kernel

  1. Calling ShowCallStack when errors are displayed
  2. Enhancement of BFSMEM memory assignation checking to avoid some invalid accesses
  3. New feature TolOprProfiler to get performance profiling of called TOL functions
  4. New BCore::MemAssignInfo to have a simpler way to detect and avoid some memory leakks by using BFSMEM system
  5. Adding explicit field BGrammar precedence_ to select the lesser one in case of ambiguity
  6. Priorized optional types for arguments of built-in functions
  7. Forcing no lazy behaviour in constructor of special series
  8. Fixes the function description field for DBClose.
  9. Bringing "Tol Extensions Loader" back to live.
  10. New StdLib::CMsg: Controlled Messages is a system to handling with any type of messages to end user in a controlled way

System tools

  1. New function OSFilCat that concats a set of files in a new one
  2. New built-in function FileCat
  3. Adding NameBlock Tpa (Tol Parallel Api) to handle with tolsh servers
  4. New functions RecDir.Count.Line and RecDir.Count.Char to count recursively number of lines and characters of specified files under a directory
  5. Fixed bug in ReportAllBatteries that avoids uptol to work fine with some paths
  6. New Input/Output built-in TOL functions (enhancement bug 555) See examples of use ​test_001 and ​test_002
    1. SysErrNum: Returns the system error code corresponding to the last happened error
    2. SysErrMsg: Returns the system error message corresponding to a specified code.
    3. FOpen: Opens a file using C function fopen and returns an integer value, which is used to refer to the file
    4. FGetText: Reads a string from a file until the character delim is readed, maxSize characters are reached or file ends
    5. FPutText: Writes a string in a open file
    6. FEof: Tests the end-of-file indicator for the stream pointed to by stream, returning non-zero EOF value if it is set
    7. FFlush: The function FFlush forces a write of all user-space buffered data for the given output or update stream via the stream's underlying write function
    8. FClose: Closes the file associated with handle that was the result of calling FOpen. The function returns 0 if successful, EOF to indicate an error

OIS

  1. Ois.DefRoot can be defined as Operative System environment variable Ois.DefRoot or Ois_DefRoot to allow use it at linux.
  2. Fixing bug 520 by using an optional parameter Real oisHasPriorityOnConflict in Ois.Load.
  3. If oisHasPriorityOnConflict is false then, when a referenceable object exist already with the same name of an object inside an OIS image, the older object will be used instead of OIS one. This option could be usefull, for example, to recover old OIS images with damaged objects, as a dating that has been bounded between default dates at building time. If oisHasPriorityOnConflict is true, that's default behaviour, OIS object will be used to reproduce exactly as was built.

Polyn

  1. Fixed bug in IsStationary that failed when the polynomial has zero coefficients

NameBlock

  1. Adding Text _.autodoc.member.<member_name> to a NameBlock force that this member has this text as description
  2. NameBlock members must treated as global or system variables when are used as the dating of a time serie
  3. Global functions must set !NameBlock::current_ to NULL to avoid invalid access to members

Set

  1. New built-in function HasIndexByName to ask if a Set is indexed by name
  2. New method StdLib::getOptArg to work with sets of optional arguments
  3. A new extended API for Classify
  4. New built-in function FindIndexByName to get the position in a indexed by name set of an element knowing its name.
  5. Best comparing of sets

Text

  1. New function StdLib::RegReplace in order to do substitutions with regular expressions

Time Algebra

  1. New robust and fast (log-lineal) implementation of SuccSuccessor and SuccPredecessor
  2. Adding TimeSet Decanual and TenYearly
  3. New built-in variable Real CpuClock that returns the CPU usage elapsed in seconds from session started
  4. Now Real Time returns the time elapsed in seconds from session started.

Matrix algebra

  1. Allowing dense-sparse operations with CholeskiSolve
  2. New built-in function Sort
  3. New thinnig parameter for BinReadRows
  4. New built-in function MatReadDimensions
  5. New VMatrix functions DifEq and BackDifEq with or without specific initial values
  6. Allowing convertion between VMatrix Blas.R.Dense and Cholmod.R.Sparse and viceverse
  7. Fixed bug in default argument Y0 for DifEq and BackDifEq
  8. Fixed bug in PreProdDiag
  9. Applying faster matrix cycling methos to some functions
  10. Forzing LLt transformation and other MatLab lchol options
  11. Adding methods related to difference equations
  12. New methods IsFinite nd InUnknown
  13. New built-in functions Pol2VMat, SubBand
  14. Forzing Structure Alignement to 8 for cholmod.h
  15. Checking dimensions on MatProd
  16. New internal method replicateTriang
  17. Fixed bug in dpotri that was returned as triangular instead of symmetric
  18. Fixed bug inBDBClassify
  19. New TOL data type VMatrix to handle with sparse and dense matrices all together. This type is not completely developped and is a feature in testing state. Please, handle with care!
    1. Current matrix subtypes are CHOLMOD sparse, factor and triplet and BLAS column major dense.
    2. System has been designed to be easily upgraded with new subtypes as Toeplitz, Hankel, Vandermonde, ...
    3. There are all kind of matrix related operations, not all developed already, that has been implemented to be the more efficient and the more transparent for the end user. System can convert internally, if needed, from dense to sparse, sparse to triplet, etc.; to make an operation be posible or faster between differently stored matrices
      1. arithmetic: sum, rest, product, ...
      2. linear algebra: Choleski decomposition and system solving, ...
      3. access and building: cell, band and block access, concatenation, ...
      4. input/output: (including OIS storement) etc)
  20. API of TOL for using BLAS and LAPACK functions
    1. Using BLAS dgemm in generic matrix products
    2. Using BLAS dsyrk in MtMSqr
    3. Using BLAS dtrmm in triangular matrix products
    4. Using BLAS dtrsm in TrSolve, LTSolve and LTInverse
    5. Using LAPACK dpotrf in choleski method
    6. Using LAPACK dpotrf and dpotri in CholeskiInverse
    7. Using BLAS-LAPACK methods in LinearRegression and LinReg
  21. Using open source ATLAS as default implementation of BLAS and LAPACK
  22. Adding new method InnerPoint to get a point matching a system of linear inequations
  23. Fixing bug 522. Ensuring access to valid cells in Matrix Sub.
  24. Added warning message of invalid access at MatDat and PutMatDat

Statistics

  1. New Matrix Quantile API to get a set of quantiles saving time
  2. New built-in function ARIMAAlmagroEval for evaluation of logarithm of likelihood of ARIMA models
  3. Avoiding white noise model in ARMAAutoCovarianzeVector
  4. Using faster LevinsonARMA instead of standard Levinson
  5. Fixed bug calculating logLH in ARIMALevinsonEval. The error did not change the maximum of likelihood.
  6. ARIMA MLE method Estimate is internally evaluated in local scope to avoid export unexpected symbols
  7. BLR:New strategy simulation that offers a much better efficiency without any loss of accuracy or numerical strength ​stdlib/math/stat/models/bayesian/linear/_blrc_kernel.tol
  8. NameBlok's to call R
    1. Rapi: TOL API for R ​stdlib/math/R
    2. Rglpk: an interface to the GNU Linear Programing Kit for solving large-scale linear programming (LP), mixed integer linear programming (MILP) and other related problems ​stdlib/math/optim/LP
    3. Rquadprog: Functions to solve Quadratic Programming Problems (stdlib/math/optim/QP)
    4. RCoda: output analysis and diagnostics for MonteCarlo Markov Chains (MCMC) (stdlib/math/stat/models/bayesian/diagnosis)
    5. New StdLib::Rkde: API between TOL and R KDE (Kernel Density Estimation)
  9. New StdLib::BysMcmc: Generic Bayesian MonteCarlo Markov Chain Methods and applications
  10. New StdLib::BysMcmc:BSR (Bayesian Sparse Regression)
    1. Estimates a sparse linear, constrained or not, regression by MCMC Gibbs method
    2. Can use Boost Spirit to parse a human-like-formated description of the model
    3. Stores data as sparse CHOLMOD matrix and implements all steps
    4. The system is designed to be easily upgraded to handle with non linearities (missing data, ARIMA, decaying transfer functions, heterocedasticity, ...)
    5. BSR new blocks: Input and output missing, ARIMA
    6. Handling with vectorial or time series data in a transparent way
    7. Non linear filters of output and input
    8. Handling with normal prior information about parameters of non linear filters. Prior information is given as a log density additive function
    9. Avoiding to store all nodes to save memory
    10. Enhanced SQL scripting
    11. Storing failed factorization matrix
    12. Allowing fixed sigma in observational nodes
    13. User can specify the number of simulations to be resumed (do.resume>0) or las simulations to be skiped (do.resume<0)
    14. Adding SLICE scalar sampler
    15. New API OneNode to define simple BSR models with just one observational node and optional prior information and order relations
    16. New non linear filter for probit models
    17. Ensuring input matrix has changed. Else it will reuse old factorization
    18. Allowing Inverse-Gamma prior over \sigma^2 $$ of BSR segments
    19. Using faster method LogDens.Almagro.Z_cond_U in draw.ARMS
    20. New BSR specialization DynHlm for Dynamic Hierarchical Linear Model
    21. New general API for importing model definition
    22. New methods BysMcmc::Bsr::Gibbs::Estim and Bsr::Gibbs::Instance(...)::eval
    23. Adding optional time info assigned to each segment of noise
    24. Adding and grouping documental information
    25. New option Text bsr.arimaFilter = BysMcmc::Options::Arima.Filter::FastCholSea; to speed up simulation of ARMA with large period
    26. New option Real bsr.arimaSkipIter to skeep iteratios of ARMA block
  1. New AlgLib Hypothesis testing:
    1. AlgLib.JarqueBeraTest
    2. AlgLib.StudentTtest1
    3. AlgLib.StudentTtest2
    4. AlgLib.UnequalVarianceTTest
    5. AlgLib.OneSampleSignTest
    6. AlgLib.WilcoxonSignedRankTest
    7. AlgLib.MannWhitneyUtest
    8. AlgLib.VarianceChiSquareTest
    9. AlgLib.VarianceFTest
    10. AlgLib.PearsonCorrelationSignificance
    11. AlgLib.SpearmanRankCorrelationSignificance
    12. AlgLib.Mann.Whitney.U.test

TCL

  1. Tcl_EvalEx, accept a Set as a Tcl command
  2. Supporting NameBlock's

Last modified 16 years ago Last modified on Mar 11, 2009, 6:46:08 PM