What's new in Analytica 4.0?
The forthcoming new 4.0 release of Analytica has dozens of major new features and hundreds of smaller enhancements:
Contents
Output formats
Graphs and Charts
We have completely rewritten Analytica's charting and graphing engine, adding a wide range of new styles and options, smarter default behavior, and the flexibility to create much more elegant charts.
- Chart styles: New styles include stacked bars, filled areas, and using symbol size and color to indicate extra dimensions. You can flip x and y axes -- e.g. to create horizontal bars for tornado diagrams.
- Axes: Log scales and reversed scales. Flexible date formats.
Label resizing and rotation for clear labelling without overlaps.
- Appearance: Color and pattern for chart backgrounds, including color gradients. Line widths. Font face, size, and color for titles, axis labels, and key. Bars with 3D effects -- cylinders or boxes.
- Chart style -- e.g. continuous or categorical -- associates with the chart variable and indexes, so it switches appropriately when you pivot the chart or change uncertainty views. You can save the range and other properties of an axis as the default for the underlying index variable, so these properties will apply to all graphs using that index.
- You can save Graph Style Templates to reuse a set of styles and appearance in other graphs.
- You can save a graph in a variety of standard file formats, including JPEG, PNG, BMP, TIFF, and EMF, using Export from the FIle menu. For more, see Graph settings
Tables
- You can include dropdown menus in an Edit table, letting end users select from a predefined set of options in a tabular format, by using the Choice() function in cells of the table.
- Select Find... from Object menu, or press control-F, to search for text in a table.
- Dozens of new mouse and keyboard shortcuts let you navigate and select regions from tables -- the same as in Excel. For example, use the mouse wheel to scroll. Drag the mouse from one cell to another to select the rectangular region between them.
- When you reopen a table or graph, Analytica remembers all settings from the last time it was displayed, including pivoting (index chosen for horizontal and vertical dimensions), slicers (values to show for other dimensions).
- Copy and paste from a table now uses full numerical precision. You can control the order of indexes in a copy of a multidimensional table by reordering its slicer indexes.
- Exported table data includes any external variable columns or coordinate index columns.
- With Analytica Enterprise or Power Player, you can now scroll through Huge index dimensions with over 32,000 elements.
- The Subtable function lets you define a variable as part of another Edit table. Any change to an element of a Subtable is reflected in its parent table, and vice versa.
Influence Diagrams
- Make neater diagrams: New options in the Diagram menu include Align nodes to top, left, bottom, or center; Make same size for height, width, or both; and Space evenly horizontally or vertically.
- URLs in a text node appear as hyperlinks in the diagram.
- Optional red flag in node shows which objects have Descriptions.
- When you've selected a Button object, Control-e now opens its Script. As before, for a Variable or Function, Control-e opens its Definition.
- The mouse wheel scrolls a diagram vertically or, pressing the control key, horizontally.
Non-US text and numbers
Recognizing that nearly half Analytica users are outside the US -- Analytica is finally getting a bit less US-centric:
- Date and currency formats: The Number format dialog now lets you select a much wider variety of formats, including non-US dates, currency symbols, and negative numbers. Date formats respond to the Windows region setting.
- You can now paste text containing accents, symbols, and characters from non-English languages (for ascii>127) into Object windows and diagram nodes.
- We know we still have some distance to go towards full internationalization, but this is a start!
The Application
- Auto save: Analytica automatically saves each model change to a backup file. So you can no longer lose changes after a software or hardware crash. Next time you start the model after an unplanned exit, it asks if you want to continue using the changes from the backup or revert to the previously saved version.
- Analytica now shares CPU nicely with other applications. Earlier versions hogged 100% of CPU cycles when the Analytica was in focus -- i.e. its application window is on top.
- It can now read model files that use LF (linefeed) as the end-of-line character, instead of CR (carriage return). This convention is standard for Unix and Linux, and sometimes happens to files that are stored or emailed from a Unix server.
- If your computer has multiple screens to show a larger desktop, you can now edit Analytica diagrams on any screen or across multiple screens.
- Analytica recovers more gracefully when a calculation causes it to run out of memory.
- Analytica 4.0 can handle up to 32,000 objects -- variables, modules, functions, etc. --- double that in previous releases.
New and Enhanced Functions
- Date functions provide much more flexibility for computing with dates, to go along with the much greater flexibility in displaying dates from the Number format dialog from Result menu.
- MdxQuery supports the standard MDX language for querying and writing to multidimensional OLAP hypercube databases, such as Microsoft SQL Server Analysis Services. It opens up Analytica for integration with Business Intelligence applications. (Analytica Enterprise only).
Probability distribution functions and statistical functions
- You can specify an array of independent probability distributions by specifying an Over index i to any probability distribution function -- for example, Normal(0, 1, Over: i).
- By default Statistical functions, such as Mean, Variance, Frequency, Correlation, and many others, operate over the Run Index: Thus, Mean(x) gives the mean over a random Monte Carlo sample from x, indexed by Run. You can now specify a different index to operate over -- for example, Mean(x, k), useful for statistical analysis of an input dataset x indexed by k.
- Lognormal will work with parameters mean and stddev (standard deviation), instead of the default median and gsdev (geometric standard deviation). In fact, it will work given any pair of parameters from median, gsdev, mean, and stddev.
- Truncate(x,min,max) : Now accepts min and/or max thresholds parameters (both optional, but at least one required). It now preserves sample ordering, and so rank correlation.
- Uniform(min, max, integer) adds an optional parameter, integer to specify that values be integers in the range.
- Random( ) generates single random variate using a specified distribution.
- Shuffle(x, i) randomly shuffles an array.
- Pdf and Cdf return the estimated probability density (or mass) function or cumulative distribution function as arrays, so the model can work with them. Previously, they were only available to view as a graph or table.
- Importance weighting is a way to get more information from fewer Monte Carlo (or Latin hypercube) samples. Instead of treating all samples as equally likely, you can set a weighting vector, SampleWeighting. For example, in a risk analysis with a low probability of extremely bad outcomes, you may want to increase the number of samples that lead to bad outcomes to get more detail in that region with correspondingly smaller weight to keep the distributions unbiased. You can also modify the SampleWeighting to reflect different input distributions and rapidly see the effects on the resulting distributions without having to rerun the simulation. Every probability distribution and statistical function uses the importance weights. The default uses equal weights, as before, so you don't have to worry about importance sampling unless you want to use it.
- System variable 'IsSampleEvalMode returns true when evaluation is probabilistic, false otherwise.
- Several distribution functions much faster, especially with uncertain parameters (hierarchical distribution). Gamma, Binomial, GammaIInv are more accurate for extremely large or small parameter values.
System Functions
- RunConsoleProcess lets you run another application from Analytica. It can pass data via a parameter or a file. It can run the process in parallel or wait for result to be returned as a file.
- GetRegistryValue( ) gives specified values from the computer registry, such as the name of default directory for model or datafiles.
- ShowPdfFile( ) shows an Adobe PDF file -- for example, to open PDF documentation for a model.
- Function AnalyticaLicenseInfo can return the expiration date, purchaser ID, user ID of the license, whether it has Educational, Beta status, or supports the Optimizer. System variable AnalyticaEdition gives the Edition (Player, Professional, Enterprise, etc.)
Array functions
- The new operator @J returns the position of each element in index J -- i.e. the integers from 1 to Size(J) indexed by J. You can also use it in a subscript: A[@J = 2] is equivalent to Slice(A, J, 2).
- PositionInIndex(a, u, i) returns the position n (an integer 1 to Size(i)) in index i for which a[i=n] = u. It is the equivalent of SubIndex, except that it returns a position instead of index value. Useful when i may contain duplicate values.
- The subscript syntax now lets you to subscript an expression -- e.g., (A+B)[I=x].
- You can assign to a slice of a local variable x -- e.g. x[i=y] := b -- letting you implement some algorithms much more efficiently.
- The new function Argmin(x, i) gives the last value of i for which x is at a minimum over i, like the existing function ArgMax(x, i). These functions can now find a min or max over one or more dimensions by specifying multiple indexes, and can return index value or position.
- Rank now lets you specify mid, lower, or upper rank in the event of a tie.
- Area adds an optional fifth parameter: Area(R,I,X1,X2,J)
- MdTable now lets you specify the first N columns of X as coordinates and the rest as measures, as used in a fact table, the format used to specify OLAP hypercubes. It also supports a user-defined conglomeration function for combining elements.
- Functions ArgMax, SubIndex, LinearInterp, StepInterp and CubicInterp are now in the Array library (formerly in Special).
- The new IndexValue function provides access to a self-indexed variable's index value from a value context.
Special Functions
- Added inverse functions ArcCos, ArcSin, ArcTan, and hyperbolic functions CosH, SinH, and TanH to the standard Sin, Cos, and Tan. All use degrees, not radians.
- TypeOf(X): Returns the type of each atom in X as a text.
- A VarTerm is a pointer to an object, such as a Variable or Module. Varterms lets you write functions and models that navigate and change a model by working with Objects -- for example, to find the inputs or earlier ancestors of a variable. They allow an Analytica programmer to write lots of nifty functions, which the rest of us will benefit from.
Language Features
- Function Parameter Qualifiers specify what type, dimensions, or range of values are expected for parameters. New qualifiers include "..." meaning one or more parameters of the same type, NonNegative, ContextSamp' (the default), and Optional indexes You can also supply default values for optional parameters, to be used when the parameter is absent.
- A variable used as an IndexType parameter gives its Indexval if it has a self-index, when evaluated in the function.
- The Domain attribute specifies the set of acceptable values of a variable. When displaying a probability distribution, Analytica uses the Domain to decide whether to generate a probability density or mass function, according to whether the Domain is continuous or discrete. If the Domain is Automatic, Analytica guesses from the actual values (usually correctly). If the Domain is Index, it gets the domain as a list of values from an Index variable.
- 0*NaN and 0*INF now give a warning and return NaN, consistent with the IEEE 754 and SANE arithmetic standards. Earlier releases simply returned 0.
- Release-gated comments. The syntax {!40000|expr} will be treated as a comment in releases prior to 4.0.0, but as the expression expr in 4.0.0 or after. {!-40001|expr} or {!40000-40001|expr} acts as expr in releases 4.0.0 through 4.0.1, but as a comment in earlier or later releases. This syntax is accepted in both typescript and expressions.
Libraries
- New distributions in "Multivariate Distributions.ana"
- MultiUniform
- Uniform_Spherical
- Distribution variations
(Note: these distributions can also be used within the Random function for single variate generation)
- Smooth_fractile
- Warp_dist
- Erlang
- Pareto
- Rayleigh
- Lorenzian
- NegBinomial
- InverseGaussian
- Wald
- Functions in the new "Generalized Regression.ana" library
- Logistic_Regression
- Probit_Regression
(These require Analytica Optimizer)
The Optimizer
The Analytica Optimizer offers powerful solver engines, licensed from Frontline Systems, to solve linear programms (LP) and mixed integer programs (MIP), quadratic programs (QP), and general nonlinear programs (NLP). The Optimizer is an optional extension to Analytica Enterprise, Power Player and ADE.
- Analytica Optimizer 4.0 uses the new 7.0 release of Frontline Systems Solver, which offers significant improvements in performance and functionality.
- The new Grouped integer variable type lets you define a set of N integer decision variables, where a solution must assign a different integer from 1 to N to each of the variables in the group.
- QpDefine now supports quadratic constraints. (Before QP problems supported quadratic objective but constraints had to be linear.)
- The new frontline comes with some licensing restrictions that were not present previously. These are detailed at Analytica Optimizer Licensing.
- If your optimization problems are really large, you can use yet more powerful add-on solver engines, including OptQuest, Knitro NLP, Mosek SOCP and NLP, and Xpress LP, QP and MIP (priced separately).
- Function SolverInfo provides information about any installed solver engine, including engine type, all parameters, search-control settings, and their allowed ranges.
- ObjNl and LhsNl parameters add the 'Q' option to specify quadratic dependence.
- Function LpFindIIS lets you relax variable bounds (not just constraints, as before). Two new parameters control this, and whether a subset of constraints is returned or a new LP.
- LpWrite, LpRead, and LpWriteIIS support three file formats: LP, MPS, and LPFML to help you document and diagnose solver formulation and behavior.
- Tracefile helps debug NLP optimizations.
- Hooks for array abstraction in NLPs: Enhancements to NlpDefine give more flexibility to structure a large model so NLP optimization array abstracts properly -- for example, for probabilistic or WhatIf analysis with NLP.
- Almost all parameters to NlpDefine are now optional, making it simpler to define an optimization problem with a scalar decision variable, only zero or one constraints, or to find feasible solutions only.
- To speed up NLP, you can give NlpDefine an explicit expression to compute the Hessian -- second derivatives of the objective and LHS.
Enable comment auto-refresher