The Regression Shelf

VECTOR Vol.7 No.4

concerns itself with the statistical procedures appropriate to estimating the unknown linear regression parameters a,b, ...d, together with the nuisance/ scale parameter k. As such it is merely one example of the basic paradigm Observation = Signal + Noise

Given a collection of values of the response variable, and the associated values of the predictor variables, the machinery of statistical linear models concerns itself with: (a) Choosing appropriate models;

(b) chosen model;

{c) Estimating the parameters;

(d) Using the model for prediction.

Hence, functions are required to enable the user to:

(a) Specify a model;

(b) Estimate the parameters of the model;

() Test hypotheses concerning that model;

(A) Perform diagnostic tests and graphical plots to investigate the validity and appropriateness of the model.

Generalized Linear Models Over the last twenty years the success of statistical linear models has been strengthened by relaxing two of their basic assumptions. First the assumption that the response variable has a Normal distribution is relaxed. Alternatives (from the Exponential family’) such as Poisson, Binomial, and Gamma distributions are allowed extending the scope considerably. For example if Y is a count of radioactive particles emitted from a source time period of size x, then, rather than assuming that y had a Normal distribution with mean atbx, it is likely to be more appropriate (since the y values are counts) to assume that Y has a Poisson distribution with mean atbx - hence the idea of Poisson Regression’.

The second assumption that is relaxed is the idea that the linear predictor’, i.e. the function atbutcv+..dw, need not specify the expected response E(Y). Rather, the linear predictor specifies a function of the expected response. For example in Poisson regression we may assume that atbUteV+.dW = log(E(Y)). This

41

VECTOR Vol.7 No.4

concept of a Link Function’ extends the scope even further. Generalized Linear Modelling is now an accepted part of an applied statistician’s armoury of tools. It is not surprising that the success of the package GLIM has resulted in the inclusion of such tools in many other statistical packages, such as SAS, GENSTAT and now ASL!

The Facilities Within the Regression Shelf The first, draft, version of the Regression Shelf was distributed to interested parties during and after the second ASL conference in October 1990. Feedback gained result exercise has resulted in some modifications to the shelf which IJ shall detail below.

In the remainder of this article, I shall introduce the facilities offered in the Regression Shelf and future article I shall discuss of the functions supplied within the shelf in more detail.

A Database of Examples In order that the Regression Shelf may be tried and tested, a number of databases have been provided in a separate workspace. Each database is stored under a variable name, which specifies a character matrix with a simple but specific format:

1 Introductory Lines, specifying appropriate references

2 Blank Line

Names; one per line with at least one space between the name and the (tabulated) start of the explanation of the variable names

4 Blank line

5 Lines of Data

As an example consider the data set HOUSEPRICES

HOUSEPRICES DATA, SOURCE: Genstat 5 An Introduction, by Lane, Galwey and Alvey, Page 55. 20 similar houses for sale in Harpenden in 1977

VECTOR Vol.7 No.4

PRICE House price in pounds sterling SPACE Total floor space of the house in square-metres GARDEN Area of the garden in square-metres AGE Age of the house in years

11500 88 15500 154 245 70 12950 66 14000 121 180 43 16500 17 17600 23 12450 112 90 52 18500 124 120 10 14900 141 16250 149 350 36 18400 170 320 7 11950 93 450 19 10400 62 17250 12 13450 148 190 23 10950 128 160 75 13950 152 11500 401 450 42 17500 145 275 0

Functions are supplied to allow you access to the explanation, the names and the data. A function is also provided to extract the data into numerical form attached to the names provided.

EXTRACT HOUSEPRICES PRICE +SPACE +GARDEN +AGE

(N.B. the explicit result of using EXTRACT is a string variable containing the variable names, separated by plus signs - this is useful for feeding into the regression routines. But in addition of course, PRICE, SPACE, GARDEN and AGE are now in the workspace.)

An Introductory Example

A simple example of the basic regression directives using the database on HOUSEPRICES as described above gives a flavour of the Regression Shelf.

EXTRAC? HOUSEPRICES PRICE +SPACE +GARDEN +AGE

43

VECTOR Vol.7 No.4

The aim of the use of regression here is to attempt to explain the variation in house prices in terms of the information about the houses contained in the variables SPACE, GARDEN and AGE. Intuitively, a large house with a large garden will command a higher price than a smaller house. As for age, perhaps its influence on house price is debatable.

Since the object is to predict house prices, the variable PRICE, or some transformation of it must be declared to be the y- variable’. This is done using the directive YVAR.

YVAR 'PRICE/1000"' Y variable is PRICE/1000 units set

To predict this variable, we have the choice of using all or some of the variables SPACE, GARDEN, AGE, together with any variables that can be constructed using them, such as for example SPACExSPACE, SPACEXxGARDEN, +AGE, etcetera. In addition, we usually require a constant term, which is denoted by Gt . Let's try using just the variable SPACE, and of course a constant. Predicting PRICE using GM+SPACE  is then equivalent to finding the regression line of PRICE on SPACE.

The linear predictor is specified by using the directive LP.

LP 'GM+SPACE!'

To fit, follow this with the directive FIT.

FIT

Response Variable is PRICE/1000 Fitted Model is GM+SPACE

source SS af ms

Due to model 66.223 1 66,223 Residual 46.089 17 2.711

Total Corrected 112.306 18 6.239

Percentage Variation Accounted for = 58.97 F-statistic = 24.43 p~value= +00

VECTOR

This table presents a summary of the ability of the chosen linear predictor to predict PRICE+1000. It takes the form of an ANOVA  table which breaks down the variation in the y-variable  (in this case 112.306), into that explained by the model (66.223) and the residual variation of the data about the fitted regression line (46.083). Hence we can see that using SPACE as a predictor variable explains 100x66.223+112.306 or 58.97% of the variation.

This summary does not however give the estimated coefficients of the predictor variables (GM and SPACE). The directive DISPLAY does.

DISPLAY

Var Est st

GM 2.576 SPACE 086

Now it should be clear from this simple one-variable model how to fit more complicated models. Let’s use all the supplied variables, for example.

LP 'GM+SPACE+GARDEN+A FIT

Response Variable is PRICE Fitted Model is GM+SPACE+G.

source ss

Due to model 96.986 Residual 45.321

Potal Corrected 112.306

Percentage Variation Accou F-statistic * p-value

DISPLAY

GM 6.079 SPACE 077 GARDEN -.001 AGE ~.048 d Err t-stat p-val

2.410 4.4 300 017 4.9 «G00

GE!

41000 ARDEN+AGE

df

3 15

18

nted for * 86,36 31.65 +00

Std Err t-stat

1.769 3.4 2011 7.0 2003 -.3 010 ~4.9

45

VECTOR Vol.7 No.4

From these results, we can see that with the additional variables, the total prediction capability is 86.36%. Note from the estimates, that the coefficient of AGE is negative (older houses tend to be sold for less than younger houses, other things being equal), and that the coefficient of GARDEN is less than its standard error, indicating that its predictive capabilities in the presence of the other variables is very small.

Fitting a model such as the above is fraught with problems. There are so many assumptions involved which should be checked. There is a vast array of suggested techniques for doing this and the regression shelf includes a number of the most popular regression diagnostics as well as a repertoire of functions to run and test more complex models. In the next article on the Regression Shelf I shall illustrate the full scope of the code supplied in the Regression Shelf.

Since January 1991, ] have been working on a complete revision of the Regression Shelf. The basic weighted least-squares routine has been changed from on Householder’s method to a version of Beaton’s method.

The criticism raised by early testers that the Shelf was too packaged’ has been considered seriously. However it does seem that it is difficult to provide useful functions including output without a good deal of packaging. So, in the revision, we have tried to reach a compromise - the package is there, but it is more modular, with core functions, such as the weighted-least-squares routine available to the user directly. With of good documentation, a user has a choice of using the Regression package or choosing core routines for his own use.

The numerical computations have now been checked by reference to published examples in the GLIM manual and in Aitken’s book on Generalised Linear Modelling. Additional functions have been added including Summary Statistics, correlation matrices (both for estimates and regression variables), and Box-Cox loglikelihood for power transformation.

46

VECTOR Vol.7 No.4

Sterling’s Function: a Case Study

by D R Appleton

Theme Several different APL functions, for the evaluation of Stirling numbers of the second kind, are presented. Their advantages and disadvantages are discussed.

Exposition A colleague doing some theoretical statistical work on the analysis of an experiment which had resulted in counts from the truncated Poisson distribution being observed, recently asked me had an APL function to evaluate the Stirling numbers of the second kind. I did not have such a function, told me they were defined for positive integers n and t by

a formula obtainable from Abramowitz and Stegun (1964). With APL it is easy to turn this expression into executable form, for example: < STIRI1 >

However this function can be improved: it ignores the fact that the first term is identically zero, it does it clear that the result is always positive, and APL has a more concise way of summing alternating series. A better function is therefore given by < STIR2 >

It may or may not be numerically preferable to factorise out the n! and use < STIR3 >

Of course APL’s power comes from its ability to handle several different values simultaneously. We could easily have a vector for T. < STIRG >

47

VECTOR Vol.7 No.4

However, it is likely that the values of 7 we would require, at least to form a table of the Stirling numbers (which might be useful in the long run to save calculation) would be all the positive integers up to T. This leads to < STIRS >

and we could have, had we wished, developed STIR2 to this form instead of STIR3.

If we wish to produce and save the numbers we may build STIRS5 into

<

STIRG

>

It is now time to look at the table of numbers this function produces; this is shown in Table 1 for W=6 and T=6. It becomes very clear function which wastes its time calculating values for N>f is unsatisfactory. This, of course, should have become evident if a little more algebra had been done before turning to programming, or if STIR2, say, had been properly tested, but the functions are so simple it hardly seemed necessary to test them all individually! Table 6 2 oO as 31 n 3 25 30 oO 1 10 61 5 Q °o oO 1s 6 Oo o o oO

Development It is important to define more coherently the problem we are tackling. We now want to produce a table of Stirling numbers in which the element at row column p contains Op, and to solve the practical problem underlying the computing problem we would like n to take values up to at least 100 and p to reach at least 2n for as many values of n as possible. This means rewriting STIRS as

< STIR7T >

48

VECTOR Vol.7 No.4

Running this in a loop produces Table 2 for N=8 and P=6. The magnitude of the values appearing in the table gives rise to concern for the function: when will it overflow? This is certainly a matter which will have to be investigated, but there are other interesting aspects to the table. Table 2 seeeeee P 1 1L 415 31 63 227 3 6 301 966 3025 nia 10 65 350 1701 7770 34105 5 15 140 1050 6951 42525 246738 6 21 266 2646 22827 179487 1323652 7 28 462 5880 63987 627396 5715424 a 36 750 11990 159027 1999612 20912320

From row 2

it is clear that 024, = 2 03,, + 1, and from row 3 it is not too difficult to deduce that 03,43 = 3 0342 + 0242, and hence to discover the recurrence relation o,=n oF oF),

This enables us to write a function without calculating the factorials or binomial coefficients in our original definition. < STIRG

>

But what is this? An APL function with a double loop? Surely that is unnecessary. Instead of working a row at a time we must evaluate the table by columns. < STIR

>

Now we have a much faster function (taking about a quarter of the time STIR8 does for N=80 and P=80), which is approaching its final stage, but let us look more closely at the table and the function we have written. Column 1 is just +\iN and column 2 is therefore +\(1N)*+\1¥. Indeed column P is 214+(7xP)p'tx+\(1N)" an expression so concise it is hardly worth writing a function for. Notice that #14 (9xP)p'x+\(1N)"' prints out the entire table given by STIR9, although it transposes it and prints each line to a different

49

VECTOR Vol.7 No.4

format. This is avoidable, for small values of WN and P, by 4224+((12xP)p'xsQ+«10 0¥F+\K'),'+iN' but we wish to store the table, not print it, a task which is done by our final version

< STIR >

which is slightly faster even than STIR9.

Recapitulation We have seen how it is worthwhile to look at the problem from a theoretical point of view so that we may obtain different algorithms for calculation. Even with an apparently obvious formula, simply translated into APL, it may be more appropriate to use a recurrence relation. This is not only for reasons of time, which are not of vital importance if the values obtained are to be stored in any case, but for reasons of numerical accuracy. In passing we suggested that STIR2, which involves the calculation of a factorial and a set of binomial coefficients, might have different numerical properties from STIR3 which involves two sets of factorials.

To investigate this suppose N=50 and 7=60, the sort of numbers which might indeed arise as the result of a small experiment. STIR1 and STIR2 give 0% = 2.02x10% while STIR3 and STIR4 give 5.06x10%. The sensitivity of each function to the order of summing the series can be looked at by changing ~/ in STIR2 and STIR3 to -/%; the results change to 1.711026 and 7.09x102 respectively. With STIR4 we can even alter the calculated value of $ depending on which values other than 60 we include in vector 7. STIR (and STIR9 which uses the same arithmetic) gives 9.55x1024 for 6%, which I believe to be correct. Not only is the algorithm contained in #14(7*P)p'x+\ (iN)! remarkably succinct, it is numerically preferable to the others. STIR works whenever 350>N+2*P and occasionally outside that range. When the function fails it is sometimes because the workspace becomes full, and sometimes because a domain error occurs. The figure’ shows an isometric plot of the common logarithm of the Sterling numbers of the second kind, for values of n and t-n up to 80. The values given above, except of 9.55x10% , are system- dependent; the ones quoted were obtained from IBM APL on a Personal Computer, the same phenomenon occurs with different values using TRYAPL2. Other systems give a domain error instead of inaccurate results.

50

VECTOR Vol.7 No.4

Coda The style of programming illustrated in STIR can be utilised in many instances of recursion, though the Stirling numbers of the second kind may be the most elegant. Readers might like to deduce what the following expressions evaluate.

< TXTG > < TXT9 > < TXT10 >

Reference [1] Abramowitz, M. & Stegun, IA (eds). Handbook of mathematical functions with formulas, graphs, and mathematical tables. Washington: US Govt. Print. Off. 1964.

51

VECTOR Vol.7 No.4

APL Product Guide

Compiled by Alison Chatterton

VECTOR’s exclusive APL Product Guide aims to provide readers with useful information about sources of APL hardware, software and services. We welcome any comments readers may its usefulness and any suggestions for improvements.

We do depend on the alacrity of suppliers to keep us informed about their products so that we can update the Guide for each issue of VECTOR. Any are not included in the Guide should contact me to get their free entry - see address below.

We reserve the right material supplied for reasons of space or to ensure a fair market coverage.

The listings are not restricted to UK companies and international suppliers are welcome to take advantage of these pages. Where no UK distributor has yet been appointed, the vendor should indicate whether this is imminent or whether approaches for representation by existing companies are welcomed.

For convenience to readers, the product list has been divided into the following groups:

* Complete APL Systems (Hardware & Software) * APL Timesharing Services * APL Interpreters * APL Visual Display Units * APL character set printers * APL-based packages * APL Consultancy * APL Training Courses * Other services * Vendor addresses

Every effort has been made to avoid errors in these listings but no responsibility can be taken by the working group for mistakes or omissions.

Note: poa  indicates price on application’.

All contributions to the APL Product Guide should be sent to Alison Chatterton, at the address on the inside back cover.

52

VECTOR Vol.7 No.4

COMPLETE APL SYSTEMS

COMPANY PRODUCT PRICES(£)

Active Workspace Lid AWLA86 4,450

AWL 386 3,095+

APL People IBM PCs & compatibles poa

Dyadic IBM R§/6000 MDS20 11,736

[8M RS/6000 MD320 13,817

IBM RS/6000 MD320 22,656

IBM RS/6000 MD520 37,118

1BM RS/6000 MDS30 72,054

(eM RSy6000 MDS40 122,842

MicroAPL Aurora 20,000+

Spectrum 7,000+

MTC. 386sx AT 2600+

APL TIMESHARING SERVICES

COMPANY PRODUCT PRICES(£)

REUTER:FILE SHARP APL poa

Uniware APL*PLUS call

APL INTERPRETERS

COMPANY PRODUCT PRICES(£)

Active Workspace Ltd DYALOG APL DOS 986

APL Software APL*Plus/PC Release 9 535

Run-time poa DETAILS

485 based 25MHz PC, 140MB Disk, 4MB RAM, VGA Colour. (Ine. 1 year on site maint.)

386 based 25 & 33MHz PC, 140MB Disk, 4MB RAM, VGA Cofour. (inc. 4 yaar on site maint)

Includes PC, meno/cotour monitor, APL interpreter, operating system optional printars, graphics boards, addltional memory etc.

APL POWERstation (Grayscale) 27.5 MIPS, 7.4 Milops RISC Processor 8Mb RAM, 120Mb Disk 19" 1280x1024 Grayscale Graph Display AIX, OSF Motif, Dyalog APL (1-user)

APL POWERSstation (Colour) 27.5 MIPS, RISC Processor 8Mb RAM, 120Mb Disk 16" 1280x1024 Colour Graphics Display AIX, OSF Motif, Dyalog APL (1-user)

Advanced APL POWERstation 27.5 MIPS, 7.4 Mflops RISC Processor 16Mb RAM, 320Mb Disk, 150Mb Tape. 16" 1280x1024 Colour Graphics Display AIX, OSF Motif, Dyalog APL (1-user)

APL POWERsystem (8-users) 27.5 MIPS, 7.4 Milops RISC Processor 16Mb RAM, 320Mb Disk, 150Mb Tape CD-ROM Drive, 16 Ports AIX, Dyalog APL (2-9 user licence}

APL POWERsystem (16-usars) 34.5 MIPS, RISC Processor 32Mb RAM, 1.24Gb Disk, 2.3&b Tape CD-ROM Drive, 18 Ports AIX, Dyalog APL (8+ user licence)

APL POWERsystam (82-users) 41 MIPS, 19 Mflops RISC Processor 64Mb RAM, 1,7Gb Disk, 2.9Gb Tape CD-ROM Drive, 92 Ports AIX, Dyalog APL (8+ user licanca)

Mult-user APL computer using 68020 CPU. Std. configuration 2Mb RAM, 16 AS232 ports, 68 Mb hard disc, 720K diskette

Expandable mult-user APL computer using Motorola 68000. Std. conguration 1 Mb RAM, 12/36 Mb disc, 12 ports.

Entry fevel 92-bit compact AT with DYALOG APL. 2MB ram, VGA, 40MB hard disk, 1.44MB floppy disk, 102 kay keyboard, monitor, mouse, MS-DOS 4.01, WORKS and WINDOWS-388. Many expansion options availabla,

DETAILS

Intemational Network application systems and public databases.

STSC's malntrame service

DETAILS Dyadles PC 988 APL interpreter

STSC's APL for IBM PC, PC/AT and PS2. Upgrades from earlier Releases also avaliable.

Closed version of APL*Plus/PC which prevents user exposure to APL.

53

VECTOR Vol.7 No.4

APL*Ptus Ik 1600 Incorporates mainframe features & performance In a version of APL for the PC

Run-time poa

Dyalog APL 1000-10,000 2nd generation APL for UnIx systems

APL2IPC 376 IBM's APL 2 far the PC.

Cocking/Drury APL*PLUS PC Rel 10 fullfeatured API for IBM's and compatibles - Version 10 includes the quad-NA facility to Interface to non-APL software, support for MS Windows and mouse devices. The User- command processor has been built In to the interpreter. Upgrades to version 10 are avaltable from Version 9 and earlier releases,

APL*PLUS PG Run-Time pea Closed version of the Interpreter for developers, prevents user exposure to APL,

APL*PLUS I! System 1600 High powered APL interpreter for the 80386 chip. Price includes one yeas malntenance and free upgrades - volume discounts VERSION 8 NOW AVAILABLE, APL*PLUS II Developer System poa the features of the APL*PLUS II System plus unlimited free run-times to enable developers to distribute thelr applications. APL*PLUS UNX poa -STSC s 2nd generatton APL for all major Unix computers and workstations. Version 4 for existing platforms and for the IBM RS/6000 available from January.

APL*PLUS VMS generation APL for DEC VAX computers running under MS.

APL*PLUS Malntrame poa Enhances VS APL with many high performance, high productivity features. For VM/CMS and MVS/TSO offers simple upgrade fram VS APL.

Dyadic Oyalog APL. tor DOS}386 generation APL for DOS.Runs In 32-bit mode, supports very targe workspaces. Unique window-based APL, Development Environment and 6SM Screen Manager. Requires 386/486 based PC or PS/2, at least 2Mb RAM, EGA or VGA, DOS 3.9 of later.

Oyalog APL for Unix Systems995-12,000 Second generation APL for Unix systems. Avallable tor Altos, Apollo, Bull, Dec, HP, IBM 6150, 18M FIS/6000, Masscomp, Pyramid, NCR, Sun and Unisys machines, and for PCs and POj2s running Xenix Oracie interface avallable for IBM, Sun and Xenix versions. FAPL Lid JAPL/PC or RML Nimbus 4.50 1SO conforming interpreter. Supplied onty with manual, (see Other Products’ for accompanying books} I-APUBBC 4.50 Asabove

FAPUArchimedes 4.50 As above

IBM UK IBM PC APL2 948 APL for the IBM PC, Program §799-PGG/1, PRPQ number RJ- 041. From alll IBM dealers,

MicroAPL, APL.68000 Leve! | 2000 First generation APL with numerous enhancements, Multl-user version (Unix, Mirage, MCS),

APL.68000 Level It 2500 Second generation APL. Nested arrays, user defined operators, selective specification etc. Multl-user version (Unix, Mirage, MCs)

APL.68000 Level |

QL 87 First generation APL, Single user, full windowing interface, software floating point support,

Mac, Amiga 260 First generation APL, Single user, full windowing Interface, hardware floating point,

APL.68000 Level tl

st 170 Second generation APL. Full windowing Interface, software floating point support.

Amiga generation APL. Full windowing interface, Hardware and software Noating polnt support.

54

VECTOR Vol.7 No.4

Mac 820

APL"PLUS Rel 10 450 APL*PLUS II V3.0 1398 REUTER:FILE SHARP APL poa Uniware APL*PLUS/PG 495 Run-Time call

APL&PLUS/UNX call APLPLUS 1} call

APL VISUAL DISPLAY UNITS COMPANY PRODUCT PRICES(£) APL People IBM & compatibles poa Dyadic |BM 3181 599 IBM

6184 1,228

General Software Mellordata poa Shangell HDS3200/10 APL 965

HDS3200/25 APL. 1065 HOS3200/95 APL 1268 HDS3200/5C APL 1395

APL PRINTERS COMPANY PRODUCT PRICES(2) APL People Epson series 200 Quen-data & Qume etc 500 Dyadic Various poa MicroAPL Datasoutn DS180+ 4,195 Philips GP480 2497 Qume Latterproz0 49

APL PACKAGES COMPANY PRODUCT PRICES(£) Active Workspace Lid Syndicate Manager poa

APL-385, APL-385 FSM-385, DRAW-385 50(PC),125(mf) Second generation APL. Full windowing interface. Hardware and software floating point support.

for IBM mainframes STSC's full for IBM PC/XT/AT, Compag, Ollverti. Closed version of APL*PLUS/PC which prevents User exposure to APL. STSC’s full for UNIX based computers STSC's full feature APL for 386 machines.

DETAILS IBM and budget  APL VDUs - monochrome/colour/graphics. Monochrome APL/ASCII vdu with APL keyboard. Supports downloadad Dyalog APL font. Colour APL/ASCIl vdu with APL keyboard. Supports downloaded Dyalog APL. font.

18" screen, 8 page memory, windows, 80/132 columns, full overstrika. Multi-host multi-session support, ANSI X3.64, DEC Tektronix 4010/4014 1024 x 390 resolution. As above plus switchable 26 or 50 Ilne screen. 75 Hz refresh. Resolution 1024 x 780 in graphics mode. As HDS3200/26 plus local pan & zoom. 14” 75 Hz refresh. 8 or 16 colours fram palette o 256. Display memory 96 lines of 80 or 132 columns. APL processing & keyboard with full overstrike. Windows, muiti-host, ANSI X3.64, DEC VT220, VT100, VT52. emulation.

DETAILS Inexpensive dot-matrix and NLQ printers Daisy-wheel printers Range of APL printers avallabie. See Datatrade entry Matrix printer with letter & draft quallty and APL (480 cps). APUASCIl Daisy-wheei printer

DETAILS Lloyd's managing agent's syndicate / company accounting system. Stamp & Personal accounts (inc, Flun offs) including ... Screen development Screen design

55

VECTOR Vol.7 No.4

APL Software Ltd (mainframe)

(microcomputer)

APLIMPETUS Cocking/Drury (for VSAPL)

(for APL2)

(for PC's)

H.M.W.

HRH Systems

{NFOSTROY

Iinterprocess

(mainframe) DB-385 GEN-385 RDS iPLS REGGPAK

POWERTOOLS: REGGPAK RDS Impetus

E’MENTS & SHAREFILE COMPILER FILEPRINT FILECONVERT FILEMANAGER TOOLS + UTILITIES DATAPORT

SHAREFILE/AP

FMT WSDOC FILEMANAGER

APL*PLUS PC Tools: IRMA Module FIN & STAT. LIBRARY SPREADSHEET MGR XTRA Arbitrage Basket Menu-Bar APL Utiltles

Russtficator

|EDIT

AFM

Format FSM124 poa

295

990 poa

poa

poa

poa poa poa poa

poa

poa

276

250 150 poa

poa

poa

1900-3200

8200-9800

1660 1650 Relational W.S. Miscellaneous Utilities

Relation Data Base System Project Management System Regression Analysis Package

Assembler written replacement function far cammonty used CPU-consuming APL functions, includes a Forms Processor. Regrassion Analysis Package Relational Database System Hierarchical Planning System

Component files, quad- functions & nested arrays for VSAPL. under VM/CMS & MYS/TSO The First APL compllert Print APL. component files Converts non-APL files to APL Extends APL primitives to database management APL Software development tools Information Centra spreadsheet incorporating data exchange between APL, FOCUS, IFPS, SAS, APL/DI. ADRSII, Lotus?23, Visicale, Multiplan & DIF STSC’s shared access component file system for APL2. Comparable to all APL*PLUS file systems: multi-user storage of APL2 arrays with efficient disk usage Full featured FMT for APL2 Workspace documentation utilities Extends APL. database management

Uttlitias including: RAM disk, full screen data entry, menu Input, report generation, exception handling and games. 827x IRMA support. Financial & Statistical routines APL-based spreadsheet for APL*PLUS/PC. Celt arithmatic; transfers to ASCII & Lotus Front-end Foreign Exchange dealing / pos keeping Arbitrage modelling Basket currency modelling pul-dewn menu for APL*PLUS/PG PC Utitles including: APLMAC (windows); Untock (unlocks functions in AWS); DTEX (text and spreadsheet imp/exp). Mostly available In English or Franch. Drivers and documentation for use with APL*PLUS/PC system and other STSC software with Cyrillic alphabet (PC). Full screen APL2 editor with Immediata APL execution, anda full-screan debugger High parformance component and kayad file system (VS APL and APL2) A QuadFMT data formatter for VS APL and APL2Z, AP124 programming fer APL applications without GDDM (APL2)

56

VECTOR Vol.7 No.4

(Pc) Marcia

MicroAPL.

REUTER-FILE

Uniware {malntrame) PowerCoda CALYAP. ucr

STATGRAPHICS 4

Upgrade 3 to 4 Upgrade pra 3 to 4 MICROSPAN LOGOL

TWIGS

MicroTASK MicroFILé MieroPLOT MicroLink MicroEDIT MicroFORM MicroSPAN MicroGRID APLCALC MicroPLOT/PC MicroSPAN/PC STATGRAPHICS Rel 4 GLOBAL LIMITS tPSACONNECT MAILBOX MAILBOX/PC V.2 upgrade to V.2 NEWSFLASH VIEWPOINT

STSC's ENHANCEMENTS:

STSC's SHAREFILE 1300

1800 115, 684

375 250 poa

199 99

599

260

250

260

poa

250 250

poa

75 16 poa poa

poa for APL2 for calling non-APL programs (VS APL and APL2) Inter-user data transfer tor VM users via UVC Single user component and Keyed files for APL2/PC Integrated statistics/graphics system for the PC, Now with macros, Bulk and educational discounts available.

Comprehensive APL tutor Logistics management system for PC and 986. Salas Forecasting, Inventory Control, Master Scheduling, Distribution Requirements, Planning etc A modular library of tools to teach and explore state-of-the-art materials management concepts. Time serias forecasting Warehouse replenishment inventory Management Grouping requirements into EOQ's Schadullng production/purchasing

AILS modules above AIL5 modules site licence Product development alds Fite utlities and database Graphics for HP plotters etc General device communications Full screen APL editor Full screen forms design Comprehensive APL tutor Ethemet & other networking APL spreadsheet system For APL'PLUS/PC product APL saif Instruction for APL"PLUS/PC

Exposure management for banks Maintrame to micro link Electronic Mall Full screen front end to ,PSA mallbox

Real time message exchange 4G - Info centre product

Quad-tunctions & nested arrays for 18M VSAPL under VM/CMS and MVS/TSO component files for IBM VSAPL under VM/CMS and MVS/TSO- amnd for IBM APL2

57

VECTOR Vol.7 No.4

PROGRAMMER TOOLS & POA

UTILITIES

FILEPRINT poa

FILESORT poa

FILECONVERT poa

FILEMANAGER(EMMA) poa -STSO s database package

EXECUCALC poa -Mainframe spreadsheet compatible with VISICALC and part of LOTUS 1-2-3 under VSAPL{VM or TSO)

(microcomputer) STATGRAPHICS poa Statistics and graphics for PCa

STATGRAPHIGS UNISTAT poa An add-on module to STATGRAPHICS: Data analysis software, APL*PLUS TOOLS

-VOLI poa _Incl. 327 IRMA disk, full screen data entry, menu input. report generation, games

VOL2 poa _Incl. File documantor, screen editor, exception handling.

SPREADSHEET MNGR poa APL spreadsheet with built-in ASCH, LOTUS and SYMPHONY interfaces,

APL*PLUS/PC FIN & poa Collection of financial

STAT.LUBRARY and statistical utllltes.

POCKET APL poa Smaller version of APL‘PLUS/PC,

UNIASM poa Collection of assembler routines for APL*PLUSJPC users,

UNITAB poa _APL"PLUSYPC spreadsheet-ke data entry and validation system.

Tha APL DEBUGGER poa First released APL*PLUS/PC debugger.

APL2C poa _Interface between APL*PLUS/PC and DATALIGHT C language Warwick University BATS 250 Menu driven system for time serias analysis and forecasting uslng Bayasian Dynamic modalling. Price Is reduced to £36 tor academic Institutions. FAB free Training program for the above.

APL CONSULTANCY

(prices quoted are per day unless otherwise marked)

COMPANY PRODUCT PRICES(£)

Active Workspace Lid Consultancy poa

Adfea Consultancy poa

APL Peeple Consultancy poa

Buckland Management Systems Consultancy poa

Camacho Consultancy poa

Chapman Consultancy 160-300 DETAILS

PC Based APL system design, programming and Implementation.

Development, maintenance, conversion, migration, documentation,.of APL products In all APL environments

Consultants available at alt levels, with experiance in: VS APL. APL*PLUS, APL, Dyalog APL, APLS8B00, C/UnIx, TSO/MVS, VM/CMS, graphics, Operational Research ete.

Expertise In APL system design, project management, prototyping, financial applications, decision support systems, MIS, links to non-APL systems, documentation, etc.

Business and Technical systems in commerce and industry ~ designing, programming and implementing applications,

Speclalising In programming & manual writing.

24-hour programmer: APL, C, assembler, graphics; PC, mini, mainftame, network.

58

VECTOR Vol.7 No.4

Cocking/Drury Consultancy 175-275 275-360 300-450 400-600 450-750 Peter Cyriax Consultancy 100-150 420-200 160-300 Delphi Consultancy poa Dyadic Consultancy poa

E&s Consultancy poa

General Software Consultancy from 120 HIM. Consultancy poa

Jan A. Clark Consultancy poa

iNFOSTROY Consultancy poa

Inteligent Programs Ltd Consultancy 175-350 Documentation 150-250 Training 150-250 Mercia Consultancy poa MicroAPL. Consultancy poa MTC. Consultancy 240-500 Parallax Systems Inc Consultancy $750 QB On-Une Consultancy 250 REUTER-FILE Consultancy poa Rochester Group Consultancy poa Rex Swaln Consultancy poa

Wickliffe Computer Lid Consultancy poa

OTHER PRODUCTS COMPANY PRODUCT PRICES(£) Adfee Employment poa APL People Employment Agency poa

HMW Employment poa FAPL Lid An APL Tutorlal 2.80 Juntor consultant Consultant Senior consultant Principal Consultant Managing consultant Junior Consultant Consultant Consultant APL system development on mainframes and micros. APL and Unix system design, consultancy, programming and training. System prototyping: alf types of Information system, engineering software, graphics and decision support systams APL*PLUS/PC, APL2, Dyslog APL

System design consultancy, programming. HMW specialize In banking and prototyping work. Computer-based Information Systems implementation where acceptance Is critical, APL on PC and Macintosh. Human Factors of HCI; novice ease of use; online assistance; training coursas; distance-leaming materials. Localization of APL software for the Soviet Union software market Systems development, enhancements, support. Preparation of naw manuals, rewriting materials. Training for APL experts through to non-technical system users. APL*PLUS & VSAPL consultancy. Technical & applications consultancy. Business analysis and APL consultancy

Introductory APL, APL tor End-user & Advanced Topics In APL Specialising in Banking, Finaneiaf & Planning Systems.

Consultancy & support sarvice world-wide. Specialise in MIS using Sharp APL independent consultant, 15 years experience. Custom software development & training, PC andjor mainframe.

System design, consultancy, programming and documentation. Especially project management and decision support systems

DETAILS Contractors and Permanent employees placed at all lavels. Contractors supplied for shorylong-term contracts, supervised or unsupervised, Executive Search service available. Contractors and permanent employees placed. 45pp by Alvord & Thomson

59

VECTOR Vol.7 No.4

An Encyclopaedia of APL (2nd Edn)§.50 228pp by Helzer

APL in Social Studies 250 S6pp by Traberman

APL Instruction Manual (2nd Edn)2.60 SSpp by Camacho & Ziemann

APL Programs for the Mathematics Classroom (Springer-Verlag)

14,50 185pp by Thomson ** Please add ons pound packing charge per order

REUTER:FILE Productivity Tools poa _—_Utiltles for systems, operations, administration & analysts; auxiliary processors, comms software, International network. Databases poa Financial, aviation, energy and socloeconomic.

Isl Tangible Math $19 An APL Approach to Math - Shareware, includes base Sharp APL

J $24 Dictionary APL simplified and enhanced - Shareware (Mac,PC}

SharpAPLPC $74 Registered Shareware and Referenca Manual

ISt APL, $99 Improved APLPC - enhancements, performance, large workspaces Renaissance Data Systems Booksellers The widest range of APL books available anywhere, See Vector advertisements,

OVERSEAS ASSOCIATIONS

GROUP LOCATION JOURNAL OTHER SERVICES Ann.Sub. _Visa/Mcd

APL Bay Area USAN. Califomla APLBUG Monthly Meetings (2nd Monday) $15 NIN

Dutch APLU.G, Holland - Mini-congress, APL ShereWare Initiative

APL Club Austra Austria - Quarterly Meetings 200ASiperson, 1000AS per company NIN

VENDOR ADDRESSES

COMPANY CONTACT ADDRESS & TELEPHONE No.

Active Workspace Ltd Ross D Ranson Moulsham Mill Centre, Parkway, Chelmsford, Essex, CM2 7PX, Tal: (0245) 262414 ext.240

Adtes Bemard Smoor Corpsstraat 50, 4128 BZ LEXMOND, Netherlands. Tel: 91.3474,2937, fax: 31.3474.2342

APL 385 Adrian Smith Brook House, Gilling East, York, Tel: 04399-385

APLBUG Jorge Maze} 117 East Creek Dr., Menlo Park, CA 94025, USA

APL Glub Austria Erich Gall IBM Osterreich, Obere Donaustrasse 95, A-1020 Wien, Austria

APL tmpotus Ltd Cedric Heddte Rusper, Sandy Lane, Ivy Hatch, SEVENOAKS, Kent TN15 OPD Tel: 0732-885126

APL People Jill Moss The Cid Malthouse, Clarence St, BATH, BA1 5NS, Tel: 0225-462602

APL, Software David Alis The Old Malthouse, Clarsnes ST, 5NS, Tel: 0225-482602

Jil Moss

Buckland Management Systems

Anthony Camacho

Paul Chapman

Cocking & Drury Ltd. John Buckland

Romilly Cocking Westwood, 19 Grange Road, Camberley, Surrey, GU15 2DK Tek: 0276 684327

2 Blenheim Road, St. Albans, Herts AL1 4NA. Tel: St. Albans (0727) 860130

18, Trevelyan Road, London, SW17 9LN Tel: 081-767 4264

180 Tottenham Court Road, LONDON, W1P 9LE Tel: 071436 9481 Fax; 071-436 0524

60