summaryrefslogtreecommitdiff
path: root/macros/latex/contrib/linearregression/linearregression.dtx
blob: 2eeea126f85b68a957981389ef0c746af752d368 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
%\iffalse 
% file:    linearregression.dtx
% author:  Battista Benciolini
% contact: benciolinibattista at gmail dot com
% date:    see preamble 
% 
% process this file with pdflatex to obtain:
%
%  - linearregression.pfd     (full documentation, three pass needed)
%  - mainlinearregression.tex (interactive main-program document)
%  - linearregression.sty     (package)
%  - sampledata.txt           (as the name says)
% 
% The author would strongly appreciate to receive  
% any comment, criticism  and just usage report
%
%\fi
%\iffalse 
%<*ins>
\begingroup
\input docstrip.tex
\keepsilent
\preamble
------------------------------------------------------------------------
[2024-06-10]
This file is part of the (expanded) distribution of linearregression
The author of  linearregression  is  Battista Benciolini 
<benciolinibattista at gmail dot com >
------------------------------------------------------------------------
The author would strongly appreciate to receive  
any comment, criticism  and just usage report   
------------------------------------------------------------------------
This program may be used, distributed and modified under 
the conditions of the LaTeX Project Public License.
(see: http://www.latex-project.org/lppl.txt)
------------------------------------------------------------------------
\endpreamble
\askforoverwritefalse
\generate{\file{linearregression.sty}{\from{linearregression.dtx}{package}}}
\generate{\file{mainlinearregression.tex}{\from{linearregression.dtx}{main}}}
\nopreamble\nopostamble
\generate{\file{sampledata.txt}{\from{linearregression.dtx}{data}}}
\endgroup
%</ins>
%\fi
%\iffalse 
%<*driver>
\documentclass[a4paper,10pt]{ltxdoc}
\usepackage[T1]{fontenc}
\usepackage{lmodern}
\usepackage[lite,nobysame,non-compressed-cites]{amsrefs}
\usepackage{amsmath,amssymb,amsfonts}
\usepackage{multicol}  
\usepackage{linearregression} 
\usepackage{graphics} 
\DeclareRobustCommand*{\Ars}{\textsf{%
\lower -.48ex\hbox{\rotatebox{-20}{A}}\kern -.3em{rs}}%
\discretionary{-}{}{\kern -.05em}\TeX\discretionary{-}{}{%
\kern -.17em}\lower -.357ex\hbox{nica}}% excerpt from some GUIT sty file 
\NewDocumentCommand\vect{m}{\underline{#1}}           % vector
\NewDocumentCommand\barycenter{m}{\overline{#1}}      % barycenter
\NewDocumentCommand\point{}{\vect{y}}                 % point
\NewDocumentCommand\coeff{}{\vect{x}}                 % direction
\NewDocumentCommand\dx{}{\vect\delta}                 % direction variation
\NewDocumentCommand\vv{}{\vect{v}}                    % barycentric coordinates
\NewDocumentCommand\trasp{}{^{\mathsf{T}}}            % traspose
\NewDocumentCommand\Renne{}{\mathbb{R}^n}             % vector space 
\NewDocumentCommand\dor{}{f}                          % distance from origin 
\NewDocumentCommand\ipoint{}{i}                       % index for points
\NewDocumentCommand\pointsum{}{\sum_{\ipoint=1}^m}    % sum over points
\NewDocumentCommand\reff{m}{(\ref{#1})}               % ref  in ( )
\NewDocumentCommand\matr{m}{{#1}}                     % matrix
\NewDocumentCommand\mC{}{\matr{C}}                    % matrix C 
\NewDocumentCommand\mc{}{k}                           % elements of matrix C 
\NewDocumentCommand\mL{}{\matr{\Lambda}}              % matrix lambda
\NewDocumentCommand\mX{}{\matr{X}}                    % matrix X 
\NewDocumentCommand\spm{}{\phantom{-}}                % space for the sign
\NewDocumentCommand\ctext{}{caption}                  % caption (a variable !)
\NewDocumentCommand\matrixtwotwo{mmmm}{               % | 2 x 2 
\begin{pmatrix} #1 & #2 \\ #3 & #4 \end{pmatrix}}     % | matrix
\DeclareMathOperator\tr{tr}                           % trace
\DeclareMathOperator\sgn{sgn}                         % signum 
\title{Linear regression with \LaTeX}
\author{Battista Benciolini}
\NewDocumentCommand\titleauthorfootnote{}{\begingroup% Not an elegant solution
\let\thefootnote\relax                             % but it is ok at the moment
\footnote{Linear regression with LaTeX - available in CTAN}% 
\footnote{Battista Benciolini - contact: benciolinibattista at gmail dot com}%
\endgroup\setcounter{footnote}{0}}% 
\parindent=0pt
\begin{document}
\hypersetup{hidelinks}
\maketitle 
\titleauthorfootnote
\tableofcontents
\vfill
\DocInput{linearregression.dtx}
\end{document}
%</driver>
%\fi
%
% \section{Introduction: first description of the problem\label{intro}}
% I start with a quote from \Ars\ (April 2021, number 31, page 73):
% \begin{quotation}
% The physicist Mario Rossi is investigating a phenomenon, 
% presumably linear, and he performs measurements in his laboratory 
% to verify his hypothesis; he measures the quantity $x$ which generates 
% the phenomenon and he measures also one of the characteristics 
% $y$ showed by the phenomenon under the effect of the stimulation $x$. 
% \\ ... \par
% Subsequently Mario graphs the data of the table to judge if the points 
% reasonably follow a linear trend or not; in this regard he computes the 
% parameters of the regression line and he draws this line on the graph 
% in order to judge the quality of the obtained results.
% \\ ... \par
% Being a \LaTeX\ user, he thinks to kill two birds with one stone: 
% using \LaTeX\ to draw the graph with the experimental data consisting 
% in the $x$, $y$ points and, at the same time, to compute the
% parameter $a$ e $b$ of the regression line $y = ax+b$, 
% and finally to draw also this line on the same graph.
% \end{quotation}
% A summary description of the the problem is therefore the following.
% A set of data pairs  is available and each pair is represented as a point 
% in the plain. A straight line is searched that optimally approximates 
% the points. The first step is therefore the choice of an optimality criterion. 
% This choice is the topic of the next section. \par
% From the text we also know that the possible deviation of $y$ 
% with respect to the model is quite larger than the uncertainty of $x$.
% \par
% After reading the description of the problem 
% of Mario Rossi I tried to produce a solution. 
% In this work I will use $y_1$ and $y_2$ 
% instead of $x$ and $y$ for the two measured quantities that 
% will become the first and second coordinate, or abscissa and ordinate, 
% in the Cartesian plane.
% \par
% The problem can be treated as a mere problem of approximation or 
% alternatively as an estimation problem in the frame of a 
% probabilistic description of the uncertainty. The two treatments are 
% conceptually different. The probabilistic treatment produces some more 
% results, but the estimation of the parameters is the same. 
% On the other hand the treatment as an approximation problem is in some sense   
% more immediate and requires a less extended theoretical background.
% For this reason it will be preferred here.  
% I consider the original problem and also a variation 
% of it based on the assumption that the two variables are known with  
% the same uncertainty. The two considered situations will prove 
% to be quite different.
%
% \section{Geometric definition of there optimality criteria}
% \begin{figure}
% \setlength\unitlength{4cm}
% \begin{picture}(1, 0.7)(0.,0.)
% \multiput(0.,0.)(1.1,0){3}{\line(1,0){1}}
% \multiput(0.,0.)(1.1,0){3}{\line(0,1){1}}
% \multiput(1,1)(1.1,0){3}{\line(-1,0){1}}
% \multiput(1,1)(1.1,0){3}{\line(0,-1){1}}
% \thicklines
% \multiput(0.08,0.06)(1.1,0){3}{\line(4,3){0.88}}
% \multiput(0.26,0.09)(1.1,0){3}{\circle{0.03}}
% \multiput(0.45,0.65)(1.1,0){3}{\circle{0.03}}
% \multiput(0.92,0.44)(1.1,0){3}{\circle{0.03}}
% \put(0.26,0.09){\line(0,1){0.1050}}
% \put(0.45,0.65){\line(0,-1){0.3125}}
% \put(0.92,0.44){\line(0,1){0.2500}}
% \put(1.36,0.09){\line(-1,0){0.14}}
% \put(1.55,0.65){\line( 1,0){0.4167}}
% \put(2.02,0.44){\line(-1,0){0.3333}}
% \put(2.46,0.09){\line(-3,4){0.05}}
% \put(2.65,0.65){\line(3,-4){0.15}}
% \put(3.12,0.44){\line(-3,4){0.12}}
% \end{picture}
% \caption{The three kinds of segments 
% used in the definition of the objective function}
% \label{fig:criteria}
% \end{figure}
% For each point given in the plane  we can consider the corresponding point 
% with the same abscissa and belonging to the line. 
% Remember that the line is exactly what has to be determined.
% The distance between the given point and the just defined point on the line
% is a reasonable measure of the discrepancy between the empirical data and the
% corresponding theoretical model. 
% The distances we are speaking about are the length of the segments shown 
% in the leftmost scheme of figure (\ref{fig:criteria}).
% To obtain a global discrepancy measure that considers all the points 
% at once we perform the sum of the squares of the lengths 
% of the mentioned segments. It is now clear that the two coordinates 
% of the points are treated quite differently and play a different role in 
% the definition of the optimality criterion. This choice is reasonable when 
% the measuring errors only (or mainly) affect the second coordinate. 
% The optimal line is the line that minimize the just defined 
% global discrepancy. The procedure for the determination of the optimal line 
% is named linear regression. 
% In this work it is named \textit{classical linear regression}.
% We can easily exchange the role of the two quantities, i.e.\ we can   
% imagine that the first quantity is affected by errors. 
% The problem is not conceptually different. The segments plotted in the
% central picture 
% of figure (\ref{fig:criteria}) represent the discrepancy between 
% the empirical data and the model.
% This other procedure  is named \textit{classical linear regression  
% with inverted role of the coordinates}.\par
% The situation is really different if the two coordinates have to be treated 
% symmetrically. 
% In this case the discrepancy between 
% the empirical data and the model must be defined in a purely geometrical way.
% Just the line and the points enter in the definition without any special role 
% for any predefined direction. With these requirements it is quite natural 
% to use the distance of each point from the line. 
% Remember that the distance of a point 
% from a line is intended along the shortest path, i.e.\ measured in the 
% direction orthogonal to the line itself. The rightmost scheme of 
% figure (\ref{fig:criteria}) shows the segments that are considered.
% The global measure of discrepancy is again obtained as the sum 
% of the squares of the length of the mentioned orthogonal segments. 
% The procedure that obtain the optimal line that 
% minimize the just defined global discrepancy is named 
% \textit{symmetrical linear regression}.\par
% Some arguments of the present section will be repeated in section
% \ref{package} from the algebraic and computational point of view.
% 
% \section{General information on the proposed solution, including limitations}
% The code that implements the solution is recorded in two files, that are 
% a package (sty) file and a main interactive document.
% The file |linearregression.sty| provides several commands 
% that can be used in any document. The file |mainlinearregression.tex|
% provides a simple interactive user interface.
% The package described in the sections \ref{manual} and 
% \ref{package} (user manual and implementation) provides the 
% functions that execute the various needed operations, i.e.\
% data input, computations, printing the numerical results and 
% generating a graphic representation of data and results.
% Some auxiliary functions complete the package.
% The design of the output (tables and plots) includes some arbitrary choices.
% The style of the graphic output is quite minimalist 
% (e.g.:\ no colors, no variations of line styles).\par
%
% \section{Some comments about the programming aspect of the package 
% and its documentation}
% Large part of the code is written using the |expl3| language. 
% (Is it also named simply L3 ? Does expl still means experimental ?)
% I have tried to be compliant with the various recommendations and 
% prescriptions for a correct use of the language, 
% but I probably only partly succeeded.\par
% Different more elegant and more coherent solutions probably exist 
% both for the general structure of the package and for some specific part 
% of the code, but this is what I have been able to do.
% Some perhaps problematic aspects are mentioned here after\par
% Several used variables are global and they are accessed by various functions.
% This makes the various parts of the package 
% quite connected to each other and creates strong dependencies. \par
% The layered programming style is only partially applied. 
% The partition between document command and lower level functions is present, 
% but part of the low level code is directly in the document commands. 
% Variants are not used.\par 
% One more remarks  concern the documentation.
% I was uncertain about the opportunity of using the class |l3doc|. I decided to
% remain using |ltxdoc|. This is the reason why I do not use the environment  
% |macro| and the command |\cs| in the documentation of some auxiliary 
% functions named according with the |expl3| standard. 
% (I have just an interim far from optimal solution 
% for a reasonable formatting.)  
%
% \section{A ready to use simple user interface\label{main}}
% The main file asks the user for the name of a
% file containing the data and generates a one (or two) page output.
%\iffalse
%<*main>
%\fi
%    \begin{macrocode}
\documentclass[a4paper]{article}
\usepackage{lmodern}
\usepackage{linearregression}
\begin{document}
\pagestyle{empty}  
\lraskfilename 
\lrcomputation
\lrplot{12.0}{+}{+}{-}{-}
\lrprint
\end{document}
%    \end{macrocode}
%\iffalse
%</main>
%\fi
%
% \section{A user manual for the package\label{manual}}
% The various analysis of a data set and the representation of the data 
% and of the results is obtained with a sequence of several commands. 
% The main operations are:
% (i) selection of the data file, (ii) data imput and computation, 
% (iii) printing of a table,
% (iv) printing of a picture (that can be repeated with different parameters).
% It is generally convenient to put the table and the picture(s) 
% in a proper floating environment. 
% The commands for the four mentioned operations are described here after.
% The first needed operation is to set the name of the data file.
% This is done with the command \DescribeMacro{\lrfilename} 
% \cs{lrfilename}\marg{file} that has a mandatory argument. 
% The argument is the name of the data file. As an alternative the 
% command \DescribeMacro{\lraskfilename} \cs{lraskfilename} can be used. 
% It asks the user to type the  name of the data file in the terminal.
% \par
% The macro \DescribeMacro{\lrcomputation} 
% \cs{lrcomputation} reads the data 
% and performs all the computations. 
% The results of the computations remain available in internal 
% variables and are then used by the macro that print them 
% or generates a plot.
%\par
% The macro \DescribeMacro{\lrprint} 
% \cs{lrprint} generates a table with all the estimated 
% parameters and some information about the data. 
% \par 
% The macro \DescribeMacro{\lrplot} 
% \cs{lrplot}\marg{imagewidth}\marg{key1}\marg{key2}\marg{key3}\marg{key4} 
% really generates the plot. The first argument is the
% width of the plot, while the height is computed according 
% to the distribution of the points. The other four arguments are referred 
% to the data points, to the lines determined with classical regression, 
% with classical regression with inverted role of the coordinates and 
% with symmetric regression. 
% The four items, i.e.\ the set of points and the three lines, are drawn 
% or not according to the corresponding character found in |key|$i$. 
% Each item is not plotted if the character is a |-|, it is plotted in any other
% case. Furthermore the lines are accompanied by a label made by the
% corresponding |key|, unless it is just a |+|.  
% \par
% Few words are necessary about the format of the data file.
% Each record of the file hold the two values related to a point.
% The two values must be separated by any number (one is needed as a minimum) of 
% space and comma characters. No character different from space 
% can be accepted before the first value and after the  second value. 
%
% \section{An example\label{example}}
% The data  reported here after will be available in |sampledata.txt| 
% and will be used in the example presented in this section .
%\iffalse
%<*data>
%\fi
% \begin{multicols}{4}
%    \begin{macrocode}
-0.546  0.107
 1.093 -0.510
 1.440  1.995
 1.414  0.991
 0.735  1.585
-1.848 -0.235
-0.203 -0.292
 1.517  0.779
 0.559 -1.341
-0.462 -0.437
-0.785 -0.661
-0.558  0.397
 0.181 -2.616
 0.619  1.859
-0.223 -1.915
 0.629 -0.534
-1.989 -2.300
-0.241  1.098
-0.931 -1.613
-1.070  0.592
 2.341  0.413
 1.993 -0.111
-2.357 -0.312
-1.975  0.140
%    \end{macrocode}
% \end{multicols}
%\iffalse
%</data>
%\fi
%
% The analysis of the sample data and the generation of a numeric table 
% is  operated by a code similar to the following 
% (see table \ref{tab:sampledata}). \\
% |\lrfilename{sampledata.txt}| \\ |\lrcomputation| \\ 
% |\begin{table}| \\  
% |  \lrprint| \\
% |  \caption{Analysis of ... }| \\  |\label{tab:sampledata}\end{table}| 
% \par
% The generation of some different graphical representation of the data and of 
% the results  is  operated by a code similar to the following 
% (see figures \ref{fig:sampledataB} ).\\
% \RenewDocumentCommand\ctext{}{LEFT The three lines are obtained with the three 
% optimality criteria. (AA) classical linear regression; (BB) classical linear 
% regression with inverted role of the coordinates; (S) symmetric linear 
% regression. RIGHT Data points and line estimated with 
% symmetric linear regression.}
% |\begin{figure}|\\|\lrplot{10.}{-}{AA}{BB}{S}| \\
% |\lrplot{10.}{+}{-}{-}{+}|
% \\ |\caption{|\ctext|}|\\ | \label{fig:sampledataB}  \end{figure}|
%
% \lrfilename{sampledata.txt} \lrcomputation 
% \begin{table}  \lrprint     \caption{Analysis of the sample data} 
%  \label{tab:sampledata} \end{table}
%  \begin{figure}    \lrplot{6.}{-}{AA}{BB}{S} \hfill \lrplot{6.}{+}{-}{-}{+} 
%    \caption{\ctext} \label{fig:sampledataB} \end{figure}
%
% \section{A package for linear regression 
% and the theory behind it\label{package}}
%\iffalse
%<*package>
%\fi
%
% \subsection{Math preliminaries and notation \label{prelim}}
% The coordinates of a set of $m$ points on the plane are available. 
% A straight line is searched that optimally approximates the points.\par 
% The coordinates of a generic point are $y_1$ and $y_2$ 
% and they are collected in the vector $\point$.
% Any given point is identified with the index $\ipoint$.
% (Explicit indices $(\dots)_1$ or $(\dots)_2$ always refer to the first 
% or second coordinate of a point or to the first or second component 
% of a vector in the plane. 
% Symbolic index $(\dots)\ipoint$ always refers to the different points. Few 
% formulas require both indices $(\dots)_{1\ipoint}$, $(\dots)_{2\ipoint}$.)\par
% With more then two points a criterion of best approximation 
% is needed to select the optimal line that describes the data. \par
% Lower case symbols are used for scalars. Lower case underlined
% symbols are used for vectors in the plane. Upper case symbols
% are used for matrices.
% \par
% It is possible that certain data generate an ambiguity or a singularity
% in the computation. 
% The following mathematical treatment of the problem
% do not mention these situations and the code does not deal with them.
%
% \subsection{Package declaration, required package and definition of variables}
% The various macro will be provided in a package file
% that is introduced as usual. Most of the macros require 
% the \LaTeX3 syntax.
%    \begin{macrocode}
\ProvidesPackage{linearregression}[2024-06-10]
\RequirePackage{pict2e}        	
\ExplSyntaxOn
%    \end{macrocode}
% The variables used in the package are defined hereafter.
%    \begin{macrocode}
\ior_new:N  \g_BBLR_file_ior 
\tl_new:N   \g_BBLR_file_name_tl
\int_new:N  \g_BBLR_number_of_points_int
\fp_new:N   \g_BBLR_abscissa_fp 
\fp_new:N   \g_BBLR_ordinate_fp     
\fp_new:N   \g_BBLR_mean_abscissa_fp 
\fp_new:N   \g_BBLR_mean_ordinate_fp 
\fp_new:N   \g_BBLR_abscissa_SecOrdMoment_fp 
\fp_new:N   \g_BBLR_ordinate_SecOrdMoment_fp 
\fp_new:N   \g_BBLR_mixed_SecOrdMoment_fp 
\fp_new:N   \g_BBLR_slope_A_fp 
\fp_new:N   \g_BBLR_slope_B_fp   
\fp_new:N   \g_BBLR_slope_S_fp 
\fp_new:N   \g_BBLR_intercept_A_fp 
\fp_new:N   \g_BBLR_intercept_B_fp 
\fp_new:N   \g_BBLR_intercept_S_fp 
\fp_new:N   \g_BBLR_cos_fp
\fp_new:N   \g_BBLR_sin_fp 
\fp_new:N   \g_BBLR_sig_sin_fp
\fp_new:N   \g_BBLR_eig_diff_fp 
\fp_new:N   \g_BBLR_diag_diff_fp 
\tl_new:N   \g_BBLR_file_line_tl  
\fp_new:N   \g_BBLR_min_abscissa_fp   
\fp_new:N   \g_BBLR_min_ordinate_fp 
\fp_new:N   \g_BBLR_max_abscissa_fp
\fp_new:N   \g_BBLR_max_ordinate_fp 
\fp_new:N   \g_BBLR_min_draw_abscissa_fp 
\fp_new:N   \g_BBLR_max_draw_abscissa_fp  
\bool_new:N \g_BBLR_data_eof_bool
\int_new:N  \g_BBLR_record_length_int
\int_new:N  \g_BBLR_rec_count_int
\int_new:N  \g_BBLR_first_separator_int
\int_new:N  \g_BBLR_last_separator_int      
\str_const:Nn  \c_BBLR_space_str {~}      
\str_const:Nn  \c_BBLR_comma_str {,}    
\str_const:Nn  \c_BBLR_plus_str  {+}   
\str_const:Nn  \c_BBLR_minus_str {-}         
\bool_new:N  \g_BBLR_plot_points_bool        
\bool_new:N  \g_BBLR_plot_lineA_bool      
\bool_new:N  \g_BBLR_plot_lineB_bool       
\bool_new:N  \g_BBLR_plot_lineS_bool            
\fp_new:N   \g_BBLR_base_fp                
\fp_new:N   \g_BBLR_height_fp             
\fp_new:N   \g_BBLR_Xbase_fp           
\fp_new:N   \g_BBLR_Xheight_fp            
\fp_new:N   \g_BBLR_Dabscissa_fp         
\fp_new:N   \g_BBLR_Dordinate_fp         
\fp_new:N   \g_BBLR_diameter_fp            
\fp_gset:Nn \g_BBLR_diameter_fp{0.2}    
\fp_new:N   \g_BBLR_line_base_length_fp    
\fp_new:N   \g_BBLR_scale_factor_fp      
\str_new:N  \c_BBLR_point_code_str
\str_new:N  \g_BBLR_labelA_str
\str_new:N  \g_BBLR_labelB_str         
\str_new:N  \g_BBLR_labelS_str
%    \end{macrocode}
%
% \subsection{Preparing data input}
% \begin{macro}{\lrfilename}
% The command \cs{lrfilename} records the file name passed as argument.
%    \begin{macrocode}
\NewDocumentCommand{\lrfilename}{m}{
\tl_gset:Nn \g_BBLR_file_name_tl {#1}
}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\lraskfilename}
% The command \cs{lraskfilename} asks for the data file name from the terminal.
%    \begin{macrocode}
\NewDocumentCommand{\lraskfilename}{}{
\ior_get_term:nN  {filename ? } \g_BBLR_file_name_tl 
\tl_trim_spaces:N \g_BBLR_file_name_tl 
}
%    \end{macrocode}
% \end{macro}
%
% \subsection{Main command declaration, computation of 
% first and second order moments}
% \begin{macro}{\lrcomputation}
% The  command \cs{lrcomputation} reads the data file and 
% performs all the relevant computations to solve the 
% proposed problem. 
%    \begin{macrocode}
\NewDocumentCommand{\lrcomputation}{}{%
%    \end{macrocode}
%
% In the sequel it will results that the first and second order moments 
% of the data provide everything needed to solve the problem. 
% The barycenter of the data is defined as 
% \begin{equation}
% \barycenter{\point}=\frac{1}{m}\pointsum \point_\ipoint.
% \label{barycenter} \end{equation}
% It is convenient to scan the data to accumulate the sum 
% that appears in \reff{barycenter}.
% The coordinates of each point are read from the file 
% and they are immediately used. 
% It is therefore not necessary to globally record the data.
%    \begin{macrocode}
\bool_gset_false:N \g_BBLR_data_eof_bool             
\int_zero:N \g_BBLR_number_of_points_int                
\fp_zero:N  \g_BBLR_mean_abscissa_fp                     
\fp_zero:N  \g_BBLR_mean_ordinate_fp                   
\ior_open:Nn \g_BBLR_file_ior \g_BBLR_file_name_tl   
\bool_until_do:Nn \g_BBLR_data_eof_bool {            
  \ior_str_get:NN \g_BBLR_file_ior \g_BBLR_file_line_tl
  \if_eof:w \g_BBLR_file_ior                        
    \bool_gset_true:N \g_BBLR_data_eof_bool          
  \else:                                              
    \int_incr:N \g_BBLR_number_of_points_int       
    \BBLR_decode_data:                               
    \fp_gset:Nn \g_BBLR_mean_abscissa_fp               
    {\g_BBLR_mean_abscissa_fp  + \g_BBLR_abscissa_fp}   
    \fp_gset:Nn \g_BBLR_mean_ordinate_fp            
    {\g_BBLR_mean_ordinate_fp + \g_BBLR_ordinate_fp} 
  \fi:                                                 
}                                                    
%    \end{macrocode}
% Loop ended. Now close the file and divide by the number of points.
%    \begin{macrocode}
\ior_close:N \g_BBLR_file_ior                        
\fp_gset:Nn \g_BBLR_mean_abscissa_fp                  
{\g_BBLR_mean_abscissa_fp / \g_BBLR_number_of_points_int}
\fp_gset:Nn \g_BBLR_mean_ordinate_fp                  
{\g_BBLR_mean_ordinate_fp / \g_BBLR_number_of_points_int}
%    \end{macrocode}
%
% The barycentric coordinates are defined for each point
% \begin{equation} \vv_\ipoint= \point_\ipoint - \barycenter{\point}
% \label{residual} \end{equation}
%  and the empirical dispersion matrix is defined as: 
% \begin{equation}  \mC=\frac{1}{m}\pointsum \vv_\ipoint\vv_\ipoint\trasp .
% \label{matrixC} \end{equation}
% Superscript as in $()\trasp$ means transpose. The elements of  $\mC$ are the 
% second order central moments and they are denoted as: 
% \begin{equation}  \mC=\matrixtwotwo{\mc_{11}}{\mc_{12}}{\mc_{12}}{\mc_{22}}.
% \label{matrixCc} \end{equation}
% A second scan of the data is performed to compute the
% sums that appears in \reff{matrixC} and to determine the 
% the extremal values of the coordinates. Record scan can be regulated 
% by a record counter, because the the number of points is now known.
%    \begin{macrocode}
\fp_zero:N  \g_BBLR_abscissa_SecOrdMoment_fp                      
\fp_zero:N  \g_BBLR_ordinate_SecOrdMoment_fp                      
\fp_zero:N  \g_BBLR_mixed_SecOrdMoment_fp                         
\fp_gset_eq:NN  \g_BBLR_min_abscissa_fp  \g_BBLR_mean_abscissa_fp    
\fp_gset_eq:NN  \g_BBLR_min_ordinate_fp  \g_BBLR_mean_ordinate_fp        
\fp_gset_eq:NN  \g_BBLR_max_abscissa_fp  \g_BBLR_mean_abscissa_fp       
\fp_gset_eq:NN  \g_BBLR_max_ordinate_fp \g_BBLR_mean_ordinate_fp  
\ior_open:Nn   \g_BBLR_file_ior \g_BBLR_file_name_tl         
\int_zero:N    \g_BBLR_rec_count_int           
\int_do_until:nn                  
{\g_BBLR_rec_count_int = \g_BBLR_number_of_points_int} 
{                                                        
    \ior_str_get:NN \g_BBLR_file_ior \g_BBLR_file_line_tl    
    \int_incr:N \g_BBLR_rec_count_int               
    \BBLR_decode_data:                                    
    \fp_gset:Nn \g_tmpa_fp                          
    {\g_BBLR_abscissa_fp - \g_BBLR_mean_abscissa_fp}
    \fp_gset:Nn \g_tmpb_fp                          
    {\g_BBLR_ordinate_fp - \g_BBLR_mean_ordinate_fp}
    \fp_gset:Nn \g_BBLR_abscissa_SecOrdMoment_fp             
    {\g_BBLR_abscissa_SecOrdMoment_fp + \g_tmpa_fp * \g_tmpa_fp}
    \fp_gset:Nn \g_BBLR_mixed_SecOrdMoment_fp                
    {\g_BBLR_mixed_SecOrdMoment_fp + \g_tmpa_fp * \g_tmpb_fp}     
    \fp_gset:Nn  \g_BBLR_ordinate_SecOrdMoment_fp           
    {\g_BBLR_ordinate_SecOrdMoment_fp + \g_tmpb_fp * \g_tmpb_fp}              
\fp_gset:Nn  \g_BBLR_min_abscissa_fp                               
{min(\g_BBLR_min_abscissa_fp, \g_BBLR_abscissa_fp)}                 
\fp_gset:Nn  \g_BBLR_min_ordinate_fp                              
{min(\g_BBLR_min_ordinate_fp, \g_BBLR_ordinate_fp)}                 
\fp_gset:Nn  \g_BBLR_max_abscissa_fp                              
{max(\g_BBLR_max_abscissa_fp, \g_BBLR_abscissa_fp)}                 
\fp_gset:Nn  \g_BBLR_max_ordinate_fp                              
{max(\g_BBLR_max_ordinate_fp, \g_BBLR_ordinate_fp)}                 
}                                                             
\ior_close:N \g_BBLR_file_ior                                     
\fp_gset:Nn \g_BBLR_abscissa_SecOrdMoment_fp                       
{\g_BBLR_abscissa_SecOrdMoment_fp / \g_BBLR_number_of_points_int}   
\fp_gset:Nn \g_BBLR_mixed_SecOrdMoment_fp                           
{\g_BBLR_mixed_SecOrdMoment_fp / \g_BBLR_number_of_points_int}      
\fp_gset:Nn \g_BBLR_ordinate_SecOrdMoment_fp                        
{\g_BBLR_ordinate_SecOrdMoment_fp / \g_BBLR_number_of_points_int}   
\fp_gset:Nn \g_BBLR_Dabscissa_fp                               
{\g_BBLR_max_abscissa_fp - \g_BBLR_min_abscissa_fp }       
\fp_gset:Nn \g_BBLR_Dordinate_fp                            
{\g_BBLR_max_ordinate_fp - \g_BBLR_min_ordinate_fp }        
%    \end{macrocode}
% A single pass algorithm exists, but it is numerically less stable. 
%
% \subsection{Classical linear regression \label{classical}}
% A line in the plane is described by the equation  
% \begin{equation} y_2=ay_1+b \label{eqab} \end{equation}
% that contains the parameters $a$ and $b$. 
% For each point it is possible to define a distance or a discrepancy 
% of the experimental data with respect to the model.
% In the given problem the second coordinate is much more affected by 
% errors than the first coordinate. It is therefore reasonable 
% to define the approximation error of each point as
% \begin{equation} e_\ipoint=y_{2\ipoint}-ay_{1\ipoint}-b
% \label{e}\end{equation}
% i.e.\ the difference between the empirical value $y_{2\ipoint}$ 
% and its model counterpart $ay_{1\ipoint}+b$.
% The global discrepancy between the data and the model is measured by the 
% least square objective function defined by:
% \begin{equation} \psi=\pointsum e_\ipoint^2 \label{psiab} \end{equation}
% and the parameters $a$ and $b$ will be determined 
% just by the minimization of the function $\psi$ defined in \reff{psiab}.
% \par
% In the present treatment of the regression problem as a pure 
% approximation problem the definition  of  $\psi$ in \reff{psiab} 
% seams quite arbitrary. It is anyway a convenient choice. 
% \par
% Expression \reff{e}  can be rewritten in the different form
% \begin{equation} 
% e_\ipoint=v_{2\ipoint}-av_{1\ipoint}+\barycenter{y}_2-a\barycenter{y}_1-b
% \label{e2}\end{equation}
% so that the function to be minimized can be expressed 
% as the sum of two quadratic functions:
% \begin{equation} 
% \psi=
% \pointsum (v_{2\ipoint}-av_{1\ipoint})^2+
% m(\barycenter{y}_2-a\barycenter{y}_1-b)^2 
% \label{psiab2} \end{equation}
% and the minimum can be attained considering 
% the two terms one at a time. 
% The second term in the right-hand side of \reff{psiab2} 
% vanishes if the choice of $b$ is: 
% \begin{equation} b=\barycenter{y}_2-a\barycenter{y}_1.
% \label{estb} \end{equation}
% The first term in the right-hand side of \reff{psiab2} becomes:
% \begin{equation}  \psi_{(a)}=m\left(\mc_{22}-2a\mc_{12}+a^2\mc_{11}\right).
% \label{parabola} \end{equation}
% Searching the minimum of $\psi$ w.r.t.\ $a$ is therefore the search 
% of the abscissa of the vertex of a parabola  
% with axis parallel to the second coordinated axis.
% The result is:
% \begin{equation} a=\mc_{12}/\mc_{11}
% \label{esta} \end{equation}
% Now the slope $a$ and the intercept $b$ can be actually computed.
%    \begin{macrocode}
\fp_gset:Nn \g_BBLR_slope_A_fp
{\g_BBLR_mixed_SecOrdMoment_fp / \g_BBLR_abscissa_SecOrdMoment_fp }
\fp_gset:Nn \g_BBLR_intercept_A_fp
{\g_BBLR_mean_ordinate_fp - \g_BBLR_slope_A_fp * \g_BBLR_mean_abscissa_fp}
%    \end{macrocode}
% \par
% The empirical data and the estimated values of $a$ and $b$
% can be used to compute 
% the value actually attained by the residuals $e_\ipoint$ and
% by the function $\psi$. Then the index
% \begin{equation} \hat\sigma_0^2=\psi/(m-2)\end{equation}
% can be used to evaluate the general quality of the data and of the model.
% This claim is clearly quite generic. A complete understanding 
% of this evaluation would require to treat the linear regression 
% problem in the framework of the probabilistic estimation theory. 
% The used notation is derived from that theory.\par
% If the role of the two coordinates is exchanged the result 
% for $a$ becomes (still with reference to \reff{eqab})
% \begin{equation} a=\mc_{22}/\mc_{12}.\end{equation}
% A complete treatment of this different situation would include 
% the redefinition of $e_\ipoint$ and of $\psi$.
% The slope and the intercept can be computed according with 
% the different assumption.
%    \begin{macrocode}
\fp_gset:Nn \g_BBLR_slope_B_fp
{\g_BBLR_ordinate_SecOrdMoment_fp / \g_BBLR_mixed_SecOrdMoment_fp}
\fp_gset:Nn \g_BBLR_intercept_B_fp
{\g_BBLR_mean_ordinate_fp - \g_BBLR_slope_B_fp * \g_BBLR_mean_abscissa_fp}
%    \end{macrocode}
%
% \subsection{Symmetric linear regression \label{symmetric}}
% If both the coordinates of the experimental points are affected 
% by the same uncertainty it is advisable to use a more symmetric 
% optimality criterion and it is convenient to use a different model equation.
% \par
% The same line can be described by a different equation, i.e.\ 
% \begin{equation} x_1y_1+x_2y_2=\dor \end{equation}
% or in vector form:
% \begin{equation} \coeff\trasp\point=\dor. \label{eqvx} \end{equation}
% The parameters in \reff{eqvx} 
% are the scalar $\dor$ and the elements 
% of the vector $\coeff$, i.e.\ $x_1$ and $x_2$. 
% The line described by \reff{eqvx} is obviously
% invariant when the three parameters are simultaneously 
% scaled by a constant. The normalization condition 
% \begin{equation} \coeff\trasp\coeff=1,  \label{norm} \end{equation}
% supplemented by $\dor\ge 0$, 
% is quite convenient because the parameters will assume 
% a significant geometrical meaning:
% $\coeff$ is the unit vector orthogonal to the line and $\dor$ 
% is the distance of the line from the origin. 
% The expression 
% \begin{equation}  d=\dor-\coeff\trasp\point \label{distance} \end{equation}
% is the distance of the generic point $\point$ from the line 
% with a sign that is positive for points on the same side of the origin.
% \par
% The distance of each given point from the desired optimal line 
% is denoted by $d_\ipoint$.
% It has a clear intrinsic geometrical meaning and it does not 
% privileges one coordinate w.r.t.\ the other.
% The function to be minimized by the optimal line is 
% \begin{equation} 
% \phi=\frac{1}{m}\pointsum d_\ipoint^2. \label{phi1} \end{equation}
% The parameters of  \reff{eqvx} are determined by the minimization 
% of the function $\phi$ that can be expressed as:
% \begin{equation}  
% \phi=\frac{1}{m}\pointsum (\coeff\trasp\point_{\ipoint}-\dor)^{2}
% \label{phi2} \end{equation}
% and then, after some algebraic manipulations:
% \begin{equation} 
% \phi=\coeff\trasp\mC\coeff+(\dor-\coeff\trasp\barycenter{\point})^2
% \label{phi3}. \end{equation}
% The function $\phi$ is composed (as it was the function $\psi$) by the sum 
% of two parts. The second term in the right-hand side of \reff{phi3} 
% vanishes if the choice of $\dor$ is: 
% \begin{equation} \dor=\coeff\trasp\barycenter{\point}.
% \label{estd} \end{equation}
% Then it is necessary to minimize the function
% \begin{equation}   \phi_{(\coeff)} =  \coeff\trasp\mC\coeff
% \label{quadraticfun} \end{equation}
% with the constrain $\coeff\trasp\coeff=1$.
% It can be proved that the function $\phi_{(\coeff)}$ 
% is stationary if $\coeff$ is an eigenvector of \mC. \par
% The function $\phi_{(\coeff)}$ and the constrain must be combined 
% using a Lagrange multiplier:
% \begin{equation} 
% \Phi=  \coeff\trasp\mC\coeff+\lambda(1-\coeff\trasp\coeff).
% \label{Phi} \end{equation}
% Then the stationarity points of $\Phi$ must be determined. 
% Equating to zero the derivatives of $\Phi$ gives
% \begin{equation} 
% \mC\coeff=\lambda\coeff
% \label{auto} \end{equation}
% i.e.\ $\coeff$ is an eigenvector of $\mC$. \par
% The same result is obtained with the following argument.
% The function $\phi_{(\coeff)}$ is stationary if its first variation 
% is zero. The variation of $\coeff$ is named $\dx$ . 
% It must respect the constrain, that becomes $\dx\trasp\coeff=0$.
% The first variation of $\phi_{(\coeff)}$ is $2\dx\trasp\mC\coeff$, 
% and it is zero if and only if the following implication is valid:
% $\dx\trasp\coeff=0 \implies \dx\trasp\mC\coeff=0$,
% and the implication is valid if and only if the vector
% $\mC\coeff$  has the same direction of $\coeff$, i.e.\ if 
% $\coeff$ is an eigenvector of $\mC$.
% \par
% The result on the optimal line 
% can be described geometrically in the following way:
% (i) the optimal line includes the barycenter of the data;  
% (ii) the optimal line is orthogonal to the eigenvector of 
% $\mC$ corresponding to the minimum eigenvalue.\par
% The obtained result is also valid in $\Renne$.
% A set of points in $\Renne$ must be approximated by an $(n-1)$-dimensional 
% affine subspace. (Other more general situations can be considered.)
% \par
% The trace of the matrix $\mC$, denoted as $\tr(\mC)$, is a measure of the 
% global dispersion of the set of points. 
% The minimum eigenvalue $\lambda_{\textrm{min}}$ of $\mC$ is a measure 
% of the dispersion of the set of points with 
% respect to the optimal affine subspace. Therefore the index 
% \begin{equation}  \frac{n\lambda_{\textrm{min}}}{\tr(\mC)} 
% \end{equation}  
% can be used as an indicator of the relative residual 
% dispersion of the data around the optimal line.
% The defined index is dimensionless and it is 
% always between $0$ and $1$.
% \par
% For the actual computation of $\coeff$ it is convenient to consider 
% the spectral factorization of the matrix $\mC$, i.e.\ 
% $\mC=\mX\mL\mX\trasp$ where $\mL$ is a diagonal matrix 
% whose diagonal elements are the eigenvalues of $\mC$ 
% and $\mX$ is an orthonormal matrix whose columns are 
% the eigenvectors of $\mC$. The spectral factorization  exists 
% for any symmetric matrix, but it is specially simple for 
% a $2\times 2$ matrix.
% \begin{equation}
% \matrixtwotwo{\mc_{11}}{\mc_{12}}{\mc_{12}}{\mc_{22}}=
% \matrixtwotwo{c}{-s}{s}{\spm c} 
% \matrixtwotwo{\lambda_1}{0}{0}{\lambda_2} 
% \matrixtwotwo{\spm c}{s}{-s}{c}
% \label{spectral}\end{equation}
% The eigenvalues can be easily obtained because 
% their sum is the trace of $\mC$ 
% \begin{equation}
% \lambda_1 + \lambda_2 = \mc_{11}+\mc_{22}
% \label{Sum}\end{equation}
% and their product 
% is the determinant of the same matrix.
% Therefore after some manipulations it results:
% \begin{equation}
% \lambda_1 - \lambda_2 = \sqrt{(\mc_{11}-\mc_{22})^2+4\mc_{12}^2} 
% \label{Difference}\end{equation}
% and the two eigenvalues are then immediately obtained. \par
% It is convenient to compute the difference of the two diagonal elements 
% of the dispersion matrix and the difference of its eigenvalues.
%    \begin{macrocode}
\fp_gset:Nn   \g_BBLR_diag_diff_fp
{\g_BBLR_abscissa_SecOrdMoment_fp - \g_BBLR_ordinate_SecOrdMoment_fp}
\fp_gset:Nn   \g_BBLR_eig_diff_fp
{sqrt(\g_BBLR_diag_diff_fp * \g_BBLR_diag_diff_fp + 
       4 * \g_BBLR_mixed_SecOrdMoment_fp * \g_BBLR_mixed_SecOrdMoment_fp)}
%    \end{macrocode}
% The computation of $c$ and $s$ is obtained from \reff{spectral}
% taking into account that $c^2+s^2=1$.
% From \reff{spectral} it results: 
% \begin{equation} \mc_{11}-\mc_{22}=(\lambda_1-\lambda_2)(c^2-s^2)
% \label{Cos2A}\end{equation}
% and also 
% \begin{equation} \mc_{12}=(\lambda_1-\lambda_2)cs
% \label{Sin2A}\end{equation}
% that is only used to determine the sign of $cs$.
% The expression for the parameters $c$ and $s$ are:
% \begin{equation} 
% c=\sqrt{\frac{1}{2}+\frac{\mc_{11}-\mc_{22}}{2(\lambda_1-\lambda_2)}}
% \label{cos}\end{equation}
% \begin{equation} 
% s=\sgn(\mc_{12})
% \sqrt{\frac{1}{2}-\frac{\mc_{11}-\mc_{22}}{2(\lambda_1-\lambda_2)}}
% \label{sin}\end{equation}
% The parameters $s$ and $c$ are the sine and cosine 
% of the angle between the axis of $y_1$ and the eigenvector 
% corresponding to the maximum eigenvalue. \par
% They are computed using the already defined elements.
%    \begin{macrocode}
\fp_gset:Nn   \g_BBLR_cos_fp%
{sqrt((1 + \g_BBLR_diag_diff_fp / \g_BBLR_eig_diff_fp) / 2)}
\fp_gset:Nn \g_BBLR_sig_sin_fp {\fp_sign:n {\g_BBLR_mixed_SecOrdMoment_fp}}
\fp_gset:Nn   \g_BBLR_sin_fp                
{\g_BBLR_sig_sin_fp*sqrt((1-\g_BBLR_diag_diff_fp / \g_BBLR_eig_diff_fp) / 2)}
%    \end{macrocode}
% The vector $\coeff$ is :
% \begin{equation} 
% \coeff=\sgn(-s\barycenter{y}_1+c\barycenter{y}_2)
% \begin{pmatrix} -s \\ c\end{pmatrix}.
% \label{xhat}\end{equation}
%
%\par
% The parameter $a$ of model \reff{eqab} can be obtained as: 
% \begin{equation} 
% a=s/c
% \end{equation}
% Now the slope and the intercept of the optimal line corresponding to the
% symmetric criterion can be computed.
%
%    \begin{macrocode}
\fp_gset:Nn \g_BBLR_slope_S_fp
{\g_BBLR_sin_fp / \g_BBLR_cos_fp }
\fp_gset:Nn \g_BBLR_intercept_S_fp
{\g_BBLR_mean_ordinate_fp - \g_BBLR_slope_S_fp * \g_BBLR_mean_abscissa_fp}
}                    
%    \end{macrocode}
%
% The theoretical treatment of the proposed problem and the 
% implementation of its numerical solution end here.
% \end{macro}
%
% \subsection{Print of table of results}
% \begin{macro}{\lrprint}
% The  command \cs{lrprint} prints  some info on the data 
% and the results of the computations in tabular form.
%    \begin{macrocode}
\NewDocumentCommand{\lrprint}{}{
\begin{center}
\begin{tabular}{| l | r |} \hline
 Data~File:                & \g_BBLR_file_name_tl                   \\ \hline
 Number~of~points:         & \int_use:N\g_BBLR_number_of_points_int \\ \hline 
 Mean~values~of~the~coordinates:              &% 
 $\fp_use:N \g_BBLR_mean_abscissa_fp$ \\ &
 $\fp_use:N \g_BBLR_mean_ordinate_fp$ \\ \hline
 Minimum~values~of~the~coordinates:        &% 
 $\fp_use:N \g_BBLR_min_abscissa_fp$ \\ &
 $\fp_use:N \g_BBLR_min_ordinate_fp$    \\ \hline
  Maximum~values~of~the~coordinates:    &% 
 $\fp_use:N \g_BBLR_max_abscissa_fp$ \\ &
 $\fp_use:N \g_BBLR_max_ordinate_fp$    \\ \hline 
 {Second~order~moments}\phantom{xxxxxxxxx}{abscissa} &%
 $\fp_use:N \g_BBLR_abscissa_SecOrdMoment_fp$ \\
 \multicolumn{1}{|r|}{mixed}   & %
$\fp_use:N \g_BBLR_mixed_SecOrdMoment_fp$ ~ \\ 
\multicolumn{1}{|r|}{ordinate}   & %
$\fp_use:N \g_BBLR_ordinate_SecOrdMoment_fp$  \\ \hline 
  Slope~and~intercept~of~optimal~line  & $\fp_use:N \g_BBLR_slope_A_fp$ \\
 (estimated~with~errors~in~ordinate)&$\fp_use:N \g_BBLR_intercept_A_fp$\\ \hline
  Slope~and~intercept~of~optimal~line & $\fp_use:N \g_BBLR_slope_B_fp$ \\ 
 (estimated~with~errors~in~abscissa)&$\fp_use:N \g_BBLR_intercept_B_fp$\\ \hline
  Components~of~unit~vector~along~the~line & $\fp_use:N  \g_BBLR_cos_fp$ \\
  &  $\fp_use:N \g_BBLR_sin_fp$   \\
  Slope~and~intercept~of~optimal~line &$\fp_use:N \g_BBLR_slope_S_fp$ \\
(estimated~with~symmetric~regression) & 
 $\fp_use:N \g_BBLR_intercept_S_fp$\\ \hline 
\end{tabular}
\end{center}
}
%    \end{macrocode}
% \end{macro}
%
% \subsection{Plot of points and lines}   
% \begin{macro}{\lrplot}
% The  command \cs{lrplot} produce a framed plot of the data 
% and of the regression line(s). The size of the plot and its actual 
% content are determined by the arguments.
%    \begin{macrocode}
\NewDocumentCommand{\lrplot}{mmmmm}{%
%    \end{macrocode}
% The plotting area is divided into a main plotting area for 
% the representation of points and line(s) and a small surrounding free space. 
% The height is computed taking into account the distribution of the points.
%    \begin{macrocode}
\fp_gset:Nn \g_BBLR_base_fp {#1}
\fp_gset:Nn \g_BBLR_Xbase_fp {\g_BBLR_base_fp - 0.6}
\fp_gset:Nn \g_BBLR_scale_factor_fp{\g_BBLR_Xbase_fp / \g_BBLR_Dabscissa_fp}
\fp_gset:Nn \g_BBLR_Xheight_fp {\g_BBLR_Dordinate_fp * \g_BBLR_scale_factor_fp}
\fp_gset:Nn \g_BBLR_height_fp {\g_BBLR_Xheight_fp + 0.6}
%    \end{macrocode}
% The information about the items to be plotted is in the remaining arguments.
%    \begin{macrocode}
\str_gset:Nn \g_BBLR_point_code_str {#2}
\str_gset:Nn \g_BBLR_labelA_str     {#3}
\str_gset:Nn \g_BBLR_labelB_str     {#4}
\str_gset:Nn \g_BBLR_labelS_str     {#5}
\bool_gset:Nn \g_BBLR_plot_points_bool
{!(\str_if_eq_p:NN \g_BBLR_point_code_str \c_BBLR_minus_str)}
\bool_gset:Nn \g_BBLR_plot_lineA_bool
{!(\str_if_eq_p:NN \g_BBLR_labelA_str \c_BBLR_minus_str)}
\bool_gset:Nn \g_BBLR_plot_lineB_bool
{!(\str_if_eq_p:NN \g_BBLR_labelB_str \c_BBLR_minus_str)}
\bool_gset:Nn \g_BBLR_plot_lineS_bool
{!(\str_if_eq_p:NN \g_BBLR_labelS_str \c_BBLR_minus_str)}
%    \end{macrocode}
% The unit of length is $1$ centimeter. The plotting area is framed.
%    \begin{macrocode}
\setlength{\unitlength}{1.0cm}  
\fp_gset:Nn \g_tmpa_fp {\g_BBLR_Xbase_fp +0.2}
\fp_gset:Nn \g_tmpb_fp {\g_BBLR_Xheight_fp +0.1}
\begin{picture}(\fp_use:N\g_BBLR_base_fp,\fp_use:N\g_BBLR_height_fp)(-0.3,-0.3)
\put(-0.1,-0.1){\line(1,0){\fp_use:N\g_tmpa_fp}}
\put(-0.1,\fp_use:N\g_tmpb_fp){\line(1,0){\fp_use:N\g_tmpa_fp}}
\fp_gset:Nn \g_tmpa_fp {\g_tmpa_fp -0.1}
\fp_gset:Nn \g_tmpb_fp {\g_tmpb_fp +0.1}
\put(-0.1,-0.1){\line(0,1){\fp_use:N\g_tmpb_fp}}
\put(\fp_use:N\g_tmpa_fp,-0.1){\line(0,1){\fp_use:N\g_tmpb_fp}}
%    \end{macrocode}
% The plot of points and line(s) is obtained using auxiliary functions.
%    \begin{macrocode}
\thicklines
\bool_if:nT {\g_BBLR_plot_points_bool}{\BBLR_plot_points:}
\bool_if:nT {\g_BBLR_plot_lineA_bool}{
\BBLR_draw_line:NNN \g_BBLR_slope_A_fp\g_BBLR_intercept_A_fp\g_BBLR_labelA_str}
\bool_if:nT {\g_BBLR_plot_lineB_bool}{
\BBLR_draw_line:NNN \g_BBLR_slope_B_fp\g_BBLR_intercept_B_fp\g_BBLR_labelB_str}
\bool_if:nT {\g_BBLR_plot_lineS_bool}{
\BBLR_draw_line:NNN \g_BBLR_slope_S_fp\g_BBLR_intercept_S_fp\g_BBLR_labelS_str}
\end{picture}
}%                                                          
%    \end{macrocode}
% \end{macro}
%
% \subsection{Functions for internal use}
% The functions listed here after are for internal 
% use and are just minimally documented. \par
% The function |\BBLR_decode_data:|   
% \marginpar{\raggedleft\texttt{
% \textbackslash{}BBLR\textunderscore{}decode\textunderscore{}data:}}
% extract two numeric values from the string read from the file.
% Some tricky actions are necessary because 
% a so called csv file sometime do not contains the separating commas. 
%    \begin{macrocode}
\cs_new_protected:Nn \BBLR_decode_data: {             
\tl_trim_spaces:N \g_BBLR_file_line_tl                          
\int_gzero:N \g_tmpa_int                              
\int_gzero:N \g_BBLR_first_separator_int                      
\int_gzero:N \g_BBLR_last_separator_int               
\int_gset:Nn \g_BBLR_record_length_int {           
\str_count:N \g_BBLR_file_line_tl}                         
\str_map_variable:NNn \g_BBLR_file_line_tl \g_tmpa_str {
\int_gincr:N \g_tmpa_int                 
\bool_lazy_or:nnTF                                                       
{\str_if_eq_p:NN \g_tmpa_str \c_BBLR_comma_str}                   
{\str_if_eq_p:NN \g_tmpa_str \c_BBLR_space_str}                    
{\int_gset_eq:NN \g_BBLR_last_separator_int \g_tmpa_int
\int_if_zero:nTF {\g_BBLR_first_separator_int}
{\int_gset_eq:NN \g_BBLR_first_separator_int \g_tmpa_int
}{\prg_do_nothing:}                                              
}{\prg_do_nothing:}                                                
}                                        
\int_gincr:N \g_BBLR_last_separator_int                    
\int_gdecr:N \g_BBLR_first_separator_int                  
\fp_gset:Nn  \g_BBLR_abscissa_fp{                                   
\str_range:Nnn \g_BBLR_file_line_tl{1}{\g_BBLR_first_separator_int}}
\fp_gset:Nn  \g_BBLR_ordinate_fp{                                   
\str_range:Nnn \g_BBLR_file_line_tl                                 
{\g_BBLR_last_separator_int}{\g_BBLR_record_length_int}}            
}                                                
%    \end{macrocode}
% The function |\BBLR_plot_points:|  \marginpar{\raggedleft\texttt{
% \textbackslash{}BBLR\textunderscore{}plot\textunderscore{}points:}}
% scans the data file to read the coordinates and 
% it draws a circle for each point.
% 
%   \begin{macrocode}
\cs_new_protected:Nn  \BBLR_plot_points: {
\ior_open:Nn   \g_BBLR_file_ior \g_BBLR_file_name_tl              
\int_zero:N    \g_BBLR_rec_count_int        
\int_do_until:nn                        
{\g_BBLR_rec_count_int = \g_BBLR_number_of_points_int} 
{                                                          
    \ior_str_get:NN \g_BBLR_file_ior \g_BBLR_file_line_tl    
    \int_incr:N \g_BBLR_rec_count_int               
    \BBLR_decode_data:                                  
    \fp_gset:Nn \g_tmpa_fp{(\g_BBLR_abscissa_fp-\g_BBLR_min_abscissa_fp)* 
    \g_BBLR_scale_factor_fp}
    \fp_gset:Nn \g_tmpb_fp{(\g_BBLR_ordinate_fp-\g_BBLR_min_ordinate_fp)* 
    \g_BBLR_scale_factor_fp}
    \put(\fp_use:N\g_tmpa_fp, \fp_use:N\g_tmpb_fp){           
    {\circle*{\fp_use:N\g_BBLR_diameter_fp}}}                  
}                                                            
\ior_close:N \g_BBLR_file_ior                                    
}                                                
%    \end{macrocode}
% The function |\BBLR_draw_line:NNN|  \marginpar{\raggedleft\texttt{
% \textbackslash{}BBLR\textunderscore{}draw\textunderscore{}line:NNN}}
% draws the line. The first two parameters given as arguments 
% are the slope and the intercept. The third parameter is a label.
% \par The next code finds the intersection of the line with the plotting area.
%    \begin{macrocode}
\cs_new_protected:Nn  \BBLR_draw_line:NNN {
\fp_gset:Nn \fp_tmpa_fp {#1 * \g_BBLR_min_abscissa_fp + #2 }
\fp_compare:nTF{\fp_tmpa_fp > \g_BBLR_max_ordinate_fp}{
\fp_gset:Nn \g_BBLR_min_draw_abscissa_fp {(\g_BBLR_max_ordinate_fp -#2) / #1}
}{
\fp_compare:nTF{\fp_tmpa_fp < \g_BBLR_min_ordinate_fp}{
\fp_gset:Nn \g_BBLR_min_draw_abscissa_fp {(\g_BBLR_min_ordinate_fp - #2) / #1}
}{
\fp_gset:Nn \g_BBLR_min_draw_abscissa_fp { \g_BBLR_min_abscissa_fp }
}}
\fp_gset:Nn \fp_tmpa_fp {#1 * \g_BBLR_max_abscissa_fp + #2 }
\fp_compare:nTF{\fp_tmpa_fp > \g_BBLR_max_ordinate_fp}{
\fp_gset:Nn \g_BBLR_max_draw_abscissa_fp {(\g_BBLR_max_ordinate_fp -#2) / #1}
}{
\fp_compare:nTF{\fp_tmpa_fp < \g_BBLR_min_ordinate_fp}{
\fp_gset:Nn \g_BBLR_max_draw_abscissa_fp { (\g_BBLR_min_ordinate_fp - #2) / #1}
}{
\fp_gset:Nn \g_BBLR_max_draw_abscissa_fp {  \g_BBLR_max_abscissa_fp }
}}
%    \end{macrocode}
% Some parameters (i.e.\ starting point and base-length) 
% are computed and the line is drawn.
%    \begin{macrocode}
\fp_gset:Nn \fp_tmpa_fp {(\g_BBLR_min_draw_abscissa_fp -              
\g_BBLR_min_abscissa_fp)* \g_BBLR_scale_factor_fp}           
\fp_gset:Nn \fp_tmpb_fp {(#1 * \g_BBLR_min_draw_abscissa_fp + #2 -
\g_BBLR_min_ordinate_fp)* \g_BBLR_scale_factor_fp}             
\fp_gset:Nn \fp_BBLR_line_base_length_fp{(\g_BBLR_max_draw_abscissa_fp -
\g_BBLR_min_draw_abscissa_fp) * \g_BBLR_scale_factor_fp}   
\put(\fp_use:N\fp_tmpa_fp, \fp_use:N\fp_tmpb_fp){          
\line(1.,\fp_use:N #1){\fp_use:N\fp_BBLR_line_base_length_fp}}
%    \end{macrocode}
%The third parameter is used as a label, if it is not a |+|.
%    \begin{macrocode}
\bool_if:nF {\str_if_eq_p:NN #3 \c_BBLR_plus_str}{       
\fp_gset:Nn \fp_tmpa_fp                                    
{0.08 * \g_BBLR_min_draw_abscissa_fp + 0.92 * \g_BBLR_max_draw_abscissa_fp} 
\fp_gset:Nn \fp_tmpb_fp {#1 * \fp_tmpa_fp + #2 }
\fp_gset:Nn \fp_tmpa_fp 
{(\fp_tmpa_fp-\g_BBLR_min_abscissa_fp)*\g_BBLR_scale_factor_fp 
+ 0.3 * #1 /sqrt(1.+#1*#1)}
\fp_gset:Nn \fp_tmpb_fp 
{(\fp_tmpb_fp-\g_BBLR_min_ordinate_fp)* \g_BBLR_scale_factor_fp
- 0.3 /sqrt(1.+#1*#1)}
\put(\fp_use:N\fp_tmpa_fp, \fp_use:N\fp_tmpb_fp){#3}
}                                                             
}                                                 
%    \end{macrocode}
%
%    \begin{macrocode}
\ExplSyntaxOff
%    \end{macrocode}
%
%
%\iffalse
%</package>
%\fi
%
% \section{Acknowledgments}
% The colleagues Paolo Zatelli, Alfonso Vitti and Giulia Graldi
% read some preliminary version 
% of this text and suggested several improvements. \par
%
% \section{About the references}
% \subsection*{Mathematics}
% The books by Lang \cite{Lang} and by Strang \cite{Strang} give 
% all the background on linear algebra.\par
% The texts by Sansò \cites{Sanso1, Sanso2} (in italian) treat the
% teory of probability and its application to metrology. 
% See: |http://www.geolab.polimi.it/text-books/|.\par
% The paper  by Karl Pearson \cite{Pearson} is the oldest text that 
% I have found on the symmetric regression, or total regression.
% \subsection*{Programming}
% The two documents \cites{L3A, L3B} are the fountamental and official guide
% for \LaTeX3 programming. The books by Donald Knuth \cites{Knuth} 
% and Leslie Lamport \cites{Lamport} are still essential references. 
% The papers by Enrico Gregorio \cites{egreg1, egreg2, egreg3, egreg4, egreg5}
% explain some general and some special aspect of \LaTeX3 programming. 
%
% \section{References}
% \begin{biblist}[\normalsize]   
% \bib{egreg1}{article}{
% author={Gregorio, Enrico}, 
% journal={ArsTeXnica},
% number={14},pages={41\ndash 47}, date={2012},
% title={\LaTeX3: un nuovo gioco per i maghi e per diventarlo},
% }
% \bib{egreg2}{article}{
% author={Gregorio, Enrico}, 
% journal={ArsTeXnica},
% number={22},pages={69\ndash 77}, date={2016},
% title={Liste, cicli, \LaTeX3},
% }
% \bib{egreg3}{article}{
% author={Gregorio, Enrico}, 
% journal={ArsTeXnica},
% number={24},pages={37\ndash 44}, date={2017},
% title={Condizionali in \LaTeX},
% }
% \bib{egreg4}{article}{
% author={Gregorio, Enrico}, 
% journal={ArsTeXnica},
% number={30},pages={36\ndash 45}, date={2020},
% title={Funzioni e |expl3|},
% }
% \bib{egreg5}{article}{
% author={Gregorio, Enrico}, 
% journal={TUGboat},
% volume={41},number={3},pages={299\ndash 307}, date={2020},
% title={Functions and |expl3|},
% }
% \bib{Knuth}{book}{
% author={Knuth, Donald},
% title={The TeXbook},
% date={1986},
% publisher={American Mathematical Society and Addison-Wesley},
% } 
% \bib{Lang}{book}{
% author={Lang, Serge},
% title={Linear Algebra},
% date={1987},
% publisher={Springer-Verlag},
% place={Berlin Heidelberg},
% }
% \bib{Lamport}{book}{
% author={Lamport, Leslie},
% title={LaTeX - A document preparation system (2nd ed.\ )},
% date={1994},
% publisher={Addison-Wesley},
% note={something interesting in the fist edition, too},
% }
% \bib{L3A}{article}{
% title={The |expl3| package and LaTeX3 programming},
% author={The LaTeX project team}, date={2024},
% note={file: |expl3.pdf| available in CTAN in l3kernel}, 
% }
% \bib{L3B}{article}{
% title={The \LaTeX3 interface},
% author={The LaTeX project team}, date={2024},
% note={file: |interface3.pdf| available in CTAN in l3kernel} 
% }
% \bib{Pearson}{article}{
% title={On lines and planes of closest fit to systems of points in space},
% author={Pearson, Karl}, date={1901},
% journal={Philosophical Magazine},
% volume={2},number={11},pages={559\ndash 572}, 
% }
% \bib{Sanso1}{book}{
% author={Sansò, Fernando},
% title={Elementi di teoria della probabilità},
% date={1996},
% publisher={Città-Studi},
% place={Milano},
% }
% \bib{Sanso2}{book}{
% author={Sansò, Fernando},
% title={La teoria della stima},
% date={1996},
% publisher={Città-Studi},
% place={Milano},
% }
% \bib{Strang}{book}{
% author={Strang, Gilbert},
% title={Introduction to linear algebra},
% date={2009},
% publisher={Wellesley-Cambridge press,},
% }
% \end{biblist} 
%
% \par\vfill\centerline{\small ***}\vfill
% \end{document} 
% 
%\iffalse
% END OF FILE linearregression.dtx
%\fi