1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2017, 2019 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
19 #include "matrix-reader.h"
24 #include "data/casegrouper.h"
25 #include "data/casereader.h"
26 #include "data/data-out.h"
27 #include "data/dataset.h"
28 #include "data/dictionary.h"
29 #include "data/format.h"
30 #include "data/variable.h"
31 #include "language/command.h"
32 #include "libpspp/i18n.h"
33 #include "libpspp/message.h"
34 #include "libpspp/str.h"
35 #include "output/pivot-table.h"
38 #define _(msgid) gettext (msgid)
39 #define N_(msgid) msgid
44 This module interprets a "data matrix", typically generated by the command
45 MATRIX DATA. The dictionary of such a matrix takes the form:
47 s_0, s_1, ... s_m, ROWTYPE_, VARNAME_, v_0, v_1, .... v_n
49 where s_0, s_1 ... s_m are the variables defining the splits, and
50 v_0, v_1 ... v_n are the continuous variables.
54 The ROWTYPE_ variable is of type A8.
55 The VARNAME_ variable is a string type whose width is not predetermined.
56 The variables s_x are of type F4.0 (although this reader accepts any type),
57 and v_x are of any numeric type.
59 The values of the ROWTYPE_ variable are in the set {MEAN, STDDEV, N, CORR, COV}
60 and determine the purpose of that case.
61 The values of the VARNAME_ variable must correspond to the names of the varibles
62 in {v_0, v_1 ... v_n} and indicate the rows of the correlation or covariance
67 A typical example is as follows:
69 s_0 ROWTYPE_ VARNAME_ v_0 v_1 v_2
71 0 MEAN 5.0000 4.0000 3.0000
72 0 STDDEV 1.0000 2.0000 3.0000
73 0 N 9.0000 9.0000 9.0000
74 0 CORR V1 1.0000 .6000 .7000
75 0 CORR V2 .6000 1.0000 .8000
76 0 CORR V3 .7000 .8000 1.0000
77 1 MEAN 9.0000 8.0000 7.0000
78 1 STDDEV 5.0000 6.0000 7.0000
79 1 N 9.0000 9.0000 9.0000
80 1 CORR V1 1.0000 .4000 .3000
81 1 CORR V2 .4000 1.0000 .2000
82 1 CORR V3 .3000 .2000 1.0000
87 matrix_material_uninit (struct matrix_material *mm)
89 gsl_matrix_free (mm->corr);
90 gsl_matrix_free (mm->cov);
91 gsl_matrix_free (mm->n);
92 gsl_matrix_free (mm->mean_matrix);
93 gsl_matrix_free (mm->var_matrix);
96 static const struct variable *
97 find_matrix_string_var (const struct dictionary *dict, const char *name)
99 const struct variable *var = dict_lookup_var (dict, name);
102 msg (ME, _("Matrix dataset lacks a variable called %s."), name);
105 if (!var_is_alpha (var))
107 msg (ME, _("Matrix dataset variable %s should be of string type."), name);
113 struct matrix_reader *
114 matrix_reader_create (const struct dictionary *dict,
115 struct casereader *in_reader)
117 const struct variable *varname = find_matrix_string_var (dict, "VARNAME_");
118 const struct variable *rowtype = find_matrix_string_var (dict, "ROWTYPE_");
119 if (!varname || !rowtype)
122 size_t varname_idx = var_get_dict_index (varname);
123 size_t rowtype_idx = var_get_dict_index (rowtype);
124 if (varname_idx < rowtype_idx)
126 msg (ME, _("Variable %s must precede %s in matrix file dictionary."),
127 "ROWTYPE_", "VARNAME_");
131 for (size_t i = 0; i < dict_get_var_cnt (dict); i++)
133 const struct variable *v = dict_get_var (dict, i);
134 if (!var_is_numeric (v) && v != rowtype && v != varname)
136 msg (ME, _("Matrix dataset variable %s should be numeric."),
143 const struct variable **vars;
144 dict_get_vars (dict, &vars, &n_vars, DC_SCRATCH);
146 /* Different kinds of variables. */
147 size_t first_svar = 0;
148 size_t n_svars = rowtype_idx;
149 size_t first_fvar = rowtype_idx + 1;
150 size_t n_fvars = varname_idx - rowtype_idx - 1;
151 size_t first_cvar = varname_idx + 1;
152 size_t n_cvars = n_vars - varname_idx - 1;
155 msg (ME, _("Matrix dataset does not have any continuous variables."));
160 struct matrix_reader *mr = xmalloc (sizeof *mr);
161 *mr = (struct matrix_reader) {
163 .grouper = casegrouper_create_vars (in_reader, &vars[first_svar], n_svars),
164 .svars = xmemdup (vars + first_svar, n_svars * sizeof *mr->svars),
167 .fvars = xmemdup (vars + first_fvar, n_fvars * sizeof *mr->fvars),
170 .cvars = xmemdup (vars + first_cvar, n_cvars * sizeof *mr->cvars),
179 matrix_reader_destroy (struct matrix_reader *mr)
183 bool ret = casegrouper_destroy (mr->grouper);
190 Allocates MATRIX if necessary,
191 and populates row MROW, from the data in C corresponding to
192 variables in VARS. N_VARS is the length of VARS.
195 matrix_fill_row (gsl_matrix **matrix,
196 const struct ccase *c, int mrow,
197 const struct variable **vars, size_t n_vars)
202 *matrix = gsl_matrix_alloc (n_vars, n_vars);
203 gsl_matrix_set_all (*matrix, SYSMIS);
206 for (col = 0; col < n_vars; ++col)
208 const struct variable *cv = vars [col];
209 double x = case_num (c, cv);
210 assert (col < (*matrix)->size2);
211 assert (mrow < (*matrix)->size1);
212 gsl_matrix_set (*matrix, mrow, col, x);
217 find_varname (const struct variable **vars, int n_vars,
220 for (int i = 0; i < n_vars; i++)
221 if (!strcasecmp (var_get_name (vars[i]), varname))
227 matrix_reader_get_string (const struct ccase *c, const struct variable *var)
229 struct substring s = case_ss (c, var);
230 ss_rtrim (&s, ss_cstr (CC_SPACES));
235 matrix_reader_set_string (struct ccase *c, const struct variable *var,
236 struct substring src)
238 struct substring dst = case_ss (c, var);
239 for (size_t i = 0; i < dst.length; i++)
240 dst.string[i] = i < src.length ? src.string[i] : ' ';
244 matrix_reader_next (struct matrix_material *mm, struct matrix_reader *mr,
245 struct casereader **groupp)
247 struct casereader *group;
248 if (!casegrouper_get_next_group (mr->grouper, &group))
250 *mm = (struct matrix_material) MATRIX_MATERIAL_INIT;
257 *groupp = casereader_clone (group);
259 const struct variable **vars = mr->cvars;
260 size_t n_vars = mr->n_cvars;
262 *mm = (struct matrix_material) { .n = NULL };
271 struct matrix matrices[] = {
272 { .name = "CORR", .m = &mm->corr },
273 { .name = "COV", .m = &mm->cov },
275 enum { N_MATRICES = 2 };
278 for (; (c = casereader_read (group)); case_unref (c))
280 struct substring rowtype = matrix_reader_get_string (c, mr->rowtype);
283 = (ss_equals_case (rowtype, ss_cstr ("N")) ? &mm->n
284 : ss_equals_case (rowtype, ss_cstr ("MEAN")) ? &mm->mean_matrix
285 : ss_equals_case (rowtype, ss_cstr ("STDDEV")) ? &mm->var_matrix
290 *v = gsl_matrix_calloc (n_vars, n_vars);
292 for (int x = 0; x < n_vars; ++x)
294 double n = case_num (c, vars[x]);
295 if (v == &mm->var_matrix)
297 for (int y = 0; y < n_vars; ++y)
298 gsl_matrix_set (*v, y, x, n);
303 struct matrix *m = NULL;
304 for (size_t i = 0; i < N_MATRICES; i++)
305 if (ss_equals_case (rowtype, ss_cstr (matrices[i].name)))
312 struct substring varname_raw = case_ss (c, mr->varname);
313 struct substring varname = ss_cstr (
314 recode_string (UTF8, dict_get_encoding (mr->dict),
315 varname_raw.string, varname_raw.length));
316 ss_rtrim (&varname, ss_cstr (CC_SPACES));
317 varname.string[varname.length] = '\0';
319 int y = find_varname (vars, n_vars, varname.string);
323 matrix_fill_row (m->m, c, y, vars, n_vars);
327 ss_dealloc (&varname);
330 casereader_destroy (group);
332 for (size_t i = 0; i < N_MATRICES; i++)
333 if (matrices[i].good_rows && matrices[i].good_rows != n_vars)
334 msg (SW, _("%s matrix has %zu columns but %zu rows named variables "
335 "to be analyzed (and %zu rows named unknown variables)."),
336 matrices[i].name, n_vars, matrices[i].good_rows,
337 matrices[i].bad_rows);
343 cmd_debug_matrix_read (struct lexer *lexer UNUSED, struct dataset *ds)
345 struct matrix_reader *mr = matrix_reader_create (dataset_dict (ds),
350 struct pivot_table *pt = pivot_table_create ("Debug Matrix Reader");
360 const char *mm_stat_names[] = {
361 [MM_CORR] = "Correlation",
362 [MM_COV] = "Covariance",
365 [MM_STDDEV] = "Standard Deviation",
367 enum { N_STATS = sizeof mm_stat_names / sizeof *mm_stat_names };
368 for (size_t i = 0; i < 2; i++)
370 struct pivot_dimension *d = pivot_dimension_create (
372 i ? PIVOT_AXIS_COLUMN : PIVOT_AXIS_ROW,
373 i ? "Column" : "Row");
375 pivot_category_create_leaf_rc (d->root, pivot_value_new_text ("Value"),
376 PIVOT_RC_CORRELATION);
377 for (size_t j = 0; j < mr->n_cvars; j++)
378 pivot_category_create_leaf_rc (
379 d->root, pivot_value_new_variable (mr->cvars[j]),
380 PIVOT_RC_CORRELATION);
383 struct pivot_dimension *stat = pivot_dimension_create (pt, PIVOT_AXIS_ROW,
385 for (size_t i = 0; i < N_STATS; i++)
386 pivot_category_create_leaf (stat->root,
387 pivot_value_new_text (mm_stat_names[i]));
389 struct pivot_dimension *split = pivot_dimension_create (
390 pt, PIVOT_AXIS_ROW, "Split");
394 struct matrix_material mm = MATRIX_MATERIAL_INIT;
395 while (matrix_reader_next (&mm, mr, NULL))
397 pivot_category_create_leaf (split->root,
398 pivot_value_new_integer (split_num + 1));
400 const gsl_matrix *m[N_STATS] = {
404 [MM_MEAN] = mm.mean_matrix,
405 [MM_STDDEV] = mm.var_matrix,
408 for (size_t i = 0; i < N_STATS; i++)
411 if (i == MM_COV || i == MM_CORR)
413 for (size_t y = 0; y < mr->n_cvars; y++)
414 for (size_t x = 0; x < mr->n_cvars; x++)
416 pt, y + 1, x, i, split_num,
417 pivot_value_new_number (gsl_matrix_get (m[i], y, x)));
420 for (size_t x = 0; x < mr->n_cvars; x++)
422 double n = gsl_matrix_get (m[i], 0, x);
425 pivot_table_put4 (pt, 0, x, i, split_num,
426 pivot_value_new_number (n));
431 matrix_material_uninit (&mm);
433 pivot_table_submit (pt);
437 matrix_reader_destroy (mr);