1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2017 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"
23 #include <libpspp/hash-functions.h>
24 #include <libpspp/message.h>
25 #include <data/casegrouper.h>
26 #include <data/casereader.h>
27 #include <data/dictionary.h>
28 #include <data/variable.h>
31 #define _(msgid) gettext (msgid)
32 #define N_(msgid) msgid
36 This module interprets a "data matrix", typically generated by the command
37 MATRIX DATA. The dictionary of such a matrix takes the form:
39 s_0, s_1, ... s_m, ROWTYPE_, VARNAME_, v_0, v_1, .... v_n
41 where s_0, s_1 ... s_m are the variables defining the splits, and
42 v_0, v_1 ... v_n are the continuous variables.
46 The ROWTYPE_ variable is of type A8.
47 The VARNAME_ variable is a string type whose width is not predetermined.
48 The variables s_x are of type F4.0 (although this reader accepts any type),
49 and v_x are of any numeric type.
51 The values of the ROWTYPE_ variable are in the set {MEAN, STDDEV, N, CORR, COV}
52 and determine the purpose of that case.
53 The values of the VARNAME_ variable must correspond to the names of the varibles
54 in {v_0, v_1 ... v_n} and indicate the rows of the correlation or covariance
59 A typical example is as follows:
61 s_0 ROWTYPE_ VARNAME_ v_0 v_1 v_2
63 0 MEAN 5.0000 4.0000 3.0000
64 0 STDDEV 1.0000 2.0000 3.0000
65 0 N 9.0000 9.0000 9.0000
66 0 CORR V1 1.0000 .6000 .7000
67 0 CORR V2 .6000 1.0000 .8000
68 0 CORR V3 .7000 .8000 1.0000
69 1 MEAN 9.0000 8.0000 7.0000
70 1 STDDEV 5.0000 6.0000 7.0000
71 1 N 9.0000 9.0000 9.0000
72 1 CORR V1 1.0000 .4000 .3000
73 1 CORR V2 .4000 1.0000 .2000
74 1 CORR V3 .3000 .2000 1.0000
80 const struct dictionary *dict;
81 const struct variable *varname;
82 const struct variable *rowtype;
83 struct casegrouper *grouper;
85 gsl_matrix *n_vectors;
86 gsl_matrix *mean_vectors;
87 gsl_matrix *var_vectors;
89 // gsl_matrix *correlation;
90 // gsl_matrix *covariance;
93 struct matrix_reader *
94 create_matrix_reader_from_case_reader (const struct dictionary *dict, struct casereader *in_reader,
95 const struct variable ***vars, size_t *n_vars)
97 struct matrix_reader *mr = xzalloc (sizeof *mr);
100 mr->varname = dict_lookup_var (dict, "varname_");
101 if (mr->varname == NULL)
103 msg (ME, _("Matrix dataset lacks a variable called %s."), "VARNAME_");
108 mr->rowtype = dict_lookup_var (dict, "rowtype_");
109 if (mr->rowtype == NULL)
111 msg (ME, _("Matrix dataset lacks a variable called %s."), "ROWTYPE_");
117 const struct variable **dvars = NULL;
118 dict_get_vars (dict, &dvars, &dvarcnt, DC_SCRATCH);
121 *n_vars = dvarcnt - var_get_dict_index (mr->varname) - 1;
126 *vars = xcalloc (sizeof (struct variable **), *n_vars);
128 for (i = 0; i < *n_vars; ++i)
130 (*vars)[i] = dvars[i + var_get_dict_index (mr->varname) + 1];
134 /* All the variables before ROWTYPE_ (if any) are split variables */
135 mr->grouper = casegrouper_create_vars (in_reader, dvars, var_get_dict_index (mr->rowtype));
143 destroy_matrix_reader (struct matrix_reader *mr)
147 bool ret = casegrouper_destroy (mr->grouper);
154 Allocates MATRIX if necessary,
155 and populates row MROW, from the data in C corresponding to
156 variables in VARS. N_VARS is the length of VARS.
159 matrix_fill_row (gsl_matrix **matrix,
160 const struct ccase *c, int mrow,
161 const struct variable **vars, size_t n_vars)
165 *matrix = gsl_matrix_alloc (n_vars, n_vars);
167 for (col = 0; col < n_vars; ++col)
169 const struct variable *cv = vars [col];
170 double x = case_data (c, cv)->f;
171 assert (col < (*matrix)->size2);
172 assert (mrow < (*matrix)->size1);
173 gsl_matrix_set (*matrix, mrow, col, x);
178 next_matrix_from_reader (struct matrix_material *mm,
179 struct matrix_reader *mr,
180 const struct variable **vars, int n_vars)
182 struct casereader *group;
186 gsl_matrix_free (mr->n_vectors);
187 gsl_matrix_free (mr->mean_vectors);
188 gsl_matrix_free (mr->var_vectors);
190 if (!casegrouper_get_next_group (mr->grouper, &group))
193 mr->n_vectors = gsl_matrix_alloc (n_vars, n_vars);
194 mr->mean_vectors = gsl_matrix_alloc (n_vars, n_vars);
195 mr->var_vectors = gsl_matrix_alloc (n_vars, n_vars);
197 mm->n = mr->n_vectors;
198 mm->mean_matrix = mr->mean_vectors;
199 mm->var_matrix = mr->var_vectors;
201 // FIXME: Make this into a hash table.
202 unsigned long *table = xmalloc (sizeof (*table) * n_vars);
204 for (i = 0; i < n_vars; ++i)
206 const int w = var_get_width (mr->varname);
209 const char *name = var_get_name (vars[i]);
211 unsigned long h = hash_bytes (s, w, 0);
216 for ( ; (c = casereader_read (group) ); case_unref (c))
218 const union value *uv = case_data (c, mr->rowtype);
220 for (col = 0; col < n_vars; ++col)
222 const struct variable *cv
223 = vars ? vars[col] : dict_get_var (mr->dict, var_get_dict_index (mr->varname) + 1 + col);
224 double x = case_data (c, cv)->f;
225 if (0 == strncasecmp ((char *)value_str (uv, 8), "N ", 8))
226 for (row = 0; row < n_vars; ++row)
227 gsl_matrix_set (mr->n_vectors, row, col, x);
228 else if (0 == strncasecmp ((char *) value_str (uv, 8), "MEAN ", 8))
229 for (row = 0; row < n_vars; ++row)
230 gsl_matrix_set (mr->mean_vectors, row, col, x);
231 else if (0 == strncasecmp ((char *) value_str (uv, 8), "STDDEV ", 8))
232 for (row = 0; row < n_vars; ++row)
233 gsl_matrix_set (mr->var_vectors, row, col, x * x);
236 const union value *uvv = case_data (c, mr->varname);
237 const uint8_t *vs = value_str (uvv, var_get_width (mr->varname));
238 int w = var_get_width (mr->varname);
239 unsigned long h = hash_bytes (vs, w, 0);
242 for (i = 0; i < n_vars; ++i)
254 if (0 == strncasecmp ((char *) value_str (uv, 8), "CORR ", 8))
256 matrix_fill_row (&mm->corr, c, mrow, vars, n_vars);
258 else if (0 == strncasecmp ((char *) value_str (uv, 8), "COV ", 8))
260 matrix_fill_row (&mm->cov, c, mrow, vars, n_vars);
264 casereader_destroy (group);