Changeset: 8eb4729bbdb8 for MonetDB
URL: http://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=8eb4729bbdb8
Modified Files:
monetdb5/optimizer/opt_candidates.c
monetdb5/optimizer/opt_prelude.c
monetdb5/optimizer/opt_prelude.h
monetdb5/optimizer/opt_support.c
sql/backends/monet5/rel_bin.c
sql/backends/monet5/sql_statement.c
sql/backends/monet5/sql_statement.h
sql/server/rel_select.c
sql/server/sql_parser.h
sql/server/sql_parser.y
sql/server/sql_scan.c
Branch: stratified_sampling
Log Message:
Add weighted sampling to the SQL layer.
The syntax for weighted sampling is:
SELECT * FROM <table> SAMPLE <id> WITH WEIGHTS <weight_column>;
diffs (232 lines):
diff --git a/monetdb5/optimizer/opt_candidates.c
b/monetdb5/optimizer/opt_candidates.c
--- a/monetdb5/optimizer/opt_candidates.c
+++ b/monetdb5/optimizer/opt_candidates.c
@@ -64,7 +64,7 @@ OPTcandidatesImplementation(Client cntxt
setVarCList(mb,getArg(p,0));
}
else if (getModuleId(p) == sampleRef) {
- if (getFunctionId(p) == subuniformRef)
+ if (getFunctionId(p) == subuniformRef ||
getFunctionId(p) == subweightedRef)
setVarCList(mb, getArg(p, 0));
}
else if (getModuleId(p) == groupRef && p->retc > 1) {
diff --git a/monetdb5/optimizer/opt_prelude.c b/monetdb5/optimizer/opt_prelude.c
--- a/monetdb5/optimizer/opt_prelude.c
+++ b/monetdb5/optimizer/opt_prelude.c
@@ -286,6 +286,7 @@ str not_uniqueRef;
str sampleRef;
str uniqueRef;
str subuniformRef;
+str subweightedRef;
str unlockRef;
str unpackRef;
str unpinRef;
@@ -567,6 +568,7 @@ void optimizerInit(void)
not_uniqueRef= putName("not_unique");
sampleRef= putName("sample");
uniqueRef= putName("unique");
+ subweightedRef = putName("subweighted");
subuniformRef= putName("subuniform");
unlockRef= putName("unlock");
unpackRef = putName("unpack");
diff --git a/monetdb5/optimizer/opt_prelude.h b/monetdb5/optimizer/opt_prelude.h
--- a/monetdb5/optimizer/opt_prelude.h
+++ b/monetdb5/optimizer/opt_prelude.h
@@ -280,6 +280,7 @@ mal_export str not_uniqueRef;
mal_export str sampleRef;
mal_export str uniqueRef;
mal_export str subuniformRef;
+mal_export str subweightedRef;
mal_export str unpackRef;
mal_export str unpinRef;
mal_export str unlockRef;
diff --git a/monetdb5/optimizer/opt_support.c b/monetdb5/optimizer/opt_support.c
--- a/monetdb5/optimizer/opt_support.c
+++ b/monetdb5/optimizer/opt_support.c
@@ -549,7 +549,9 @@ isSlice(InstrPtr p)
int
isSample(InstrPtr p)
{
- return (getModuleId(p) == sampleRef && getFunctionId(p) ==
subuniformRef);
+ return (getModuleId(p) == sampleRef &&
+ (getFunctionId(p) == subuniformRef ||
+ getFunctionId(p) == subweightedRef));
}
int isOrderby(InstrPtr p){
diff --git a/sql/backends/monet5/rel_bin.c b/sql/backends/monet5/rel_bin.c
--- a/sql/backends/monet5/rel_bin.c
+++ b/sql/backends/monet5/rel_bin.c
@@ -530,9 +530,8 @@ exp_bin(backend *be, sql_exp *e, stmt *l
list *ops;
node *n;
int first = 1;
-
- ops = sa_list(sql->sa);
- args = e->l;
+ ops = sa_list(sql->sa);
+ args = e->l;
for( n = args->h; n; n = n->next ) {
s = NULL;
if (!swapped)
@@ -2794,12 +2793,23 @@ rel2bin_sample(backend *be, sql_rel *rel
const char *tname = table_name(sql->sa, sc);
s = exp_bin(be, rel->exps->h->data, NULL, NULL, NULL, NULL,
NULL, NULL);
-
if (!s)
s = stmt_atom_lng_nil(be);
sc = column(be, sc);
- sample = stmt_sample(be, stmt_alias(be, sc, tname, cname),s);
+
+ if (rel->exps->h->next) {
+ stmt* left = rel_bin(be, rel->l);
+ stmt* right = rel_bin(be, rel->r);
+ // weighted sampling
+ stmt* weights = exp_bin(be, rel->exps->h->next->data,
left, right, NULL, NULL, NULL, NULL);
+ if (!weights)
+ return NULL;
+
+ sample = stmt_weighted_sample(be, stmt_alias(be, sc,
tname, cname), s, weights);
+ } else {
+ sample = stmt_sample(be, stmt_alias(be, sc, tname,
cname), s);
+ }
for ( ; n; n = n->next) {
stmt *sc = n->data;
diff --git a/sql/backends/monet5/sql_statement.c
b/sql/backends/monet5/sql_statement.c
--- a/sql/backends/monet5/sql_statement.c
+++ b/sql/backends/monet5/sql_statement.c
@@ -1013,6 +1013,34 @@ stmt_sample(backend *be, stmt *s, stmt *
return NULL;
}
+stmt *
+stmt_weighted_sample(backend *be, stmt *s, stmt *sample, stmt *weights)
+{
+ MalBlkPtr mb = be->mb;
+ InstrPtr q = NULL;
+
+ if (s->nr < 0 || sample->nr < 0 || weights->nr < 0)
+ return NULL;
+
+ q = newStmt(mb, sampleRef, subweightedRef);
+ q = pushArgument(mb, q, s->nr);
+ q = pushArgument(mb, q, sample->nr);
+ q = pushArgument(mb, q, weights->nr);
+ if (q) {
+ stmt *ns = stmt_create(be->mvc->sa, st_sample);
+
+ ns->op1 = s;
+ ns->op2 = sample;
+ ns->nrcols = s->nrcols;
+ ns->key = s->key;
+ ns->aggr = s->aggr;
+ ns->flag = 0;
+ ns->q = q;
+ ns->nr = getDestVar(q);
+ return ns;
+ }
+ return NULL;
+}
stmt *
stmt_order(backend *be, stmt *s, int direction)
diff --git a/sql/backends/monet5/sql_statement.h
b/sql/backends/monet5/sql_statement.h
--- a/sql/backends/monet5/sql_statement.h
+++ b/sql/backends/monet5/sql_statement.h
@@ -208,6 +208,7 @@ extern stmt *stmt_result(backend *be, st
*/
extern stmt *stmt_limit(backend *sa, stmt *c, stmt *piv, stmt *gid, stmt
*offset, stmt *limit, int distinct, int dir, int last, int order);
extern stmt *stmt_sample(backend *be, stmt *s, stmt *sample);
+extern stmt *stmt_weighted_sample(backend *be, stmt *s, stmt *sample, stmt
*weights);
extern stmt *stmt_order(backend *be, stmt *s, int direction);
extern stmt *stmt_reorder(backend *be, stmt *s, int direction, stmt
*orderby_ids, stmt *orderby_grp);
diff --git a/sql/server/rel_select.c b/sql/server/rel_select.c
--- a/sql/server/rel_select.c
+++ b/sql/server/rel_select.c
@@ -4855,10 +4855,29 @@ rel_select_exp(mvc *sql, sql_rel *rel, S
if (sn->sample) {
list *exps = new_exp_list(sql->sa);
- sql_exp *o = rel_value_exp( sql, NULL, sn->sample, 0, ek);
- if (!o)
- return NULL;
- append(exps, o);
+ if (sn->sample->token == SQL_WEIGHTED_SAMPLE) {
+ // weighted sampling
+ // parse the sample size and weight vector and pass it
on to rel_sample
+ dlist *l = sn->sample->data.lval;
+
+ lng sample_size = l->h->data.l_val;
+ sql_exp* sample_size_exp = exp_atom_lng(sql->sa,
sample_size);
+
+ symbol* weights = l->h->next->data.sym;
+ sql_exp* weights_exp = rel_value_exp(sql, &rel,
weights, 0, ek);
+
+ if (!sample_size_exp || !weights_exp)
+ return NULL;
+ append(exps, sample_size_exp);
+ append(exps, weights_exp);
+ } else {
+ // uniform sampling
+ // parse the sample size and pass it on to rel_sample
+ sql_exp *o = rel_value_exp( sql, &rel, sn->sample, 0,
ek);
+ if (!o)
+ return NULL;
+ append(exps, o);
+ }
rel = rel_sample(sql->sa, rel, exps);
}
diff --git a/sql/server/sql_parser.h b/sql/server/sql_parser.h
--- a/sql/server/sql_parser.h
+++ b/sql/server/sql_parser.h
@@ -172,7 +172,8 @@ typedef enum tokens {
SQL_XMLQUERY,
SQL_XMLTEXT,
SQL_XMLVALIDATE,
- SQL_XMLNAMESPACES
+ SQL_XMLNAMESPACES,
+ SQL_WEIGHTED_SAMPLE
} tokens;
typedef enum jt {
diff --git a/sql/server/sql_parser.y b/sql/server/sql_parser.y
--- a/sql/server/sql_parser.y
+++ b/sql/server/sql_parser.y
@@ -602,7 +602,7 @@ SQLCODE SQLERROR UNDER WHENEVER
%token ALTER ADD TABLE COLUMN TO UNIQUE VALUES VIEW WHERE WITH
%token<sval> sqlDATE TIME TIMESTAMP INTERVAL
%token YEAR MONTH DAY HOUR MINUTE SECOND ZONE
-%token LIMIT OFFSET SAMPLE
+%token LIMIT OFFSET SAMPLE WEIGHTS
%token CASE WHEN THEN ELSE NULLIF COALESCE IF ELSEIF WHILE DO
%token ATOMIC BEGIN END
@@ -3333,6 +3333,12 @@ opt_sample:
$$ = _newAtomNode( atom_float(SA, t,
strtod($2,NULL)));
}
| SAMPLE param { $$ = $2; }
+ | SAMPLE poslng WITH WEIGHTS search_condition {
+ dlist *l = L();
+ append_lng(l, $2);
+ append_symbol(l, $5);
+ $$ = _symbol_create_list(SQL_WEIGHTED_SAMPLE, l);
+ }
;
sort_specification_list:
diff --git a/sql/server/sql_scan.c b/sql/server/sql_scan.c
--- a/sql/server/sql_scan.c
+++ b/sql/server/sql_scan.c
@@ -208,6 +208,7 @@ scanner_init_keywords(void)
keywords_insert("LIKE", LIKE);
keywords_insert("LIMIT", LIMIT);
keywords_insert("SAMPLE", SAMPLE);
+ keywords_insert("WEIGHTS", WEIGHTS);
keywords_insert("LOCAL", LOCAL);
keywords_insert("LOCKED", LOCKED);
keywords_insert("NATURAL", NATURAL);
_______________________________________________
checkin-list mailing list
[email protected]
https://www.monetdb.org/mailman/listinfo/checkin-list
