void
GaussJordan::find_pivot (const int k)
{
if (iszero (get_entry (k, k)))
swap_rows (k, find_nonzero (k));
}
/**
* Having finished scanning a putatively tabular file, apply heuristics
* to the column statistics to finally determine the statistical class
* for each column (quantitative, categorical or (maybe) ordinal).
*
* The determination depends primarily on the presence or lack of unanimity
* in the observed types (integer, float, string) in the column, then
* devolves to consideration of the many other statistics in the column
* struct.
*/
void analyze_column( struct column *c ) {
const int OBSERVED_TYPE_COUNT
= count_nonzero( c->type_vote+1 /* exclude FTY_EMPTY */, FTY_COUNT-1 );
c->stat_class = STC_UNK; // ...unless overridden below!
if( OBSERVED_TYPE_COUNT == 0 /* all must be empty */ ) {
assert( c->type_vote[ FTY_EMPTY ] > 0 );
} else
if( OBSERVED_TYPE_COUNT == 1 /* Unanimity... */ ) {
// ...doesn't necessarily determine the stat class because...
switch( find_nonzero( c->type_vote, FTY_COUNT ) ) {
case FTY_INTEGER: // ...integers can be used in many ways!
c->stat_class = _integer_inference( c );
break;
case FTY_STRING:
if( ( ! c->excess_values )
&& set_count( & c->value_set ) < c->type_vote[ FTY_STRING ]
&& c->long_field_count == 0 )
c->stat_class = STC_CAT;
break;
case FTY_FLOAT:
c->stat_class = STC_QUA;
}
} else {
/**
* If more than two types are observed and STRING is one of them,
* then everything hinges on the cardinality of observed strings
* (the contents of value_set EXCLUDING any integers it contains).
* 1. If exactly one string value is observed AND it's a potential
* missing data indicator, then inference devolves to that for
* numeric types.
* 2. If more than one string value is observed all bets are off;
* the column remains STC_UNKnown.
*/
const char *sval[2];
const bool UNIQUE_STRING
= c->type_vote[ FTY_STRING ] > 0
&& _fetch_string_values( & c->value_set, sval, 2 ) == 1;
const bool HAS_CANDIDATE_MISSING_DATA_PLACEHOLDER
= UNIQUE_STRING
&& regexec( &_compiled_re_NA, sval[0], 0, NULL, 0 ) == 0;
if( OBSERVED_TYPE_COUNT == 2 ) {
if( c->type_vote[ FTY_STRING ] > 0 ) {
if( HAS_CANDIDATE_MISSING_DATA_PLACEHOLDER ) {
if( c->type_vote[ FTY_INTEGER ] > 0 ) {
c->stat_class = _integer_inference( c );
} else {
assert( c->type_vote[ FTY_FLOAT ] > 0 );
c->stat_class = STC_QUA;
}
}
} else { // no string, just ints and floats
assert( c->type_vote[ FTY_INTEGER ] > 0 &&
c->type_vote[ FTY_FLOAT ] > 0 );
c->stat_class = STC_QUA;
}
} else { // Column contains int(s), float(s) AND string(s).
if( HAS_CANDIDATE_MISSING_DATA_PLACEHOLDER )
c->stat_class = STC_QUA;
} // 3 types observed
} // > 1 type observed.
}
