Example 1 with ValFun
use of water.rapids.vals.ValFun in project h2o-3 by h2oai.
the class AstGroup method apply.
@Override
public ValFrame apply(Env env, Env.StackHelp stk, AstRoot[] asts) {
Frame fr = stk.track(asts[1].exec(env)).getFrame();
int ncols = fr.numCols();
AstNumList groupby = check(ncols, asts[2]);
final int[] gbCols = groupby.expand4();
// Count of aggregates; knock off the first 4 ASTs (GB data [group-by] [order-by]...),
// then count by triples.
int naggs = (asts.length - 3) / 3;
final AGG[] aggs = new AGG[naggs];
for (int idx = 3; idx < asts.length; idx += 3) {
Val v = asts[idx].exec(env);
String fn = v instanceof ValFun ? v.getFun().str() : v.getStr();
FCN fcn = FCN.valueOf(fn);
AstNumList col = check(ncols, asts[idx + 1]);
if (col.cnt() != 1)
throw new IllegalArgumentException("Group-By functions take only a single column");
// Aggregate column
int agg_col = (int) col.min();
if (fcn == FCN.mode && !fr.vec(agg_col).isCategorical())
throw new IllegalArgumentException("Mode only allowed on categorical columns");
NAHandling na = NAHandling.valueOf(asts[idx + 2].exec(env).getStr().toUpperCase());
aggs[(idx - 3) / 3] = new AGG(fcn, agg_col, na, (int) fr.vec(agg_col).max() + 1);
}
// do the group by work now
IcedHashMap<G, String> gss = doGroups(fr, gbCols, aggs);
final G[] grps = gss.keySet().toArray(new G[gss.size()]);
// apply an ORDER by here...
if (gbCols.length > 0)
Arrays.sort(grps, new java.util.Comparator<G>() {
// Compare 2 groups. Iterate down _gs, stop when _gs[i] > that._gs[i],
// or _gs[i] < that._gs[i]. Order by various columns specified by
// gbCols. NaN is treated as least
@Override
public int compare(G g1, G g2) {
for (int i = 0; i < gbCols.length; i++) {
if (Double.isNaN(g1._gs[i]) && !Double.isNaN(g2._gs[i]))
return -1;
if (!Double.isNaN(g1._gs[i]) && Double.isNaN(g2._gs[i]))
return 1;
if (g1._gs[i] != g2._gs[i])
return g1._gs[i] < g2._gs[i] ? -1 : 1;
}
return 0;
}
// I do not believe sort() calls equals() at this time, so no need to implement
@Override
public boolean equals(Object o) {
throw H2O.unimpl();
}
});
// Build the output!
String[] fcnames = new String[aggs.length];
for (int i = 0; i < aggs.length; i++) fcnames[i] = aggs[i]._fcn.toString() + "_" + fr.name(aggs[i]._col);
MRTask mrfill = new MRTask() {
@Override
public void map(Chunk[] c, NewChunk[] ncs) {
int start = (int) c[0].start();
for (int i = 0; i < c[0]._len; ++i) {
// One Group per row
G g = grps[i + start];
int j;
for (// The Group Key, as a row
j = 0; // The Group Key, as a row
j < g._gs.length; // The Group Key, as a row
j++) ncs[j].addNum(g._gs[j]);
for (int a = 0; a < aggs.length; a++) ncs[j++].addNum(aggs[a]._fcn.postPass(g._dss[a], g._ns[a]));
}
}
};
Frame f = buildOutput(gbCols, naggs, fr, fcnames, grps.length, mrfill);
return new ValFrame(f);
}Example 2 with ValFun
use of water.rapids.vals.ValFun in project h2o-3 by h2oai.
the class Env method lookup.
// ----
// Variable lookup
public Val lookup(String id) {
// Lexically scoped functions first
Val val = _scope == null ? null : _scope.lookup(id);
if (val != null)
return val;
// disallow TRUE/FALSE/NA to be overwritten by keys in the DKV... just way way saner this way
if (CONSTS.containsKey(id)) {
return CONSTS.get(id).exec(this);
}
// Now the DKV
Value value = DKV.get(Key.make(expand(id)));
if (value != null) {
if (value.isFrame())
return addGlobals((Frame) value.get());
// Only understand Frames right now
throw new IllegalArgumentException("DKV name lookup of " + id + " yielded an instance of type " + value.className() + ", but only Frame is supported");
}
// Now the built-ins
AstPrimitive ast = PRIMS.get(id);
if (ast != null)
return new ValFun(ast);
throw new IllegalArgumentException("Name lookup of '" + id + "' failed");
}