Examples with ValFrame water.rapids.vals.ValFrame used on opensource projects

Example 1 with ValFrame

use of water.rapids.vals.ValFrame in project h2o-3 by h2oai.

the class AstCut method apply.

@Override
public ValFrame apply(Env env, Env.StackHelp stk, AstRoot[] asts) {
    Frame fr = stk.track(asts[1].exec(env)).getFrame();
    double[] cuts = check(asts[2]);
    Arrays.sort(cuts);
    String[] labels = check2(asts[3]);
    final boolean lowest = asts[4].exec(env).getNum() == 1;
    final boolean rite = asts[5].exec(env).getNum() == 1;
    // cap at 12
    final int digits = Math.min((int) asts[6].exec(env).getNum(), 12);
    if (fr.vecs().length != 1 || fr.vecs()[0].isCategorical())
        throw new IllegalArgumentException("First argument must be a numeric column vector");
    double fmin = fr.anyVec().min();
    double fmax = fr.anyVec().max();
    // c(0,10,100) -> 2 bins (0,10] U (10, 100]
    int nbins = cuts.length - 1;
    double width;
    if (nbins == 0) {
        if (cuts[0] < 2)
            throw new IllegalArgumentException("The number of cuts must be >= 2. Got: " + cuts[0]);
        // in this case, cut the vec into _cuts[0] many pieces of equal length
        nbins = (int) Math.floor(cuts[0]);
        width = (fmax - fmin) / nbins;
        cuts = new double[nbins];
        cuts[0] = fmin - 0.001 * (fmax - fmin);
        for (int i = 1; i < cuts.length; ++i) cuts[i] = (i == cuts.length - 1) ? (fmax + 0.001 * (fmax - fmin)) : (fmin + i * width);
    }
    // if(width == 0) throw new IllegalArgumentException("Data vector is constant!");
    if (labels != null && labels.length != nbins)
        throw new IllegalArgumentException("`labels` vector does not match the number of cuts.");
    // Construct domain names from _labels or bin intervals if _labels is null
    final double[] cutz = cuts;
    // first round _cuts to dig.lab decimals: example floor(2.676*100 + 0.5) / 100
    for (int i = 0; i < cuts.length; ++i) cuts[i] = Math.floor(cuts[i] * Math.pow(10, digits) + 0.5) / Math.pow(10, digits);
    String[][] domains = new String[1][nbins];
    if (labels == null) {
        domains[0][0] = (lowest ? "[" : left(rite)) + cuts[0] + "," + cuts[1] + rite(rite);
        for (int i = 1; i < (cuts.length - 1); ++i) domains[0][i] = left(rite) + cuts[i] + "," + cuts[i + 1] + rite(rite);
    } else
        domains[0] = labels;
    Frame fr2 = new MRTask() {

        @Override
        public void map(Chunk c, NewChunk nc) {
            int rows = c._len;
            for (int r = 0; r < rows; ++r) {
                double x = c.atd(r);
                if (Double.isNaN(x) || (lowest && x < cutz[0]) || (!lowest && (x < cutz[0] || MathUtils.equalsWithinOneSmallUlp(x, cutz[0]))) || (rite && x > cutz[cutz.length - 1]) || (!rite && (x > cutz[cutz.length - 1] || MathUtils.equalsWithinOneSmallUlp(x, cutz[cutz.length - 1]))))
                    nc.addNum(Double.NaN);
                else {
                    for (int i = 1; i < cutz.length; ++i) {
                        if (rite) {
                            if (x <= cutz[i]) {
                                nc.addNum(i - 1);
                                break;
                            }
                        } else if (x < cutz[i]) {
                            nc.addNum(i - 1);
                            break;
                        }
                    }
                }
            }
        }
    }.doAll(1, Vec.T_NUM, fr).outputFrame(fr.names(), domains);
    return new ValFrame(fr2);
}

Example 2 with ValFrame

use of water.rapids.vals.ValFrame in project h2o-3 by h2oai.

the class AstDdply 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 = AstGroup.check(ncols, asts[2]);
    int[] gbCols = groupby.expand4();
    AstRoot fun = asts[3].exec(env).getFun();
    // Current execution scope; needed to lookup variables
    AstFunction scope = env._scope;
    // Pass 1: Find all the groups (and count rows-per-group)
    IcedHashMap<AstGroup.G, String> gss = AstGroup.doGroups(fr, gbCols, AstGroup.aggNRows());
    final AstGroup.G[] grps = gss.keySet().toArray(new AstGroup.G[gss.size()]);
    // apply an ORDER by here...
    final int[] ordCols = new AstNumList(0, gbCols.length).expand4();
    Arrays.sort(grps, new java.util.Comparator<AstGroup.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
        // _orderByCols.  NaN is treated as least
        @Override
        public int compare(AstGroup.G g1, AstGroup.G g2) {
            for (int i : ordCols) {
                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();
        }
    });
    // Uniquely number the groups
    for (int gnum = 0; gnum < grps.length; gnum++) grps[gnum]._dss[0][0] = gnum;
    // Pass 2: Build all the groups, building 1 Vec per-group, with exactly the
    // same Chunk layout, except each Chunk will be the filter rows numbers; a
    // list of the Chunk-relative row-numbers for that group in an original
    // data Chunk.  Each Vec will have a *different* number of rows.
    Vec[] vgrps = new BuildGroup(gbCols, gss).doAll(gss.size(), Vec.T_NUM, fr).close();
    // Pass 3: For each group, build a full frame for the group, run the
    // function on it and tear the frame down.
    // gather up the remote tasks...
    final RemoteRapids[] remoteTasks = new RemoteRapids[gss.size()];
    Futures fs = new Futures();
    for (int i = 0; i < remoteTasks.length; i++) fs.add(RPC.call(vgrps[i]._key.home_node(), remoteTasks[i] = new RemoteRapids(fr, vgrps[i]._key, fun, scope)));
    fs.blockForPending();
    // Build the output!
    final double[] res0 = remoteTasks[0]._result;
    String[] fcnames = new String[res0.length];
    for (int i = 0; i < res0.length; i++) fcnames[i] = "ddply_C" + (i + 1);
    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
                AstGroup.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]);
                double[] res = remoteTasks[i + start]._result;
                for (int a = 0; a < res0.length; a++) ncs[j++].addNum(res[a]);
            }
        }
    };
    Frame f = AstGroup.buildOutput(gbCols, res0.length, fr, fcnames, gss.size(), mrfill);
    return new ValFrame(f);
}

Example 3 with ValFrame

use of water.rapids.vals.ValFrame in project h2o-3 by h2oai.

the class AstFlatten method apply.

@Override
public Val apply(Env env, Env.StackHelp stk, AstRoot[] asts) {
    Frame fr = stk.track(asts[1].exec(env)).getFrame();
    // did not flatten
    if (fr.numCols() != 1 || fr.numRows() != 1)
        return new ValFrame(fr);
    Vec vec = fr.anyVec();
    switch(vec.get_type()) {
        case Vec.T_BAD:
        case Vec.T_NUM:
            return new ValNum(vec.at(0));
        case Vec.T_TIME:
            // check for missing values
            return vec.isNA(0) ? new ValNum(Double.NaN) : new ValNum(vec.at8(0));
        case Vec.T_STR:
            return new ValStr(vec.atStr(new BufferedString(), 0).toString());
        case // check for missing values
        Vec.T_CAT:
            return vec.isNA(0) ? new ValStr("NA") : new ValStr(vec.factor(vec.at8(0)));
        default:
            throw H2O.unimpl("The type of vector: " + vec.get_type_str() + " is not supported by " + str());
    }
}

Example 4 with ValFrame

use of water.rapids.vals.ValFrame 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 5 with ValFrame

use of water.rapids.vals.ValFrame in project h2o-3 by h2oai.

the class AstGroupedPermute method apply.

@Override
public ValFrame apply(Env env, Env.StackHelp stk, AstRoot[] asts) {
    Frame fr = stk.track(asts[1].exec(env)).getFrame();
    final int permCol = (int) asts[2].exec(env).getNum();
    AstNumList groupby = AstGroup.check(fr.numCols(), asts[3]);
    final int[] gbCols = groupby.expand4();
    final int permuteBy = (int) asts[4].exec(env).getNum();
    final int keepCol = (int) asts[5].exec(env).getNum();
    String[] names = new String[gbCols.length + 4];
    int i = 0;
    for (; i < gbCols.length; ++i) names[i] = fr.name(gbCols[i]);
    names[i++] = "In";
    names[i++] = "Out";
    names[i++] = "InAmnt";
    names[i] = "OutAmnt";
    String[][] domains = new String[names.length][];
    int d = 0;
    for (; d < gbCols.length; d++) domains[d] = fr.domains()[gbCols[d]];
    domains[d++] = fr.domains()[permCol];
    domains[d++] = fr.domains()[permCol];
    domains[d++] = fr.domains()[keepCol];
    domains[d] = fr.domains()[keepCol];
    long s = System.currentTimeMillis();
    BuildGroups t = new BuildGroups(gbCols, permuteBy, permCol, keepCol).doAll(fr);
    Log.info("Elapsed time: " + (System.currentTimeMillis() - s) / 1000. + "s");
    s = System.currentTimeMillis();
    SmashGroups sg;
    H2O.submitTask(sg = new SmashGroups(t._grps)).join();
    Log.info("Elapsed time: " + (System.currentTimeMillis() - s) / 1000. + "s");
    return new ValFrame(buildOutput(sg._res.values().toArray(new double[0][][]), names, domains));
}