use of water.Key in project h2o-3 by h2oai.
the class PersistManager method anyURIToKey.
/** Convert given URI into a specific H2O key representation.
*
* The representation depends on persistent backend, since it will
* deduce file location from the key content.
*
* The method will look at scheme of URI and based on it, it will
* ask a backend to provide a conversion to a key (i.e., URI with scheme
* 'hdfs' will be forwared to HDFS backend).
*
* @param uri file location
* @return a key encoding URI
* @throws IOException in the case of uri conversion problem
* @throws water.exceptions.H2OIllegalArgumentException in case of unsupported scheme
*/
public final Key anyURIToKey(URI uri) throws IOException {
Key ikey = null;
String scheme = uri.getScheme();
if ("s3".equals(scheme)) {
ikey = I[Value.S3].uriToKey(uri);
} else if ("hdfs".equals(scheme)) {
ikey = I[Value.HDFS].uriToKey(uri);
} else if ("s3".equals(scheme) || "s3n".equals(scheme) || "s3a".equals(scheme)) {
ikey = I[Value.HDFS].uriToKey(uri);
} else if ("file".equals(scheme) || scheme == null) {
ikey = I[Value.NFS].uriToKey(uri);
} else if (useHdfsAsFallback() && I[Value.HDFS].canHandle(uri.toString())) {
ikey = I[Value.HDFS].uriToKey(uri);
} else {
throw new H2OIllegalArgumentException("Unsupported schema '" + scheme + "' for given uri " + uri);
}
return ikey;
}
use of water.Key in project h2o-3 by h2oai.
the class AstVariance method array.
// Matrix covariance. Compute covariance between all columns from each Frame
// against each other. Return a matrix of covariances which is frx.numCols
// wide and fry.numCols tall.
private Val array(Frame frx, Frame fry, Mode mode, boolean symmetric) {
Vec[] vecxs = frx.vecs();
int ncolx = vecxs.length;
Vec[] vecys = fry.vecs();
int ncoly = vecys.length;
if (mode.equals(Mode.Everything) || mode.equals(Mode.AllObs)) {
if (mode.equals(Mode.AllObs)) {
for (Vec v : vecxs) if (v.naCnt() != 0)
throw new IllegalArgumentException("Mode is 'all.obs' but NAs are present");
if (!symmetric)
for (Vec v : vecys) if (v.naCnt() != 0)
throw new IllegalArgumentException("Mode is 'all.obs' but NAs are present");
}
CoVarTaskEverything[] cvs = new CoVarTaskEverything[ncoly];
double[] xmeans = new double[ncolx];
for (int x = 0; x < ncoly; x++) xmeans[x] = vecxs[x].mean();
if (symmetric) {
//1-col returns scalar
if (ncoly == 1)
return new ValNum(vecys[0].naCnt() == 0 ? vecys[0].sigma() * vecys[0].sigma() : Double.NaN);
int[] idx = new int[ncoly];
for (int y = 1; y < ncoly; y++) idx[y] = y;
int[] first_index = new int[] { 0 };
//compute covariances between column_i and column_i+1, column_i+2, ...
Frame reduced_fr;
for (int y = 0; y < ncoly - 1; y++) {
idx = ArrayUtils.removeIds(idx, first_index);
reduced_fr = new Frame(frx.vecs(idx));
cvs[y] = new CoVarTaskEverything(vecys[y].mean(), xmeans).dfork(new Frame(vecys[y]).add(reduced_fr));
}
double[][] res_array = new double[ncoly][ncoly];
//fill in the diagonals (variances) using sigma from rollupstats
for (int y = 0; y < ncoly; y++) res_array[y][y] = vecys[y].naCnt() == 0 ? vecys[y].sigma() * vecys[y].sigma() : Double.NaN;
//arrange the results into the bottom left of res_array. each successive cvs is 1 smaller in length
for (int y = 0; y < ncoly - 1; y++) System.arraycopy(ArrayUtils.div(cvs[y].getResult()._covs, (fry.numRows() - 1)), 0, res_array[y], y + 1, ncoly - y - 1);
//copy over the bottom left of res_array to its top right
for (int y = 0; y < ncoly - 1; y++) {
for (int x = y + 1; x < ncoly; x++) {
res_array[x][y] = res_array[y][x];
}
}
//set Frame
Vec[] res = new Vec[ncoly];
Key<Vec>[] keys = Vec.VectorGroup.VG_LEN1.addVecs(ncoly);
for (int y = 0; y < ncoly; y++) {
res[y] = Vec.makeVec(res_array[y], keys[y]);
}
return new ValFrame(new Frame(fry._names, res));
}
// Launch tasks; each does all Xs vs one Y
for (int y = 0; y < ncoly; y++) cvs[y] = new CoVarTaskEverything(vecys[y].mean(), xmeans).dfork(new Frame(vecys[y]).add(frx));
// 1-col returns scalar
if (ncolx == 1 && ncoly == 1) {
return new ValNum(cvs[0].getResult()._covs[0] / (fry.numRows() - 1));
}
// Gather all the Xs-vs-Y covariance arrays; divide by rows
Vec[] res = new Vec[ncoly];
Key<Vec>[] keys = Vec.VectorGroup.VG_LEN1.addVecs(ncoly);
for (int y = 0; y < ncoly; y++) res[y] = Vec.makeVec(ArrayUtils.div(cvs[y].getResult()._covs, (fry.numRows() - 1)), keys[y]);
return new ValFrame(new Frame(fry._names, res));
} else {
if (symmetric) {
if (ncoly == 1)
return new ValNum(vecys[0].sigma() * vecys[0].sigma());
CoVarTaskCompleteObsMeanSym taskCompleteObsMeanSym = new CoVarTaskCompleteObsMeanSym().doAll(fry);
long NACount = taskCompleteObsMeanSym._NACount;
double[] ymeans = ArrayUtils.div(taskCompleteObsMeanSym._ysum, fry.numRows() - NACount);
// 1 task with all Ys
CoVarTaskCompleteObsSym cvs = new CoVarTaskCompleteObsSym(ymeans).doAll(new Frame(fry));
double[][] res_array = new double[ncoly][ncoly];
for (int y = 0; y < ncoly; y++) {
System.arraycopy(ArrayUtils.div(cvs._covs[y], (fry.numRows() - 1 - NACount)), y, res_array[y], y, ncoly - y);
}
//copy over the bottom left of res_array to its top right
for (int y = 0; y < ncoly - 1; y++) {
for (int x = y + 1; x < ncoly; x++) {
res_array[x][y] = res_array[y][x];
}
}
//set Frame
Vec[] res = new Vec[ncoly];
Key<Vec>[] keys = Vec.VectorGroup.VG_LEN1.addVecs(ncoly);
for (int y = 0; y < ncoly; y++) {
res[y] = Vec.makeVec(res_array[y], keys[y]);
}
return new ValFrame(new Frame(fry._names, res));
}
CoVarTaskCompleteObsMean taskCompleteObsMean = new CoVarTaskCompleteObsMean(ncoly, ncolx).doAll(new Frame(fry).add(frx));
long NACount = taskCompleteObsMean._NACount;
double[] ymeans = ArrayUtils.div(taskCompleteObsMean._ysum, fry.numRows() - NACount);
double[] xmeans = ArrayUtils.div(taskCompleteObsMean._xsum, fry.numRows() - NACount);
// 1 task with all Xs and Ys
CoVarTaskCompleteObs cvs = new CoVarTaskCompleteObs(ymeans, xmeans).doAll(new Frame(fry).add(frx));
// 1-col returns scalar
if (ncolx == 1 && ncoly == 1) {
return new ValNum(cvs._covs[0][0] / (fry.numRows() - 1 - NACount));
}
// Gather all the Xs-vs-Y covariance arrays; divide by rows
Vec[] res = new Vec[ncoly];
Key<Vec>[] keys = Vec.VectorGroup.VG_LEN1.addVecs(ncoly);
for (int y = 0; y < ncoly; y++) res[y] = Vec.makeVec(ArrayUtils.div(cvs._covs[y], (fry.numRows() - 1 - NACount)), keys[y]);
return new ValFrame(new Frame(fry._names, res));
}
}
use of water.Key in project h2o-3 by h2oai.
the class AstRm method apply.
@Override
public ValNum apply(Env env, Env.StackHelp stk, AstRoot[] asts) {
Key id = Key.make(env.expand(asts[1].str()));
Value val = DKV.get(id);
if (val == null)
return new ValNum(0);
if (val.isFrame())
// Remove unshared Vecs
env._ses.remove(val.<Frame>get());
else
// Normal (e.g. Model) remove
Keyed.remove(id);
return new ValNum(1);
}
use of water.Key in project h2o-2 by h2oai.
the class Parse2 method serve.
@Override
protected Response serve() {
PSetup p = _source.value();
CustomParser.ParserSetup setup = p != null ? p._setup._setup : new CustomParser.ParserSetup();
setup._singleQuotes = _sQuotes.value();
destination_key = Key.make(_dest.value());
try {
// Make a new Setup, with the 'header' flag set according to user wishes.
Key[] keys = p._keys.toArray(new Key[p._keys.size()]);
Job parseJob = ParseDataset2.forkParseDataset(destination_key, keys, setup, delete_on_done.value());
job_key = parseJob.self();
// Allow the user to specify whether to block synchronously for a response or not.
if (_blocking.value()) {
// block until the end of job
parseJob.get();
assert Job.isEnded(job_key) : "Job is still running but we already passed over its end. Job = " + job_key;
}
return Progress2.redirect(this, job_key, destination_key);
} catch (Throwable e) {
return Response.error(e);
}
}
use of water.Key in project h2o-2 by h2oai.
the class PutValue method serve.
@Override
public Response serve() {
JsonObject response = new JsonObject();
Key k = Key.make(_key.value()._kb, (byte) (int) _rf.value());
Value v = new Value(k, _value.value().getBytes());
UKV.put(k, v);
response.addProperty(KEY, k.toString());
response.addProperty(REPLICATION_FACTOR, k.desired());
response.addProperty(VALUE_SIZE, v._max);
return Response.done(response);
}
Aggregations