Search in sources :

Example 41 with MatrixContext

use of com.tencent.angel.ml.matrix.MatrixContext in project angel by Tencent.

the class UpdateRowsTest method setup.

@Before
public void setup() throws Exception {
    // set basic configuration keys
    Configuration conf = new Configuration();
    conf.setBoolean("mapred.mapper.new-api", true);
    conf.setBoolean(AngelConf.ANGEL_JOB_OUTPUT_PATH_DELETEONEXIST, true);
    conf.set(AngelConf.ANGEL_TASK_USER_TASKCLASS, DummyTask.class.getName());
    // use local deploy mode and dummy dataspliter
    conf.set(AngelConf.ANGEL_DEPLOY_MODE, "LOCAL");
    conf.setBoolean(AngelConf.ANGEL_AM_USE_DUMMY_DATASPLITER, true);
    conf.set(AngelConf.ANGEL_INPUTFORMAT_CLASS, CombineTextInputFormat.class.getName());
    conf.set(AngelConf.ANGEL_SAVE_MODEL_PATH, LOCAL_FS + TMP_PATH + "/out");
    conf.set(AngelConf.ANGEL_TRAIN_DATA_PATH, LOCAL_FS + TMP_PATH + "/in");
    conf.set(AngelConf.ANGEL_LOG_PATH, LOCAL_FS + TMP_PATH + "/log");
    conf.setInt(AngelConf.ANGEL_WORKERGROUP_NUMBER, 1);
    conf.setInt(AngelConf.ANGEL_PS_NUMBER, 2);
    conf.setInt(AngelConf.ANGEL_WORKER_TASK_NUMBER, 1);
    conf.setInt(AngelConf.ANGEL_MODEL_PARTITIONER_PARTITION_SIZE, 1000);
    conf.setBoolean("use.new.split", false);
    conf.setInt(AngelConf.ANGEL_PSAGENT_CACHE_SYNC_TIMEINTERVAL_MS, 10);
    conf.setInt(AngelConf.ANGEL_WORKER_HEARTBEAT_INTERVAL_MS, 1000);
    conf.setInt(AngelConf.ANGEL_PS_HEARTBEAT_INTERVAL_MS, 1000);
    conf.setInt(AngelConf.ANGEL_WORKER_MAX_ATTEMPTS, 1);
    conf.setInt(AngelConf.ANGEL_PS_MAX_ATTEMPTS, 1);
    // get a angel client
    angelClient = AngelClientFactory.get(conf);
    // add dense double matrix
    MatrixContext dMat = new MatrixContext();
    dMat.setName(DENSE_DOUBLE_MAT);
    dMat.setRowNum(rowNum);
    dMat.setColNum(feaNum);
    dMat.setMaxRowNumInBlock(blockRowNum);
    dMat.setMaxColNumInBlock(blockColNum);
    dMat.setRowType(RowType.T_DOUBLE_DENSE);
    dMat.setPartitionerClass(ColumnRangePartitioner.class);
    angelClient.addMatrix(dMat);
    // add sparse double matrix
    MatrixContext sMat = new MatrixContext();
    sMat.setName(SPARSE_DOUBLE_MAT);
    sMat.setRowNum(rowNum);
    sMat.setColNum(feaNum);
    sMat.setMaxRowNumInBlock(blockRowNum);
    sMat.setMaxColNumInBlock(blockColNum);
    sMat.setRowType(RowType.T_DOUBLE_SPARSE);
    angelClient.addMatrix(sMat);
    // add dense float matrix
    MatrixContext dfMat = new MatrixContext();
    dfMat.setName(DENSE_FLOAT_MAT);
    dfMat.setRowNum(rowNum);
    dfMat.setColNum(feaNum);
    dfMat.setMaxRowNumInBlock(blockRowNum);
    dfMat.setMaxColNumInBlock(blockColNum);
    dfMat.setRowType(RowType.T_FLOAT_DENSE);
    dfMat.setPartitionerClass(ColumnRangePartitioner.class);
    angelClient.addMatrix(dfMat);
    // add sparse float matrix
    MatrixContext sfMat = new MatrixContext();
    sfMat.setName(SPARSE_FLOAT_MAT);
    sfMat.setRowNum(rowNum);
    sfMat.setColNum(feaNum);
    sfMat.setMaxRowNumInBlock(blockRowNum);
    sfMat.setMaxColNumInBlock(blockColNum);
    sfMat.setRowType(RowType.T_FLOAT_SPARSE);
    angelClient.addMatrix(sfMat);
    // add dense float matrix
    MatrixContext diMat = new MatrixContext();
    diMat.setName(DENSE_INT_MAT);
    diMat.setRowNum(rowNum);
    diMat.setColNum(feaNum);
    diMat.setMaxRowNumInBlock(blockRowNum);
    diMat.setMaxColNumInBlock(blockColNum);
    diMat.setRowType(RowType.T_INT_DENSE);
    diMat.setPartitionerClass(ColumnRangePartitioner.class);
    angelClient.addMatrix(diMat);
    // add sparse float matrix
    MatrixContext siMat = new MatrixContext();
    siMat.setName(SPARSE_INT_MAT);
    siMat.setRowNum(rowNum);
    siMat.setColNum(feaNum);
    siMat.setMaxRowNumInBlock(blockRowNum);
    siMat.setMaxColNumInBlock(blockColNum);
    siMat.setRowType(RowType.T_INT_SPARSE);
    angelClient.addMatrix(siMat);
    // add dense long matrix
    MatrixContext dlMat = new MatrixContext();
    dlMat.setName(DENSE_LONG_MAT);
    dlMat.setRowNum(rowNum);
    dlMat.setColNum(feaNum);
    dlMat.setMaxRowNumInBlock(blockRowNum);
    dlMat.setMaxColNumInBlock(blockColNum);
    dlMat.setRowType(RowType.T_LONG_DENSE);
    dlMat.setPartitionerClass(ColumnRangePartitioner.class);
    angelClient.addMatrix(dlMat);
    // add sparse long matrix
    MatrixContext slMat = new MatrixContext();
    slMat.setName(SPARSE_LONG_MAT);
    slMat.setRowNum(rowNum);
    slMat.setColNum(feaNum);
    slMat.setMaxRowNumInBlock(blockRowNum);
    slMat.setMaxColNumInBlock(blockColNum);
    slMat.setRowType(RowType.T_LONG_SPARSE);
    angelClient.addMatrix(slMat);
    // add sparse long-key double matrix
    MatrixContext dLongKeysMatrix = new MatrixContext();
    dLongKeysMatrix.setName(SPARSE_DOUBLE_LONG_MAT);
    dLongKeysMatrix.setRowNum(rowNum);
    dLongKeysMatrix.setColNum(feaNum);
    dLongKeysMatrix.setMaxRowNumInBlock(blockRowNum);
    dLongKeysMatrix.setMaxColNumInBlock(blockColNum);
    dLongKeysMatrix.setRowType(RowType.T_DOUBLE_SPARSE_LONGKEY);
    angelClient.addMatrix(dLongKeysMatrix);
    // add sparse long-key float matrix
    MatrixContext slfMatrix = new MatrixContext();
    slfMatrix.setName(SPARSE_FLOAT_LONG_MAT);
    slfMatrix.setRowNum(rowNum);
    slfMatrix.setColNum(feaNum);
    slfMatrix.setMaxRowNumInBlock(blockRowNum);
    slfMatrix.setMaxColNumInBlock(blockColNum);
    slfMatrix.setRowType(RowType.T_FLOAT_SPARSE_LONGKEY);
    angelClient.addMatrix(slfMatrix);
    // add sparse long-key int matrix
    MatrixContext sliMatrix = new MatrixContext();
    sliMatrix.setName(SPARSE_INT_LONG_MAT);
    sliMatrix.setRowNum(rowNum);
    sliMatrix.setColNum(feaNum);
    sliMatrix.setMaxRowNumInBlock(blockRowNum);
    sliMatrix.setMaxColNumInBlock(blockColNum);
    sliMatrix.setRowType(RowType.T_INT_SPARSE_LONGKEY);
    angelClient.addMatrix(sliMatrix);
    // add sparse long-key long matrix
    MatrixContext sllMatrix = new MatrixContext();
    sllMatrix.setName(SPARSE_LONG_LONG_MAT);
    sllMatrix.setRowNum(rowNum);
    sllMatrix.setColNum(feaNum);
    sllMatrix.setMaxRowNumInBlock(blockRowNum);
    sllMatrix.setMaxColNumInBlock(blockColNum);
    sllMatrix.setRowType(RowType.T_LONG_SPARSE_LONGKEY);
    angelClient.addMatrix(sllMatrix);
    // Start PS
    angelClient.startPSServer();
    // Start to run application
    angelClient.run();
    Thread.sleep(5000);
    psId = new ParameterServerId(0);
    psAttempt0Id = new PSAttemptId(psId, 0);
    WorkerGroupId workerGroupId = new WorkerGroupId(0);
    workerId = new WorkerId(workerGroupId, 0);
    workerAttempt0Id = new WorkerAttemptId(workerId, 0);
}
Also used : CombineTextInputFormat(org.apache.hadoop.mapreduce.lib.input.CombineTextInputFormat) MatrixContext(com.tencent.angel.ml.matrix.MatrixContext) Configuration(org.apache.hadoop.conf.Configuration) PSAttemptId(com.tencent.angel.ps.PSAttemptId) WorkerAttemptId(com.tencent.angel.worker.WorkerAttemptId) ParameterServerId(com.tencent.angel.ps.ParameterServerId) WorkerId(com.tencent.angel.worker.WorkerId) WorkerGroupId(com.tencent.angel.worker.WorkerGroupId) Before(org.junit.Before)

Example 42 with MatrixContext

use of com.tencent.angel.ml.matrix.MatrixContext in project angel by Tencent.

the class GetRowTest method setup.

@Before
public void setup() throws Exception {
    // set basic configuration keys
    Configuration conf = new Configuration();
    conf.setBoolean("mapred.mapper.new-api", true);
    conf.setBoolean(AngelConf.ANGEL_JOB_OUTPUT_PATH_DELETEONEXIST, true);
    conf.set(AngelConf.ANGEL_TASK_USER_TASKCLASS, DummyTask.class.getName());
    // use local deploy mode and dummy dataspliter
    conf.set(AngelConf.ANGEL_DEPLOY_MODE, "LOCAL");
    conf.setBoolean(AngelConf.ANGEL_AM_USE_DUMMY_DATASPLITER, true);
    conf.set(AngelConf.ANGEL_INPUTFORMAT_CLASS, CombineTextInputFormat.class.getName());
    conf.set(AngelConf.ANGEL_SAVE_MODEL_PATH, LOCAL_FS + TMP_PATH + "/out");
    conf.set(AngelConf.ANGEL_TRAIN_DATA_PATH, LOCAL_FS + TMP_PATH + "/in");
    conf.set(AngelConf.ANGEL_LOG_PATH, LOCAL_FS + TMP_PATH + "/log");
    conf.setInt(AngelConf.ANGEL_WORKERGROUP_NUMBER, 1);
    conf.setInt(AngelConf.ANGEL_PS_NUMBER, 1);
    conf.setInt(AngelConf.ANGEL_WORKER_TASK_NUMBER, 1);
    conf.setInt(AngelConf.ANGEL_MODEL_PARTITIONER_PARTITION_SIZE, 1000);
    conf.setInt(AngelConf.ANGEL_PSAGENT_CACHE_SYNC_TIMEINTERVAL_MS, 10);
    conf.setInt(AngelConf.ANGEL_WORKER_HEARTBEAT_INTERVAL_MS, 1000);
    conf.setInt(AngelConf.ANGEL_PS_HEARTBEAT_INTERVAL_MS, 1000);
    conf.setInt(AngelConf.ANGEL_WORKER_MAX_ATTEMPTS, 1);
    conf.setInt(AngelConf.ANGEL_PS_MAX_ATTEMPTS, 1);
    // get a angel client
    angelClient = AngelClientFactory.get(conf);
    // add dense double matrix
    MatrixContext dMat = new MatrixContext();
    dMat.setName(DENSE_DOUBLE_MAT);
    dMat.setRowNum(1);
    dMat.setColNum(feaNum);
    dMat.setMaxColNumInBlock(colmunSize);
    dMat.setRowType(RowType.T_DOUBLE_DENSE);
    dMat.setPartitionerClass(ColumnRangePartitioner.class);
    angelClient.addMatrix(dMat);
    // add sparse double matrix
    MatrixContext sMat = new MatrixContext();
    sMat.setName(SPARSE_DOUBLE_MAT);
    sMat.setRowNum(1);
    sMat.setColNum(feaNum);
    sMat.setMaxColNumInBlock(colmunSize);
    sMat.setRowType(RowType.T_DOUBLE_SPARSE);
    sMat.setPartitionerClass(HashPartitioner.class);
    angelClient.addMatrix(sMat);
    // add dense float matrix
    MatrixContext dfMat = new MatrixContext();
    dfMat.setName(DENSE_FLOAT_MAT);
    dfMat.setRowNum(1);
    dfMat.setColNum(feaNum);
    dfMat.setMaxColNumInBlock(colmunSize);
    dfMat.setRowType(RowType.T_FLOAT_DENSE);
    dfMat.setPartitionerClass(ColumnRangePartitioner.class);
    angelClient.addMatrix(dfMat);
    // add sparse float matrix
    MatrixContext sfMat = new MatrixContext();
    sfMat.setName(SPARSE_FLOAT_MAT);
    sfMat.setRowNum(1);
    sfMat.setColNum(feaNum);
    sfMat.setMaxColNumInBlock(colmunSize);
    sfMat.setRowType(RowType.T_FLOAT_SPARSE);
    angelClient.addMatrix(sfMat);
    // add dense float matrix
    MatrixContext diMat = new MatrixContext();
    diMat.setName(DENSE_INT_MAT);
    diMat.setRowNum(1);
    diMat.setColNum(feaNum);
    diMat.setMaxColNumInBlock(colmunSize);
    diMat.setRowType(RowType.T_INT_DENSE);
    diMat.setPartitionerClass(ColumnRangePartitioner.class);
    angelClient.addMatrix(diMat);
    // add sparse float matrix
    MatrixContext siMat = new MatrixContext();
    siMat.setName(SPARSE_INT_MAT);
    siMat.setRowNum(1);
    siMat.setColNum(feaNum);
    siMat.setMaxColNumInBlock(colmunSize);
    siMat.setRowType(RowType.T_INT_SPARSE);
    angelClient.addMatrix(siMat);
    // add dense long matrix
    MatrixContext dlMat = new MatrixContext();
    dlMat.setName(DENSE_LONG_MAT);
    dlMat.setRowNum(1);
    dlMat.setColNum(feaNum);
    dlMat.setMaxColNumInBlock(colmunSize);
    dlMat.setRowType(RowType.T_LONG_DENSE);
    dlMat.setPartitionerClass(ColumnRangePartitioner.class);
    angelClient.addMatrix(dlMat);
    // add sparse long matrix
    MatrixContext slMat = new MatrixContext();
    slMat.setName(SPARSE_LONG_MAT);
    slMat.setRowNum(1);
    slMat.setColNum(feaNum);
    slMat.setMaxColNumInBlock(colmunSize);
    slMat.setRowType(RowType.T_LONG_SPARSE);
    angelClient.addMatrix(slMat);
    // add sparse long-key double matrix
    MatrixContext dLongKeysMatrix = new MatrixContext();
    dLongKeysMatrix.setName(SPARSE_DOUBLE_LONG_MAT);
    dLongKeysMatrix.setRowNum(1);
    dLongKeysMatrix.setColNum(feaNum);
    dLongKeysMatrix.setMaxColNumInBlock(colmunSize);
    dLongKeysMatrix.setRowType(RowType.T_DOUBLE_SPARSE_LONGKEY);
    angelClient.addMatrix(dLongKeysMatrix);
    // add sparse long-key float matrix
    MatrixContext slfMatrix = new MatrixContext();
    slfMatrix.setName(SPARSE_FLOAT_LONG_MAT);
    slfMatrix.setRowNum(1);
    slfMatrix.setColNum(feaNum);
    slfMatrix.setMaxColNumInBlock(colmunSize);
    slfMatrix.setRowType(RowType.T_FLOAT_SPARSE_LONGKEY);
    angelClient.addMatrix(slfMatrix);
    // add sparse long-key int matrix
    MatrixContext sliMatrix = new MatrixContext();
    sliMatrix.setName(SPARSE_INT_LONG_MAT);
    sliMatrix.setRowNum(1);
    sliMatrix.setColNum(feaNum);
    sliMatrix.setMaxColNumInBlock(colmunSize);
    sliMatrix.setRowType(RowType.T_INT_SPARSE_LONGKEY);
    angelClient.addMatrix(sliMatrix);
    // add sparse long-key long matrix
    MatrixContext sllMatrix = new MatrixContext();
    sllMatrix.setName(SPARSE_LONG_LONG_MAT);
    sllMatrix.setRowNum(1);
    sllMatrix.setColNum(feaNum);
    sllMatrix.setMaxColNumInBlock(colmunSize);
    sllMatrix.setRowType(RowType.T_LONG_SPARSE_LONGKEY);
    angelClient.addMatrix(sllMatrix);
    // Start PS
    angelClient.startPSServer();
    // Start to run application
    angelClient.run();
    Thread.sleep(5000);
    psId = new ParameterServerId(0);
    psAttempt0Id = new PSAttemptId(psId, 0);
    WorkerGroupId workerGroupId = new WorkerGroupId(0);
    workerId = new WorkerId(workerGroupId, 0);
    workerAttempt0Id = new WorkerAttemptId(workerId, 0);
}
Also used : CombineTextInputFormat(org.apache.hadoop.mapreduce.lib.input.CombineTextInputFormat) MatrixContext(com.tencent.angel.ml.matrix.MatrixContext) Configuration(org.apache.hadoop.conf.Configuration) PSAttemptId(com.tencent.angel.ps.PSAttemptId) WorkerAttemptId(com.tencent.angel.worker.WorkerAttemptId) ParameterServerId(com.tencent.angel.ps.ParameterServerId) WorkerId(com.tencent.angel.worker.WorkerId) WorkerGroupId(com.tencent.angel.worker.WorkerGroupId) Before(org.junit.Before)

Example 43 with MatrixContext

use of com.tencent.angel.ml.matrix.MatrixContext in project angel by Tencent.

the class MatrixRowGetTest method submit.

@Override
public void submit(Configuration conf) throws Exception {
    conf.setBoolean(AngelConf.ANGEL_AM_USE_DUMMY_DATASPLITER, true);
    AngelClient angelClient = AngelClientFactory.get(conf);
    // test the colNum is 1000000, 1 million
    // rowNum = conf.getInt("row", 1);
    // colNum = conf.getInt("col", 1000000);
    // nnz = conf.getInt("nnz", 10000);
    // blockRowNum = conf.getInt("blockRow", 1);
    // blockColNum = conf.getInt("blockCol", 20000);
    // rowNum = conf.getInt("row", 1);
    // colNum = conf.getInt("col", 10);
    // nnz = conf.getInt("nnz", 2);
    // blockRowNum = conf.getInt("blockRow", 1);
    // blockColNum = conf.getInt("blockCol", 10);
    // add sparse longkey float matrix of 1 million
    // MatrixContext dMat = new MatrixContext();
    // dMat.setName(SPARSE_LONGKEY_FLOAT_MAT);
    // dMat.setRowNum(rowNum);
    // dMat.setColNum(colNum);
    // dMat.setMaxRowNumInBlock(blockRowNum);
    // dMat.setMaxColNumInBlock(blockColNum);
    // dMat.setRowType(RowType.T_FLOAT_SPARSE_LONGKEY);
    // dMat.set(MatrixConf.MATRIX_SAVE_PATH, conf.get("angel.save.model.path"));
    // angelClient.addMatrix(dMat);
    // test the colNum is Long.MAX_VALUE
    rowNum = conf.getInt("row", 1);
    longcolNum = conf.getLong("col", Long.MAX_VALUE);
    longnnz = conf.getInt("nnz", 100000000);
    blockRowNum = conf.getInt("blockRow", 1);
    longBlockColNum = conf.getLong("blockCol", Long.MAX_VALUE / 200);
    // add sparse longkey float matrix of Long.MAX_VALUE
    MatrixContext longMat = new MatrixContext();
    longMat.setName(SPARSE_LONGKEY_FLOAT_MAT);
    longMat.setRowNum(rowNum);
    longMat.setColNum(longcolNum);
    longMat.setMaxRowNumInBlock(blockRowNum);
    longMat.setMaxColNumInBlock(longBlockColNum);
    longMat.setRowType(RowType.T_FLOAT_SPARSE_LONGKEY);
    angelClient.addMatrix(longMat);
    angelClient.startPSServer();
    angelClient.run();
    angelClient.waitForCompletion();
    angelClient.stop(0);
}
Also used : MatrixContext(com.tencent.angel.ml.matrix.MatrixContext) AngelClient(com.tencent.angel.client.AngelClient)

Example 44 with MatrixContext

use of com.tencent.angel.ml.matrix.MatrixContext in project angel by Tencent.

the class AngelClient method load.

public void load(Set<String> matrixNames) {
    // Check need load matrices
    String loadPath = conf.get(AngelConf.ANGEL_LOAD_MODEL_PATH);
    if (loadPath != null && !loadPath.isEmpty()) {
        ModelLoadContext loadContext = new ModelLoadContext(loadPath);
        int needLoadMatrixCount = 0;
        for (String name : matrixNames) {
            MatrixContext matrix = nameToMatrixMap.get(name);
            if (matrix.getAttributes().get(MatrixConf.MATRIX_LOAD_PATH) != null) {
                loadContext.addMatrix(new MatrixLoadContext(name));
                needLoadMatrixCount++;
            }
        }
        if (needLoadMatrixCount > 0) {
            load(loadContext);
        }
    }
}
Also used : MatrixContext(com.tencent.angel.ml.matrix.MatrixContext) MatrixLoadContext(com.tencent.angel.model.MatrixLoadContext) ModelLoadContext(com.tencent.angel.model.ModelLoadContext)

Example 45 with MatrixContext

use of com.tencent.angel.ml.matrix.MatrixContext in project angel by Tencent.

the class AngelPSClient method startPS.

/**
 * Start Angel ps
 *
 * @return Angel ps running context
 */
public AngelContext startPS() throws AngelException {
    // load user job resource files
    String userResourceFiles = conf.get(AngelConf.ANGEL_APP_USER_RESOURCE_FILES);
    LOG.info("userResourceFiles=" + userResourceFiles);
    if (userResourceFiles != null) {
        try {
            ConfUtils.addResourceFiles(conf, userResourceFiles);
        } catch (Throwable x) {
            throw new AngelException(x);
        }
    }
    int psNum = conf.getInt(AngelConf.ANGEL_PS_NUMBER, AngelConf.DEFAULT_ANGEL_PS_NUMBER);
    if (psNum <= 0) {
        throw new AngelException("Invalid parameter:Wrong ps number!");
    }
    conf.set(AngelConf.ANGEL_RUNNING_MODE, RunningMode.ANGEL_PS.toString());
    client.addMatrix(new MatrixContext("init", 1, psNum, -1, -1, -1, 1, 1, new ArrayList<>(), RowType.T_DOUBLE_SPARSE));
    client.startPSServer();
    client.run();
    return new AngelContext(client.getMasterLocation(), conf);
}
Also used : AngelException(com.tencent.angel.exception.AngelException) MatrixContext(com.tencent.angel.ml.matrix.MatrixContext) ArrayList(java.util.ArrayList)

Aggregations

MatrixContext (com.tencent.angel.ml.matrix.MatrixContext)52 Configuration (org.apache.hadoop.conf.Configuration)35 CombineTextInputFormat (org.apache.hadoop.mapreduce.lib.input.CombineTextInputFormat)32 PSAttemptId (com.tencent.angel.ps.PSAttemptId)27 ParameterServerId (com.tencent.angel.ps.ParameterServerId)27 Before (org.junit.Before)27 WorkerAttemptId (com.tencent.angel.worker.WorkerAttemptId)26 WorkerGroupId (com.tencent.angel.worker.WorkerGroupId)26 WorkerId (com.tencent.angel.worker.WorkerId)26 TaskId (com.tencent.angel.worker.task.TaskId)14 AngelClient (com.tencent.angel.client.AngelClient)7 AngelException (com.tencent.angel.exception.AngelException)6 DummyTask (com.tencent.angel.master.DummyTask)4 IOException (java.io.IOException)3 ArrayList (java.util.ArrayList)3 BeforeClass (org.junit.BeforeClass)3 InvalidParameterException (com.tencent.angel.exception.InvalidParameterException)2 Test (org.junit.Test)2 ServiceException (com.google.protobuf.ServiceException)1 PartitionKey (com.tencent.angel.PartitionKey)1