TADManager.java

package org.nd4j.linalg.cache;

import org.apache.commons.math3.util.Pair;
import org.nd4j.linalg.api.buffer.DataBuffer;
import org.nd4j.linalg.api.ndarray.INDArray;

/**
 * This interface describes TAD caching.
 *
 * While working with tensors, all operations are happening on some demensions,
 * and since training tasks are repetitive, we can pre-calculate TAD shapes and offsets once,
 * and reuse them later during whole training process.
 *
 *
 *
 * @author raver119@gmail.com
 */
public interface TADManager {

    /**
     * This method returns TAD shapeInfo and all offsets
     * for specified tensor and dimensions.
     *
     * @param array Tensor for TAD precalculation
     * @param dimension
     * @return
     */
    Pair<DataBuffer, DataBuffer> getTADOnlyShapeInfo(INDArray array, int... dimension);

    /**
     * This method removes all cached shape buffers
     */
    void purgeBuffers();
}