}

    @Override
    public ModelSource resolve(ModelLocator locator, String relative) {
        String norm = relative.replace('\\', File.separatorChar).replace('/', File.separatorChar);
        Path path = getPath().getParent().resolve(norm);
        Path relatedPom = locator.locateExistingPom(path);
        if (relatedPom != null) {
            return new PathSource(relatedPom.normalize(), null);

    private static final String KEEP_EXPIRES_DOCS = "keep_expires_docs";

    protected long crawlingExecutionInterval = Constants.DEFAULT_CRAWLING_EXECUTION_INTERVAL;

    protected int crawlerPriority = Thread.NORM_PRIORITY;

    protected final List<DataCrawlingThread> dataCrawlingThreadList = Collections.synchronizedList(new ArrayList<>());

    public void crawl(final String sessionId) {

    protected long crawlingExecutionInterval = Constants.DEFAULT_CRAWLING_EXECUTION_INTERVAL;

    protected int indexUpdaterPriority = Thread.MAX_PRIORITY;

    protected int crawlerPriority = Thread.NORM_PRIORITY;

    protected final List<Crawler> crawlerList = Collections.synchronizedList(new ArrayList<>());

    public void crawl(final String sessionId, final List<String> webConfigIdList, final List<String> fileConfigIdList) {

        metrics, preprocessing layers, and stateful RNN layers)
    *   Update Keras batch normalization layer to use the running mean and
        average computation in the `fused_batch_norm`. You should see
        significant performance improvements when using `fused_batch_norm` in
        Eager mode.

*   `tf.lite`:

    *   Enable TFLite experimental new converter by default.

*   XLA