Label phase not working in the third round as FTD is not finding any new training data

[sania-16]

[sonalgoyal]

this happens during model grid search. Since we try to build generic params here and case a wider net, some values end up being cmpletely off. Changing the params and multiple combinations does not help.

To avoid this case, labelDataSampleSize can be adjusted to get fresh samples on which a model can get formed.

So far, no other case has been reported of a model not being created and ultimately the convergence does happen. To avoid the log, moving the log level to fatal for this scenario. d3b5136

If we still see this issue, will need a much further debug.