added hyper param search feature in jasper_infer.py

[harisankarh]

Added grid search-based alpha-beta hyper param tuning code in jasper_infer.py. Incorporated suggestions from pull request #55.

Added following changes:

• beam_width parameter to override default 128
• alpha_step, beta_step parameters to define steps for grid search
• rename alpha_from/beta_from to just alpha/beta so that it is possible to do inference with a single set of alpha/beta values (without tuning over a grid)

Signed-off-by: Hariankar H [email protected]

[Slyne]

I think there is no need to add greedy_hypotheses here for each round evaluation of alpha & beta.

[harisankarh]

@Slyne thanks for pointing it out. Removed it.

[okuchaiev]

the CI fails because of PEP8 issue:

Could you please run "pycodestyle" before the check-in? (see below)

+ pycodestyle . --exclude=./scripts/get_librispeech_data.py,./scripts/process_beam_dump.py,./tests/other/jasper.py,./tests/other/jasper_zero_dl.py,./collections/nemo_nlp/nemo_nlp/utils/metrics/sacrebleu.py,./collections/nemo_nlp/nemo_nlp/utils/metrics/fairseq_tokenizer.py,./nemo/setup.py,./docs/sources/source/conf.py,./docs/sources/source/tutorials/infer.py,./docs/sources/source/tutorials/test.py,./collections/nemo_nlp/build

./examples/asr/jasper_infer.py:254:1: E305 expected 2 blank lines after class or function definition, found 1

script returned exit code 1

[harisankarh]

fixed the pep8 issues

[blisc]

Do you plan on using the infer mode? If not, I plan on removing infer mode and renaming the file to jasper_eval.py since the intended use case for that file is mainly eval.