[GNMT v2/Tensorflow] API Support for Single Queries (similar to google translate api)

[abbyDC]

For translate mode, I think it was optimized for doing batch translations and I can say it is already fast for that use case. The problem is if I just wanted to translate 1 sentence, it takes the same time to process a batch of statements.

Feature wanted:
I wish there is a mode like the google translate api where in I could input a single statement and it will return the corresponding translation of that sentence almost instantly.

**I may be wrong so if this feature is already present, I'd appreciate it if you'd comment it here so I could close the issue.

[mwawrzos]

Can you give an example, how google translate api can be used in the mode you mean?

It would be great if you could share some bash/python script, that shows step by step, how it looks like.

[abbyDC]

@mwawrzos Similar to this example from https://pythonprogramming.net/translation-api-google-cloud-tutorial/. So there is a function which takes in a text and it will return the translation. This is a quick way to get a translation for a single string.

from google.cloud import translate
def translate_text(text,target='en'):
    """
    Target must be an ISO 639-1 language code.
    https://cloud.google.com/translate/docs/languages
    """
    translate_client = translate.Client()
    result = translate_client.translate(
        text,
        target_language=target)

    print(u'Text: {}'.format(result['input']))
    print(u'Translation: {}'.format(result['translatedText']))
    print(u'Detected source language: {}'.format(
        result['detectedSourceLanguage']))
example_text ="Hola saludos" 
translate_text(example_text.decode('utf-8'),target='en')
### OUTPUT ###
"""
Text: Hola saludos
Translation: Hello greetings
Detected source language: es
"""

Issue with current translation: Upon checking the code, it seems like the Estimator.predict() causes the inference to slow down for a single query since it needs to create a graph for each prediction.