Custom pipes for Docker and Snowflake and deprecate nlu_ref

[C-K-Loan]

Improvements to Docker integration

• Docs have been updated to reflect changes

Instead of providing a nlu model reference for deploying a model, users have now 2 options

1) Deploy Pipeline via Name, Language and Bucket

nlp.build_image(pipeline_name=pipe_name,  pipeline_language=pipe_lang,  pipeline_bucket=pipe_bucket)

2.1) Deploy custom pipeline by providing a fitted pipeline object

custom_pipe = nlp.Pipeline(stages=[...]).fit(df)
nlp.build_image(custom_pipe=custom_pipe)

2.2) Deploy NLU pipelines as container

Any nlu pipeline can still be deployed as a custom pipe

# Load Model
pipe = nlp.load(model_nlu_ref)  
# Predict so that under the hood pipeline is fitted and vanilla_transformer_pipe attribute is avaiable
pipe.predict('')  
# Now we can just handle it like a custom pipeline
nlp.build_image(custom_pipe=pipe.vanilla_transformer_pipe)

---

Improvements to Snowflake integration

• Docs have been updated to reflect changes
• Compute pool parameters compute_pool_min_nodes,compute_pool_max_nodes,compute_pool_instance_family can now be configured in nlp.snowflake_common_setup with values defined in the Snowflake Documentation

Since the Snowflake integration uses the Docker, all of the changes for the docker utilities are reflected into Snowflake as well.

Instead of providing a nlu model reference for deploying a model, users have now 2 options

Deploy Pipeline via Name, Language and Bucket

nlp.build_image(pipeline_name=pipe_name,  pipeline_language=pipe_lang,  pipeline_bucket=pipe_bucket)

Deploy custom pipeline by providing a fitted pipeline object

custom_pipe = nlp.Pipeline(stages=[...]).fit(df)
nlp.build_image(custom_pipe=custom_pipe)

Deploy NLU pipelines as container
Any nlu pipeline can still be deployed

# Load Model
pipe = nlp.load(model_nlu_ref)  
# Predict so that under the hood pipeline is fitted and vanilla_transformer_pipe attribute is avaiable
pipe.predict('')  
# Now we can just handle it like a custom pipeline
nlp.build_image(custom_pipe=pipe.vanilla_transformer_pipe)