awaiskaleem

def create_experiment(proj_id,project_description):
        for i in [1]:
            try:
                mlflow.create_experiment(proj_id)
            except:
                continue
        client.set_experiment_tag(mlflow.get_experiment_by_name(proj_id).experiment_id
                                                          ,"mlflow.note.content"
                                                          ,project_description)

awaiskaleem

def set_run(proj_id):
        '''
        + This creates a file called "active_run" in S3 and writes current run_id into it.
        + If a file named "live_run" does not exist, it creates one and throws active_run into it 
        '''
        s3_client = boto3.Session(profile_name=None).client('s3')
        s3_resource = boto3.resource('s3')
        artifact_bucket = 'YOUR ARTIFACT BUCKET ON S3'
        active_run_id = mlflow.active_run().info.run_id

awaiskaleem

def get_run(proj_id):
        '''
        get active and live runs for this model
        '''
        s3_client = boto3.Session(profile_name=None).client('s3')
        artifact_bucket = 'YOUR ARTIFACT BUCKET ON S3'
        
        s3_object = s3_client.get_object(Bucket=artifact_bucket, Key='mlflow/'+proj_id+'/active_model_run')
        active_run = s3_object['Body'].read().decode("utf-8")
        s3_object = s3_client.get_object(Bucket=artifact_bucket, Key='mlflow/'+proj_id+'/live_model_run')

awaiskaleem

def update_learning(recall,recall_live):
        
        s3_client = boto3.Session(profile_name=None).client('s3')
        s3_resource = boto3.resource('s3')
        artifact_bucket = 'YOUR ARTIFACT BUCKET ON S3'
        
        if recall>recall_live:
            # Push live champion to history
            try:
                object = s3_client.get_object(Bucket=artifact_bucket, Key='mlflow/'+proj_id+'/live_model_run_history')

awaiskaleem

def get_parameter(run_id, param_name):
  return ast.literal_eval(mlflow.get_run(run_id).data.params[param_name])
   

awaiskaleem

def save_model(model,model_name):
  mlflow.spark.log_model(model,model_name)
    

awaiskaleem

def load_model(run_id, model_name):
  model = mlflow.get_run(run_id).info.artifact_uri+"/model_name/sparkml"
  return PipelineModel.load(model)

awaiskaleem

import logging
import mlflow


logger = logging.getLogger(__name__)

class BaseMLLogger:
    """
    Base class for tracking experiments.
    This class can be extended to implement custom logging backends like MLFlow, Tensorboard, or Sacred.