Logging module

MLOps has a log parser that clean logs and tries to find useful information. This is the module to access this functionalities. Since there are a lot of logger types is hard to create the perfect parser, so using this one helps your model script being parsed by MLOps and sending cleaner messages.

Logger

class mlops_codex.logging.Logger(model_type: str)

Bases: object

Class to custom logger for model scripts.

model_type

Attribute that designates the type of the model being executed. Can be ‘Sync’ or ‘Async’

Type:

str

Raises:

InputError – Invalid input for the logging functions

Example

The logger needs to be implemented inside the function being executed by MLOps like this:

from joblib import load
import pandas as pd
from mlops_codex.logging import Logger

def score(data_path, model_path):
    logger = Logger('Async')

    logger.debug("USING LOGGER")

    model = load(model_path+"/model.pkl")

    df = pd.read_csv(data_path+'/input.csv')

    if len(df) < 5:
        logger.warning("DF is less than 5 lines")

    df['score'] = 1000 * (1-model.predict_proba(df)[:,1])

    output = data_path+'/output.csv'

    df.to_csv(output, index=False)

    return output

callback(output: str | int | float | list | dict) → str

Compile the logs with the response for Sync models only. Should be the return of function being executed. This output should be able to be parsed as a JSON, so if you are using a non-primitive type as your return, make sure it can be parsed by json.dumps.

Example

def score(data, base_path):
    logger = Logger('Sync')

    logger.debug("USING LOGGER")

    model = load(base_path+"/model.pkl")

    df = pd.DataFrame(data=json.loads(data), index=[0])

    return logger.callback({"score": 1000 * (1-float(model.predict_proba(df)[0,1]))})

Parameters:

output (str) – Output of the function being executed.

debug(message: str) → None

Logs a DEBUG message.

Parameters:

message (str) – Message that will be logged

error(message: str) → None

Logs a ERROR message.

Parameters:

message (str) – Message that will be logged

warning(message: str) → None

Logs a WARNING message.

Parameters:

message (str) – Message that will be logged