Training module#

Module with the classes that allows to manage your training experiments.

MLOpsTrainingExecution#

class mlops_codex.training.MLOpsTrainingExecution(*, training_id: str, group: str, exec_id: str, login: str | None = None, password: str | None = None, url: str | None = None)[source]#

Bases: MLOpsExecution

Class to manage trained models.

Parameters:

  • training_id (str) – Training id (hash) from the experiment you want to access

  • group (str) – Group the training is inserted.

  • exec_id (str) – Execution id for that specific training run

  • login (str) – Login for authenticating with the client. You can also use the env variable MLOPS_USER to set this

  • password (str) – Password for authenticating with the client. You can also use the env variable MLOPS_PASSWORD to set this

  • environment (str) – Environment of MLOps you are using.

  • run_data (dict) – Metadata from the execution.

Raises:

Example

from mlops_codex.training import MLOpsTrainingClient
from mlops_codex.base import MLOpsExecution

client = MLOpsTrainingClient('123456')
client.create_group('ex_group', 'Group for example purpose')
training = client.create_training_experiment('Training example', 'Classification', 'ex_group')
print(client.get_training(training.training_id, 'ex_group').training_data)

data_path = './samples/train/'

run = training.run_training('First test', data_path+'dados.csv', training_reference='train_model', training_type='Custom', python_version='3.9', requirements_file=data_path+'requirements.txt', wait_complete=True)

print(run.get_training_execution(run.exec_id))
print(run.download_result())

run.promote_model('Teste notebook promoted custom', 'score', data_path+'app.py', data_path+'schema.json',  'csv')
get_status() → dict[source]#

Gets the status of the related execution.

Raises:

ExecutionError – Execution unavailable

Returns:

Returns the execution status.

Return type:

dict

promote_model(*, model_name: str, model_reference: str, source_file: str, input_type: str, operation: str | None = 'Sync', schema: str | dict | None = None, extra_files: list | None = None, requirements_file: str | None = None, env: str | None = None, wait_complete: bool | None = False) → SyncModel | AsyncModel[source]#

Upload models trained inside MLOps.

Parameters:

  • model_name (str) – The name of the model, in less than 32 characters

  • model_reference (Optional[str], optional) – The name of the scoring function inside the source file

  • source_file (Optional[str], optional) – Path of the source file. The file must have a scoring function that accepts two parameters: data (data for the request body of the model) and model_path (absolute path of where the file is located)

  • schema (Union[str, dict], optional) – Path to a JSON or XML file with a sample of the input for the entrypoint function. A dict with the sample input can be sending as well

  • extra_files (list, optional) – A optional list with additional files paths that should be uploaded. If the scoring function refer to this file they will be on the same folder as the source file

  • requirements_file (str, optional) – Path of the requirements file. This will override the requirements used in trainning. The packages versions must be fixed eg: pandas==1.0

  • env (str, optional) – Flag that choose which environment (dev, staging, production) of MLOps you are using. Default is True

  • operation (str) – Defines which kind operation is being executed (Sync or Async). Default value is Sync

  • input_type (str) – The type of the input file that should be ‘json’, ‘csv’ or ‘parquet’

  • wait_complete (Optional[bool], default=False) – Wait for model to be ready and returns a MLOpsModel instance with the new model

Raises:

TrainingError – The training execution shouldn’t be succeeded to be promoted

Returns:

The new training model

Return type:

Union[SyncModel, AsyncModel]

MLOpsTrainingExperiment#

class mlops_codex.training.MLOpsTrainingExperiment(*, training_id: str, login: str | None = None, password: str | None = None, group: str = 'datarisk', url: str = 'https://neomaril.datarisk.net/')[source]#

Bases: BaseMLOps

Class to manage models being trained inside MLOps

Parameters:

  • login (str) – Login for authenticating with the client. You can also use the env variable MLOPS_USER to set this

  • password (str) – Password for authenticating with the client. You can also use the env variable MLOPS_PASSWORD to set this

  • training_id (str) – Training id (hash) from the experiment you want to access

  • group (str) – Group the training is inserted.

  • environment (str) – Flag that choose which environment of MLOps you are using. Test your deployment first before changing to production. Default is True

  • executions (List[int]) – Ids for the executions in that training

Raises:

Example

from mlops_codex.training import MLOpsTrainingClient
from mlops_codex.base import MLOpsExecution

client = MLOpsTrainingClient('123456')
client.create_group('ex_group', 'Group for example purpose')
training = client.create_training_experiment('Training example', 'Classification', 'ex_group')
print(client.get_training(training.training_id, 'ex_group').training_data)

data_path = './samples/train/'

run = run = training.run_training('First test', data_path+'dados.csv', training_reference='train_model', training_type='Custom', python_version='3.9', requirements_file=data_path+'requirements.txt', wait_complete=True)

print(run.get_training_execution(run.exec_id))
print(run.download_result())
get_all_training_executions() → List[MLOpsTrainingExecution][source]#

Get all executions from that experiment.

Returns:

All executions from that training

Return type:

List[MLOpsTrainingExecution]

get_training_execution(exec_id: str | None = None) → MLOpsTrainingExecution[source]#

Get the execution instance.

Parameters:

exec_id (Optional[str], optional) – Execution id. If not informed we get the last execution.

Returns:

The chosen execution

Return type:

MLOpsExecution

log_train(*, name, X_train, y_train, description: str | None = None, save_path: str | None = None)[source]#
run_training(*, run_name: str, training_type: str = 'External', description: str | None = None, train_data: str | None = None, dataset: str | MLOpsDataset | None = None, training_reference: str | None = None, python_version: str = '3.10', conf_dict: str | dict | None = None, source_file: str | None = None, requirements_file: str | None = None, extra_files: list | None = None, env: str | None = None, X_train=None, y_train=None, model_outputs=None, model_file: str | None = None, model_metrics: str | dict | None = None, model_params: str | dict | None = None, model_hash: str | None = None, wait_complete: bool | None = False) → dict | MLOpsExecution[source]#

Runs a prediction from the current model.

Parameters:

  • run_name (str) – The name of the model, in less than 32 characters

  • train_data (str) – Path of the file with train data.

  • training_reference (Optional[str], optional) – The name of the training function inside the source file. Just used when training_type is Custom

  • training_type (str) – Can be Custom, AutoML or External

  • description (Optional[str], optional) – Description of the experiment

  • python_version (Optional[str], optional) – Python version for the training environment. Available versions are 3.8, 3.9, 3.10. Defaults to ‘3.10’

  • conf_dict (Union[str, dict]) – Path to a JSON file with the AutoML configuration. A dict can be sending as well. Just used when training_type is AutoML

  • source_file (Optional[str], optional) – Path of the source file. The file must have a training function that accepts one parameter: model_path (absolute path of where the file is located). Just used when training_type is Custom

  • requirements_file (str) – Path of the requirements file. The packages versions must be fixed eg: pandas==1.0. Just used when training_type is Custom

  • env (Optional[str], optional) – .env file to be used in your training enviroment. This will be encrypted in the server.

  • extra_files (Optional[list], optional) – A optional list with additional files paths that should be uploaded. If the scoring function refer to this file they will be on the same folder as the source file. Just used when training_type is Custom

  • wait_complete (Optional[bool], optional) – Boolean that informs if a model training is completed (True) or not (False). Default value is False

Raises:

InputError – Some input parameters its invalid

Returns:

The return of the scoring function in the source file for Sync models or the execution class for Async models.

Return type:

Union[dict, MLOpsExecution]

Example

>>> execution = run = training.run_training('First test', data_path+'dados.csv', training_reference='train_model', python_version='3.9', requirements_file=data_path+'requirements.txt', wait_complete=True)

MLOpsTrainingClient#

class mlops_codex.training.MLOpsTrainingClient(login: str | None = None, password: str | None = None, tenant: str | None = None, url: str | None = None)[source]#

Bases: BaseMLOpsClient

Class for client for accessing MLOps and manage models

Parameters:

  • login (str) – Login for authenticating with the client. You can also use the env variable MLOPS_USER to set this

  • password (str) – Password for authenticating with the client. You can also use the env variable MLOPS_PASSWORD to set this

  • url (str) – URL to MLOps Server. Default value is https://neomaril.datarisk.net, use it to test your deployment first before changing to production. You can also use the env variable MLOPS_URL to set this

Raises:

Example

from mlops_codex.training import MLOpsTrainingClient

client = MLOpsTrainingClient('123456')
client.create_group('ex_group', 'Group for example purpose')
training = client.create_training_experiment('Training example', 'Classification',  'Custom', 'ex_group')
print(client.get_training(training.training_id, 'ex_group').training_data)
create_training_experiment(*, experiment_name: str, model_type: str, group: str, force: bool | None = False) → MLOpsTrainingExperiment[source]#

Create a new training experiment on MLOps.

Parameters:

  • experiment_name (str) – The name of the experiment, in less than 32 characters

  • model_type (str) – The name of the scoring function inside the source file.

  • group (str) – Group the model is inserted. Default to ‘datarisk’ (public group)

  • force (Optional[bool], optional) – Forces to create a new training with the same model_type, experiment_name, group

Raises:
Returns:

A MLOpsTrainingExperiment instance with the training hash from training_id

Return type:

MLOpsTrainingExperiment

Example

>>> training = client.create_training_experiment('Training example', 'Classification', 'ex_group')
get_training(*, training_id: str, group: str) → MLOpsTrainingExperiment[source]#

Acess a model using its id

Parameters:

  • training_id (str) – Training id (hash) that needs to be acessed

  • group (str) – Group the model is inserted.

Raises:
Returns:

A MLOpsTrainingExperiment instance with the training hash from training_id

Return type:

MLOpsTrainingExperiment

Example

>>> training = get_training('Tfb3274827a24dc39d5b78603f348aee8d3dbfe791574dc4a6681a7e2a6622fa')