Working with Results#

Once flow is interpreted and executed, you may obtain results of the execution.

Let’s consider the following example:

graph LR A{{Training Data}} --> P1[Preprocess] P1 --> P2([Training]) B{{Training Labels}} --> P2 P2 -->|trained_model| P3[Evaluation] P2 -->|trained_model| P4[Training Report Generation] P2 -->|metrics| P5 P3 --> P5([Final Report Generation]) P4 --> P5

To compose and execute this flow, you would write code as follows:

from malevich.app import preprocess, train, evaluate, training_report, final_report
from malevich import flow, collection, CoreInterpreter

@flow
def train_and_evaluate():
    # Define the start of the flow with collections.
    training_data = collection('training_data')
    training_labels = collection('training_labels')

    # Passing a collection to a processor function will
    # automatically register the dependency (Training Data --> Preprocess).
    preprocessed_data = preprocess(training_data)

    # Passing results of previous processors will also register
    # dependencies (Preprocess --> Training).
    trained_model, metrics = train(preprocessed_data, training_labels)

    # Processors for generating reports and evaluation results.
    training_report_result = training_report(trained_model)
    evaluation_result = evaluate(trained_model)

    # The final report is generated based on previous outcomes.
    return trained_model, final_report(training_report_result, evaluation_result, metrics)

# Deploy and execute the flow.
task = train_and_evaluate()
task.interpret(CoreInterpreter(core_auth=('example', 'Welcome to Malevich!')))
task.prepare()
task.run()
task.stop()

results = task.results()

Note

Now, the object results will contain results of flow execution. We fetch outputs of operations you have returned from the flow function (ones in rounded rectangles). In this case, the output of train processor is assumed to be an asset and a collection. The output of final_report processor is assumed to be a collection. So, the variable results will be a list of two objects of type malevich.models.results.base.BaseResult.

Different interpreters address results in different ways. We have run task on Malevich Core, so we actually get a list of malevich.models.results.core.CoreResult objects. Here is a way to extract actual results:

# CoreResultPayload
train_results = results[0].get()
# pandas.DataFrame
report = results[1].get_df()

# train returns a
model, metrics = train_results[0].data, train_results[1].data
Show Source