Skip to content

You can download the code of this step here or all the steps here.

For Notebooks

The "Getting Started" Notebook is available here. In Taipy GUI, the process to execute a Jupyter Notebook is different from executing a Python Script. It is important to check the Notebook content and see the documentation.

Step 12: Compare Scenarios

Cycles are helpful to keep track of KPI over time. The goal of this step is to compare the primary scenario of every cycle and its pipelines over time.

To achieve this:

  • A new dataframe has to be initialized. It will store the metrics for the baseline and ml pipeline.

  • Then, a part will use a boolean to show or not the comparison.

  • Finally, a selector will change the displayed metrics of the graph.

# Initial dataset for comparison
comparison_scenario = pd.DataFrame({"Scenario Name":[],
                                    "RMSE baseline":[], "MAE baseline":[],
                                    "RMSE ML":[], "MAE ML":[]})

# Indicates if the comparison is done
comparison_scenario_done = False

# Selector for metrics
metric_selector = ["RMSE", "MAE"]
selected_metric = metric_selector[0]

First, a function has to be created to compare the primary scenario of all the cycles. tp.get_primary_scenarios() is the useful function to use for this effect. compare() goes through all of these scenarios and pipelines and add the metrics in lists. In the end, state.comparison_scenario is updated and comparison_scenario_done set to True.

from sklearn.metrics import mean_absolute_error, mean_squared_error

def compute_metrics(historical_data, predicted_data):
    rmse = mean_squared_error(historical_data, predicted_data)
    mae = mean_absolute_error(historical_data, predicted_data)
    return rmse, mae

def compare(state):
    # Initial lists for comparison
    scenario_names = []
    rmses_baseline = []
    maes_baseline = []
    rmses_ml = []
    maes_ml = []

    # Go through all the primary scenarios
    all_scenarios = tp.get_primary_scenarios()
    all_scenarios_ordered = sorted(all_scenarios, key=lambda x: x.creation_date.timestamp())

    for scenario in all_scenarios_ordered:
        print(f"Scenario {}")
        # Go through all the pipelines
        for pipeline in scenario.pipelines.values():
            print(f"     Pipeline {pipeline.config_id}")
            # Get the predictions dataset with the historical data
            only_prediction_dataset = create_predictions_dataset(pipeline)[]

            # Series to compute the metrics (true values and predicted values)
            historical_values = only_prediction_dataset["Historical values"]
            predicted_values = only_prediction_dataset["Predicted values"]

            # Compute the metrics for this pipeline and primary scenario
            rmse, mae = compute_metrics(historical_values, predicted_values)

            # Add values to the appropriate lists
            if "baseline" in pipeline.config_id:
            elif "ml" in pipeline.config_id:

        scenario_names.append(scenario.creation_date.strftime("%A %d %b"))

    # Update comparison_scenario
    state.comparison_scenario = pd.DataFrame({"Scenario Name":scenario_names,
                                              "RMSE baseline":rmses_baseline,
                                              "MAE baseline":maes_baseline,
                                              "RMSE ML":rmses_ml,
                                              "MAE ML":maes_ml})

    # When comparison_scenario_done will be set to True,
    # the part with the graphs will be finally rendered
    state.comparison_scenario_done = True

Let's create a page related to this comparison. As said before, this page will contain a graph to compare scenarios and pipelines; and a selector to choose the metric on which to compare. When pressed the button at the bottom of the page calls the compare() function. When finished, the render property of the part will render the rest of the page. Also, a new Taipy's block is present in the Markdown: expandable.

# Performance page
page_performance = """




<|{comparison_scenario}|chart|type=bar|x=Scenario Name|y[1]=RMSE baseline|y[2]=RMSE ML|height=100%|width=100%|>

<|{comparison_scenario}|chart|type=bar|x=Scenario Name|y[1]=MAE baseline|y[2]=MAE ML|height=100%|width=100%|>


<|Compare primarys|button|on_action=compare|>

Page Performance

# Add the page_performance section to the menu   
multi_pages = """
<|menu|label=Menu|lov={["Data Visualization", "Scenario Manager", "Performance"]}|on_action=on_menu|>

<|part|render={page=="Data Visualization"}|""" + page_data_visualization + """|>
<|part|render={page=="Scenario Manager"}|""" + page_scenario_manager + """|>
<|part|render={page=="Performance"}|""" + page_performance + """|>



Compare Scenarios