Experiment mode

With experiment mode, you can save and name your application as an experiment version. When running an experiment version in experiment mode, Taipy only considers the entities attached to this version. It filters out all other entities from different versions, so you can continue your development process without worrying about losing entities of your other versions.

Use case

The experiment mode is handy in the process of improving a stable version of a Taipy application. In that case, you need to run one or multiple experiments like tuning algorithm parameters, updating configuration by adding KPI data nodes, testing new algorithms, etc. These require some configuration changes so that each experiment has its own configuration. Each run of an experiment must be analyzed, evaluated, and eventually compared to others.

In that case, you want to keep your experiments in a dedicated version with the corresponding configuration and entities. You also need to be able to re-run the application eventually with some configuration changes without deleting the experiment's entities.

In the following, we consider the basic Taipy Core application main.py:

import taipy.core as tp
from taipy import Config


def my_print_algo(entry: str):
    # does nothing!
    return entry


input_cfg = Config.configure_data_node("input", default_data="a_string")
output_cfg = Config.configure_data_node("output")
task_cfg = Config.configure_task("my_print_algo", my_print_algo, input_cfg, output_cfg)
scenario_cfg = Config.configure_scenario_from_tasks("my_scenario", [task_cfg])

if __name__ == "__main__":
    tp.Core().run()
    tp.create_scenario(scenario_cfg)
    print(f"nb scenarios: {len(tp.get_scenarios())}")

Create a version¶

To create an experiment version you can run a Taipy application on your command line interface with --experiment option.

$ python main.py -l
Version number                         Mode                   Creation date
26e56e84-9e7e-4a26-93f6-443d9aa541d9   Development (latest)   2023-01-25 12:20:33

$ python main.py --experiment
[2023-01-25 12:20:56,474][Taipy][INFO] job JOB_my_print_algo_e1c49bdb-9284-40c5-a096-db0235697cb3 is completed.
nb scenarios: 1

$ python main.py -l
Version number                         Mode                   Creation date
325d0618-6f9e-459b-9597-48fa93a57a23   Experiment (latest)    2023-01-25 12:20:56
26e56e84-9e7e-4a26-93f6-443d9aa541d9   Development            2023-01-25 12:20:33

An experiment version is created and stored. As you can see the number of scenarios displayed is 1. That means Taipy only considers the scenarios of the experiment version, and filters out the entities of the other development version. By default, a random name is used to identify the version of your application.

Name a version¶

You can explicitly define the name of an experiment version by providing the name in the command line.

$ python main.py -l
Version number                         Mode                   Creation date
325d0618-6f9e-459b-9597-48fa93a57a23   Experiment (latest)    2023-01-25 12:20:56
26e56e84-9e7e-4a26-93f6-443d9aa541d9   Development            2023-01-25 12:20:33

$ python main.py --experiment 1.0
[2023-01-25 12:24:19,613][Taipy][INFO] job JOB_my_print_algo_9b6232f9-601e-4a85-852e-2ada7bc1e459 is completed.
nb scenarios: 1

$ python main.py -l
Version number                         Mode                   Creation date
1.0                                    Experiment (latest)    2023-01-25 12:24:19
325d0618-6f9e-459b-9597-48fa93a57a23   Experiment             2023-01-25 12:20:56
26e56e84-9e7e-4a26-93f6-443d9aa541d9   Development            2023-01-25 12:20:33

With --experiment 1.0 option, an experiment version is created and stored under the name "1.0".

In this example, you can see the number of scenarios displayed is still 1. Taipy only considered the scenarios of the 1.0 version, and filters out the entities of other versions.

Run an existing version¶

To run an existing experiment version, the configuration of the application must be the same. No change must be done. You can run a Taipy application on your command line interface with --experiment VERSION option and the name of the existing version.

$ python main.py -l
Version number                         Mode                   Creation date
1.0                                    Experiment (latest)    2023-01-25 12:24:19
325d0618-6f9e-459b-9597-48fa93a57a23   Experiment             2023-01-25 12:20:56
26e56e84-9e7e-4a26-93f6-443d9aa541d9   Development            2023-01-25 12:20:33

$ python main.py --experiment 1.0
[2023-01-25 12:28:54,963][Taipy][INFO] job JOB_my_print_algo_2133dc18-643b-4351-872b-aedfc2c65c9c is completed.
nb scenarios: 2

$ python main.py -l
Version number                         Mode                   Creation date
1.0                                    Experiment (latest)    2023-01-25 12:24:19
325d0618-6f9e-459b-9597-48fa93a57a23   Experiment             2023-01-25 12:20:56
26e56e84-9e7e-4a26-93f6-443d9aa541d9   Development            2023-01-25 12:20:33

As you can see, this time the number of scenarios displayed is 2. Indeed, we run the 1.0 version twice, so we have two scenarios attached to it. Scenarios from other versions are filtered out.

Clean entities of an existing version¶

To run an experiment version with a fresh start, you can run on experiment mode with --clean-entities option. Taipy deletes the entities of the version provided before running the application again.

$ python main.py -l
Version number                         Mode                   Creation date
1.0                                    Experiment (latest)    2023-01-25 12:24:19
325d0618-6f9e-459b-9597-48fa93a57a23   Experiment             2023-01-25 12:20:56
26e56e84-9e7e-4a26-93f6-443d9aa541d9   Development            2023-01-25 12:20:33

$ python main.py --experiment 1.0 --clean-entities
[2023-01-25 12:36:05,598][Taipy][INFO] Clean all entities of version 1.0
[2023-01-25 12:36:05,777][Taipy][INFO] job JOB_my_print_algo_494bf4a7-afa2-4916-9221-fabd8de1738a is completed.
nb scenarios: 1

$ python main.py -l
Version number                         Mode                   Creation date
1.0                                    Experiment (latest)    2023-01-25 12:24:19
325d0618-6f9e-459b-9597-48fa93a57a23   Experiment             2023-01-25 12:20:56
26e56e84-9e7e-4a26-93f6-443d9aa541d9   Development            2023-01-25 12:20:33

As you can see in the previous example the number of scenarios became 1 again. Indeed, Taipy cleaned all entities of that experiment version before running it again (which created a new scenario). Scenarios from other versions are not deleted and still filtered out.

Change config of an existing version¶

As explained before, there is a constraint when re-running an existing version. The configuration of the application must be the same. No change must be made.

But don't worry; if there is any change to your configuration, Taipy will show a warning message before exiting.

Let's assume a change has been made in the configuration in main.py. A custom property (description) has been added to the output data node config. Here is the new configuration.

import taipy.core as tp
from taipy import Config


def my_print_algo(entry: str):
    # does nothing!
    return entry


input_cfg = Config.configure_data_node("input", default_data="a_string")
output_cfg = Config.configure_data_node("output", description="What a description")
task_cfg = Config.configure_task("my_print_algo", my_print_algo, input_cfg, output_cfg)
scenario_cfg = Config.configure_scenario_from_tasks("my_scenario", [task_cfg])

if __name__ == "__main__":
    tp.Core().run()
    tp.create_scenario(scenario_cfg)
    print(f"nb scenarios: {len(tp.get_scenarios())}")

$ python main.py --experiment 1.0
[2023-01-25 12:52:05,484][Taipy][WARNING] The Configuration of version 1.0 is conflicted with the current Python Config.
Added object:
        DATA_NODE "output" has attribute "description" added: What a description

To force running the application with the changes, run your application with --taipy-force option.

In the example above, when re-running version 1.0, Taipy detects and displays all the differences, so you precisely know what has been changed and can decide what to do. Either you revert your configuration changes, or you can run the application with the --taipy-force option to force Taipy to update the configuration of the provided version before re-running the application.

$ python main.py --experiment 1.0 --taipy-force
[2023-01-25 12:55:05,484][Taipy][WARNING] The Configuration of version 1.0 is conflicted with the current Python Config.
[2023-01-25 12:52:05,692][Taipy][WARNING] Overriding version 1.0 ...
[2023-01-25 12:52:05,890][Taipy][INFO] job JOB_my_print_algo_96ed74ed-183b-4dff-86c9-3b733d4d9bd9 is completed.
nb scenarios: 2

As you can see on the previous example, the application run correctly after updating the configuration. A new scenario has been created.

Warning

By forcing the configuration update, you must be aware that old entities instantiated before the configuration change may not be compatible.

On the previous example, note that two scenarios are attached to version 1.0. Similarly, two output data nodes as well. One has been instantiated before the configuration update and one after. That means the description property only exists for one of the two output data nodes. It is your responsibility to handle the changes.

Hint: you can migrate your old entities so they become compatible with the new configuration or you can ensure your code is compatible with both versions of data nodes.