Skip to content

Latest commit

 

History

History
196 lines (123 loc) · 8.2 KB

webapp_exp_tuning.md

File metadata and controls

196 lines (123 loc) · 8.2 KB
title
Tuning Experiments

Tune experiments and edit an experiment's execution details, then execute the tuned experiments on local or remote machines.

To Tune an Experiment and Execute it Remotely:

  1. Locate the experiment. Open the experiment's Project page from the Dashboard or the main Projects page.

    • On the Dashboard,
      • Click on an experiment from RECENT EXPERIMENTS
      • In RECENT PROJECTS > click on a project card > click experiment
      • In RECENT PROJECTS > click VIEW ALL > click the project card > click experiment
    • On the Projects page, click project card, or the All projects card > click experiment

  2. Clone the experiment. In the experiments table:

    1. Click Clone, and a Clone experiment box will pop up.
    2. In the Project textbox, select or create a project. To search for another project, start typing the project name. To create a new project, type new experiment name and click Create New.
    3. Enter an optional description.
    4. Click CLONE.

    The cloned experiment's status is now Draft.

  3. Edit the experiment. See modifying experiments.

  4. Enqueue the experiment for execution. Right-click the experiment > Enqueue > Select a queue > ENQUEUE.

    The experiment's status becomes Pending. When the worker assigned to the queue fetches the Task (experiment), the status becomes Running. The experiment can now be tracked and its results visualized.

Modifying Experiments

Experiments whose status is Draft are editable (see the user properties exception). In the ClearML Web UI, edit any of the following

:::note User parameters are editable in any experiment, except experiments whose status is Published (read-only). :::

Execution Details

Source Code

Modify code execution by changing any of the following:

  • Repository, commit (select by ID, tag name, or choose the last commit in the branch), script, working directory, and/or binary.
  • The Python packages to be installed and/or their versions - Edit the package list, or clear it to have the ClearML Agent either not install any packages or use an existing repo requirements.txt file. If the experiment is based on a run in which the packages used were eventually different to the ones originally specified, you can easily or reset the packages to originally recorded values ("Original Pip").
  • Uncommitted changes - Edit or clear all.

To modify the source code, hover over the relevant sections in the EXECUTION tab to access Edit, Clear/Discard, and/or Reset functions.

Base Docker Image

Select a pre-configured Docker that ClearML Agent will use to remotely execute this experiment (see Building Docker containers).

To add, change, or delete a base Docker image:

  • In EXECUTION > AGENT CONFIGURATION > BASE DOCKER IMAGE > hover > EDIT > Enter the base Docker image.

:::important For a ClearML Agent to execute the task in a container, the agent must be running in Docker mode:

clearml-agent daemon --queue <execution_queue_to_pull_from> --docker [optional default docker image to use]

For more information, see Docker Mode. :::

Output Destination

Set an output destination for model checkpoints (snapshots) and other artifacts. Examples of supported types of destinations and formats for specifying locations include:

  • A shared folder: /mnt/share/folder
  • S3: s3://bucket/folder
  • Non-AWS S3-like services (e.g. MinIO): s3://host_addr:port/bucket
  • Google Cloud Storage: gs://bucket-name/folder
  • Azure Storage: azure://<account name>.blob.core.windows.net/path/to/file

To add, change, or delete an artifact output destination:

  • In EXECUTION > OUTPUT > DESTINATION > hover > EDIT > edit > SAVE.

:::note Set Output Destination for Artifacts Also set the output destination for artifacts in code (see the output_uri parameter of the Task.init method), and in the ClearML configuration file for all experiments (see default_output_uri on the ClearML Configuration Reference page). :::

Configuration

Hyperparameters

Add, change, or delete hyperparameters, which are organized in the ClearML Web UI in the following sections:

  • Args - Automatically logged argument parser parameters (e.g. argparse, click, hydra).

  • TF_DEFINE - TensorFlow definitions (from code, TF_DEFINEs automatic logging).

  • General - Parameter dictionaries (from code, connected to the Task by calling Task.connect()).

  • Environment variables - Tracked if variables were listed in the CLEARML_LOG_ENVIRONMENT environment variable or the sdk.development.log_os_environments field of the clearml.conf file (see this FAQ).

  • Custom named parameter groups (see the name parameter in Task.connect).

To add, change, or delete hyperparameters:

  • In the CONFIGURATION tab > HYPERPARAMETERS > parameter group > hover > click EDIT > add, change, or delete keys and/or values > click SAVE.

User Properties

User properties allow storing any descriptive information in key-value pair format. They are editable in any experiment, except experiments whose status is Published (read-only).

To add, change, or delete user properties:

  • In CONFIGURATION > USER PROPERTIES > Properties > hover > click EDIT > add, change, or delete keys and/or values > click SAVE.

Configuration Objects

To add, change, or delete the Task model configurations:

  • In CONFIGURATION > CONFIGURATION OBJECTS > object name > hover > EDIT or CLEAR (if the configuration is not empty).

Artifacts

Initial Weights Input Model

Edit model configuration and label enumeration, choose a different initial input weight model for the same project or any other project, or remove the model.

:::note The models are editable in the MODELS tab, not the EXPERIMENTS tab. Clicking the model name hyperlink shows the model in the MODELS tab. :::

To select a different model:

  1. In ARTIFACTS > Input Model > Hover and click EDIT.
  2. If a model is associated with the experiment, click Edit Pencil.
  3. In the SELECT MODEL dialog, select a model from the current project or any other project.

To edit a model's configuration or label enumeration:

  1. Click the model name hyperlink. The model details appear in the MODELS tab.

  2. Edit the model configuration or label enumeration:

    • Model configuration - In the NETWORK tab > Hover and click EDIT. > CLick EDIT or CLEAR (to remove the configuration).

      Users can also search for the configuration (hover over the configuration textbox, the search box appears) and copy the configuration to the clipboard (hover and click Copy Clipboard).

    • Label enumeration - In the LABELS tab > Hover and click EDIT > Add, change, or delete label enumeration key-value pairs.

To remove a model from an experiment:

  • Hover and click EDIT > Click Trash