Skip to content

Configure the Comet SDK

When getting started with Comet experiment tracking, you need to make sure your API key is correctly configured and you may also want to make sure that you have customized Comet exactly for the kind of experiment logging you want.

Set the API key

You should use the init() method to provide your API key to Comet, either from command line or in your Python code:

1
comet init

With this method run, Comet also prompts you to customize the creation of an example script, on properties such as framework and information to log, that you can use as a learning ground. To skip this step, run the command as comet init --api-key instead. Find more details in the Command Line Utilities docs.

1
2
3
import comet_ml

comet_ml.init()

With either approach, the init() method interactively prompts you for your API key which, once provided, is automatically accessed from your ~/.comet.config configuration file.

Tip

If you would like to log data to the Comet platform but don't want to create an account, you can use comet_ml.init(anonymous=True). The data will be logged to a public Comet project which you can then claim if you wish.
Please don't use this modality with proprietary or other sensitive data.

Warning

While it is possible to pass your API key using comet init --api_key="your_api_key or comet_ml.init(api_key="your_api_key"), it is not recommended since code is automatically logged by Comet and you could accidentally commit your API key to Git too.

Configure Comet logging

Comet has a wide range of configuration options allowing you to enable and disable features in a very granular way.

You can configure the comet_ml Python SDK in three ways:

  1. From the .comet.config file: You can directly edit the configuration file, which is automatically created for you when running init() as detailed in the section above.
    This is the recommended approach.
  2. With Comet environment variables: You can define the Comet configuration as environment variables from terminal or in your Python code directly with the os functionalities.
    Consider this approach when it is not convenient to create and store a configuration file.
  3. For the Experiment: You can specify the configuration when creating a Comet Experiment through the experiment.init() method.
    Consider this approach when wanting to enable or disable features in-place for the single Experiment only.

From the .comet.config file

The ~/.comet.config configuration file follows the format:

[section_name]
config_name=value

So if you wanted to add your API key to your configuration file, you could set it using:

[comet]
api_key="your_api_key"

Find the list of Comet configuration names in the Explore the configuration options section.

With Comet environment variables

You can set environment variables to update the Comet configuration temporarily for the current session or permanently for all sessions.

Run export YOUR_VARIABLE_NAME="your_value" from your terminal.

1. Open one of .bashrc, .bash_profile, or .profile configuration files depending on your set-up.

2. Add this line to the file: export YOUR_VARIABLE_NAME="your_value".

3. Save and execute the file, e.g. with source ~/.bashrc.

For more information and instructions, read more about setting environment variables in Python.

Comet environment variables are capitalized, and follow the format:

COMET_<parameter-name>
For example, you could set the Comet workspace by running export COMET_WORKSPACE="your_workspace".

Find the list of Comet environment parameter names in the Explore the configuration options section.

For the Experiment

The Experiment.init() method supports a wide set of arguments that allow you to customize the automated logging performed by Comet.

Many of the configuration options described in the Explore Comet configuration options section below can be set with this method. Please refer to the linked reference for a complete list of supported configuration options.

Explore Comet configuration options

When first setting up Comet for your project / workspace / team, you may find yourself choosing how Comet automatically logs information for you and what information it should log.

The options you are likely to update more frequently are listed at the top and highlighted in bold.

Comet Config Section/NameEnvironment Parameter Name
Description
[comet]
api_key
COMET_API_KEYThe API key used for creating Experiments and using the Python API
[comet]
workspace
COMET_WORKSPACEUse this workspace when creating new experiments
[comet]
project_name
COMET_PROJECT_NAMEUse this project when creating new experiments
[comet]
url_override
COMET_URL_OVERRIDEURL to upload the experiments to, this needs to be updated for on-premise deployments. Defaults to https://www.comet.com/clientlib/.
[comet_logging]
file
COMET_LOGGING_FILEUse the given file for storing log messages. This can also contain patterns such as "comet-{project}.log" (See below)
[comet_logging]
file_level
COMET_LOGGING_FILE_LEVELBy default, the log outputs will contains INFO and higher level (WARNING and ERROR) log messages. This configuration item can be used to change the level of the logged messages. Defaults to INFO.
[comet]
offline_directory
COMET_OFFLINE_DIRECTORYSet the offline directory for OfflineExperiment(). Defaults to .cometml-runs.
[comet]
git_directory
COMET_GIT_DIRECTORYSet the directory that contains your git repo, if you are not under it
[comet]
display_summary_level
COMET_DISPLAY_SUMMARY_LEVELControls the level of detail displayed in the experiment summary. If set to 0, the summary is hidden, but the summary notification is still sent. If set to 2, the experiment summary shows everything that has been logged. Defaults to 1.
[comet]
rich_output
COMET_RICH_OUTPUTWhen enabled, Comet console log messages will be rendered using formatted methods from the Rich library. Defaults to True.
[comet]
system_cpu_interval
COMET_SYSTEM_CPU_INTERVALControls the interval at which system CPU metrics (CPU utilization and memory utilization) are logged. The value must be in seconds and cannot be less than 1s. Defaults to 60.
[comet]
system_gpu_interval
COMET_SYSTEM_GPU_INTERVALControls the interval at which system GPU metrics are logged. The value must be in seconds and cannot be less than 1s. Defaults to 60.
[comet]
system_network_interval
COMET_SYSTEM_NETWORK_INTERVALControls the interval at which system network metrics are logged. The value must be in seconds and cannot be less than 1s. Defaults to 60.
[comet]
system_disk_interval
COMET_SYSTEM_DISK_INTERVALControls the interval at which system disk metrics are logged. The value must be in seconds and cannot be less than 1s. Defaults to 60.
[comet]
http_session.retry_total
COMET_HTTP_SESSION_RETRY_TOTALNumber of times the SDK will try to upload a message before triggering the fallback to offline mechanism. Defaults to 3.
[comet]
http_session.retry_backoff_factor
COMET_HTTP_SESSION_RETRY_BACKOFF_FACTORControls how much time is applied between two different attempts to send data to Comet. Comet will wait for backoff_factor * (2 ** (attempt_number - 1)) seconds between attemps. Defaults to 2.
[comet_fallback_streamer]
keep_offline_zip
COMET_FALLBACK_STREAMER_KEEP_OFFLINE_ZIPIf True, save a local copy of experiment data for all training runs, not the just the ones that encountered connectivity issues. Defaults to False.
[comet_logging]
hide_api_key
COMET_LOGGING_HIDE_API_KEYIf True, redact the Comet API key in logs. Defaults to True.
[comet_logging]
file_overwrite
COMET_LOGGING_FILE_OVERWRITEOverwrites the log file on each run, if True. Defaults to False.
[comet_logging]
console
COMET_LOGGING_CONSOLESet the logging level for console messages (for example, INFO, DEBUG).
[comet_logging]
metrics_ignore
COMET_LOGGING_METRICS_IGNOREList of framework:metric names to not log automatically. Defaults to keras:batch_size,keras:batch_batch
[comet_logging]
parameters_ignore
COMET_LOGGING_PARAMETERS_IGNOREList of framework:parameter names to not log automatically. Defaults to keras:verbose,keras:do_validation,keras:validation_steps
[comet_logging]
others_ignore
COMET_LOGGING_OTHERS_IGNOREList of framework:other names do not log automatically.
[comet_timeout]
cleaning
COMET_TIMEOUT_CLEANINGHow many times (in seconds) does the Python SDK wait until all experiment data has been uploaded. Default is 3600 seconds / 1 hour.
[comet_auto_log]
code
COMET_AUTO_LOG_CODEAuto Log source code for the current Experiment Run. Defaults to True.
[comet_auto_log]
git_metadata
COMET_AUTO_LOG_GIT_METADATAAuto Log git metadata for the current Experiment Run. Defaults to True.
[comet_auto_log]
git_patch
COMET_AUTO_LOG_GIT_PATCHAuto Log the git patch for current Experiment Run. Set to True to enable.
[comet_auto_log]
graph
COMET_AUTO_LOG_GRAPHAuto Log the model graph. This will depend on the framework being used. Set to True to enable.
[comet_auto_log]
parameters
COMET_AUTO_LOG_PARAMETERSAuto Log the model parameters. This will depend on the framework being used. Set to True to enable.
[comet_auto_log]
weights
COMET_AUTO_LOG_WEIGHTSAuto Log the model weights. This will depend on the framework being used. Set to True to enable.
[comet_auto_log]
metrics
COMET_AUTO_LOG_METRICSAuto Log the model metrics. This will depend on the framework being used. Set to True to enable.
[comet_auto_log]
cli_arguments
COMET_AUTO_LOG_CLI_ARGUMENTSAuto Log CLI arguments passed to the training script. Set to True to enable.
[comet_auto_log]
output_logger
COMET_AUTO_LOG_OUTPUT_LOGGERAuto Log console and Jupyter Notebook standard output and error. Set to True to enable.
[comet_auto_log]
co2
COMET_AUTO_LOG_CO2Auto Log CO2 emissions from the run using the code-carbon package. Set to True to enable.
[comet_auto_log]
env_details
COMET_AUTO_LOG_ENV_DETAILSAuto Log details about the compute environment. Set to True to enable, set to False to quickly disable all environment details logging.
[comet_auto_log]
env_gpu
COMET_AUTO_LOG_ENV_GPUAuto Log the collection of GPU details and system metrics (utilization, memory usage etc..). Logging Environment Details must also be set to True.
[comet_auto_log]
env_host
COMET_AUTO_LOG_ENV_HOSTAuto Log the host information (ip, host name, python version, user etc...). Logging Environment Details must also be set to True.
[comet_auto_log]
env_cpu
COMET_AUTO_LOG_ENV_CPUAuto Log the collection of CPU details and system metrics (utilization, memory usage etc..). Logging Environment Details must also be set to True.
[comet_auto_log]
env_cpu_per_core
COMET_AUTO_LOG_ENV_CPU_PER_COREAuto Log the collection of CPU details and system metrics (utilization etc..) for each CPU core. Logging Environment Details must also be set to True. Defaults to False.
[comet_auto_log]
env_network
COMET_AUTO_LOG_ENV_NETWORKAuto Log the collection of network system metrics. Logging Environment Details must also be set to True. Defaults to True.
[comet_auto_log]
env_disk
COMET_AUTO_LOG_ENV_DISKAuto Log the collection of disk system metrics. Logging Environment Details must also be set to True. Defaults to True.
[comet_auto_log]
disable
COMET_AUTO_LOG_DISABLEDisable all network communication with the Comet backend. It is useful when you want to test to make sure everything is working, without actually logging anything. Defaults to False.

In addition, the following configurations are available only as environment variables:

Environment ParameterDescription
COMET_INIA path to another .comet.config file to take into account
COMET_OPTIMIZER_IDCurrent optimizer ID; for use with the comet optimize CLI command
COMET_OPTIMIZER_PROCESS_IDCurrent job number; set when using comet optimize CLI command
COMET_OPTIMIZER_PROCESS_JOBSTotal number of parallel jobs; set when using comet optimize CLI command
COMET_DISABLE_AUTO_LOGGINGDisable all automatic logging (see next section)
May. 17, 2024