Changelog¶
All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog.
[1.11.1.post1] - 2024-11-14¶
Fixed¶
- Connecting to different DAI server versions in the same Python process.
[1.11.1] - 2024-09-25¶
Added¶
- Support for Driverless AI 1.11.1.
- Support for following in the admin API.
- List and delete model diagnostics of a user.
- List and delete projects of a user.
- List and delete visualizations of a user.
- List and delete interpretations of a user.
- List and abort running tasks of a user.
[1.11.0] - 2024-08-10¶
Added¶
- Support for listing and downloading previous predictions from experiments.
- Support for tagging datasets linked to projects.
Fixed¶
- Validations to prevent uploading unsupported file types to create datasets.
- Returning duplicated usernames from
Admin.list_current_users()
.
[1.10.7.3] - 2024-11-13¶
Added¶
- Support for Driverless AI 1.10.7.3.
Fixed¶
- Connecting to different DAI server versions in the same Python process.
[1.10.7.2] - 2024-09-19¶
Added¶
- Support for Driverless AI 1.10.7.2.
Fixed¶
RecursionError: maximum recursion depth exceeded
error when instantiatingClient
objects in a loop.- Set the log level to INFO only if a default log level is not set.
[1.10.7.1] - 2024-08-16¶
Added¶
- Support for Driverless AI 1.10.7.1.
[1.10.7] - 2024-01-23¶
Added¶
- Support for Python 3.12.
- Support to compare the settings and setups of an experiment with another experiment.
- Support for getting all the linked projects of an experiment.
- Support for importing experiments from external file systems.
- Support for adding and updating the tags of an experiment that is linked to a project.
- Retrieve and list Interpretations from
MLI
API. - Get summary of parameters of an interpretation.
- Get AutoViz visualizations, datasets, experiments, and projects by their name.
- Support for adding a description to a dataset.
columns
,creation_timestamp
,data_source
,description
,file_path
,file_size
, andshape
properties for datasets in the Admin API.- Support for using additional configured authentication methods to connect to Driverless AI.
- Improve docs content and the examples.
Fixed¶
- Bugs related to retrieving the variable importance of last iteration of an experiment using the
Experiment.variable_importance
method. - Unclear or missing information in docstrings.
- Bugs related to MLI Plots and Explainers on standalone model.
- Retry if a server call gets redirected to the Steam login page.
Removed¶
- Support for Python 3.7.
- Support for Driverless AI 1.9.x versions.
TimeseriesMethods
fromMLI
.
Deprecated¶
- Support for
IIDMethods
inMLI
. UseMLI.get()
andMLI.list()
instead.
Changed¶
- The
client.server.version
property now returns an instance of_server.Version
, which is compatible withstr
. - The
Interpretation.explanation_plots
property now returns a mapping ofExplanationPlot
s. - Removed
Experiment.summary()
method, addedExperiment.summary
property that returns astr
.
[1.10.6.3] - 2024-07-22¶
Added¶
- Support for Driverless AI 1.10.6.3.
- Support to set the connection timeout duration when downloading artifacts from the server.
[1.10.6.2] - 2023-11-01¶
Added¶
- Support for Driverless AI 1.10.6.2.
- Retrying when Driverless AI instances in H2O Steam are temporarily unavailable.
Deprecated¶
- Deprecated support for Python 3.7.
- Deprecated the
Experiment.summary()
method. TheExperiment.summary
property will be introduced in the 1.10.7 release.
[1.10.6.1] - 2023-09-04¶
Added¶
- Support for Driverless AI 1.10.6.1.
- Set the default log level to
INFO
.
Fixed¶
- Log entries being duplicated if multiple
Client
objects are instantiated in the same script or notebook. - By default, logs being sent to
sys.stderr
instead ofsys.stdout
. - Print output of
experiments.preview()
no longer hasINFO
in all lines.
[1.10.6] - 2023-08-28¶
Added¶
- Support for Python 3.11.
- Support for configuring the
logger
for thedriverlessai
package. - AutoDoc API for creating and downloading AutoDocs.
- Deployments API for interacting with deployments in the DAI server.
- Support for model deployments in NVIDIA Triton inference server.
- Support to access
actual_vs_predicted_chart
,gains_chart
,ks_chart
,lift_chart
,prec_recall_curve
,residual_histogram
,residual_plot
,roc_curve
, andconfusion_matrix
plots of a Model Diagnostics. - Support to access
partial_dependence_plot
,shapley_summary_plot_for_original_features
,shapley_values_for_original_features
,shapley_values_for_transformed_features
, andsurrogate_decision_tree
plot of an MLI explanation. - Support to get the CPU, memory, disk, GPU usages and stats of worker nodes of the DAI server.
- Options to show the description, name in DAI, and valid values of settings when searching setting of an MLI explanation.
- Admin:
- Support to list and access DAI server logs.
- Support to access creation timestamp, datasets, run duration, size, and setting of experiments.
- AutoViz visualizations:
- Heatmaps, outliers.
- Support to download the logs of a visualization.
- Datasets:
- Directly accessing the details webpage of a dataset.
- Summarize a dataset using OpenAI GPT.
- Option to pass a
DataRecipe
tomodify_by_recipe
. - API to get all experiments a dataset has been used as a train, test, or validation dataset.
- Experiments:
- Support to pass a Pandas DataFrame as a dataset to predict on it.
- Support to access
actual_vs_predicted_chart
,gains_chart
,ks_chart
,lift_chart
,prec_recall_curve
,residual_histogram
,residual_plot
,roc_curve
, andconfusion_matrix
plots. - Options to get the variable importance of a given iteration and of a given model index.
- Support to access the generated AutoDoc.
- Support to export the model as an NVIDIA Triton model.
- Support to create a fit and transformation.
- Recipes:
- Option to pass a BitBucket URL to create a custom recipe.
- Support for pre-transformer recipes.
- Support to activate, deactivate, and check the active state of custom recipes.
- Access the code and update the code of a custom recipe.
- Documentation:
- Example on interacting with the plots of a Model Diagnostics.
- Examples on how to connect to DAI instances in H2O AI Cloud.
- Example on interacting with AutoDocs.
- Examples on running a specific explainer, using custom explainers, and accessing explainer data of an MLI explanation.
Fixed¶
- Detection of image columns in the train dataset when creating an experiment.
AttributeError: 'ModelDiagnosticJob' object has no attribute 'key'
error when listing Model Diagnostics.- MLI artifact downloading
Removed¶
- Support for Driverless AI 1.8.x versions.
Deprecated¶
- Deprecate support for Driverless AI 1.9.x versions.
[1.10.5.1] - 2023-08-15¶
Added¶
- Support for Driverless AI 1.10.5.1.
[1.10.5.post1] - 2023-05-03¶
Fixed¶
- Fix
RuntimeError: The Driverless AI server does not provide a universally unique identifier for this object type.
error when getting an experimentclient.experiments.get(key)
. - Declare
urllib3
as a direct dependency and set the minimum required version to v1.26.0.
[1.10.5] - 2023-04-25¶
Added¶
- Support for Driverless AI 1.10.5.
- Improve support for Autoviz visualizations
- APIs to access plots and graphs of a visualization
- Option to add custom plots to a visualization.
- Support for running model diagnostics, including download predictions.
- Add support to retrieve explainers from an MLI interpretation.
- APIs to interact with
Explainer
s, explainer frames, explainer data, and artifacts. - Add admin API that facilitates performing administration tasks.
- Support to transform dataset from using a trained model.
- API to interact with the
Transformation
and download transformed data. - Introduce Beta APIs.
- Option to pass Azure Blob Storage credentials
azure_blob_account_name
,azure_blob_account_key
,azure_connection_string
when importing dataset using the Azure Blob Storage connector. - Option to pass Minio credentials
minio_access_key_id
,minio_secret_access_key
when importing dataset using the Minio connector. - Documentation:
- Add examples on using the Autoviz visualization API.
- Add examples on using the MLI
Explainer
API. - Add example on using the Admin API.
- Add example on transforming a dataset using a trained model.
- Add example on importing datasets using the Minio connector.
- Add information about the H2O Feature Store connector support.
Fixed¶
experiment.retrain_async(final_models_only=True)
resulting DAI server error.
Removed¶
- Support for Python 3.6.
Deprecated¶
- Deprecate support for Driverless AI 1.8.x versions.
[1.10.4.3] - 2023-01-30¶
Added¶
- Support for Driverless AI 1.10.4.3.
[1.10.4.2] - 2023-01-16¶
Added¶
- Support for Driverless AI 1.10.4.2.
- Option to pass AWS credentials
aws_access_key_id
,aws_secret_access_key
when importing dataset using the AWS S3 connector. - Documentation:
- Add example on importing datasets using the AWS S3 connector.
- Add example on importing a folder as a dataset.
[1.10.4.1] - 2022-12-15¶
Added¶
- Support for Driverless AI 1.10.4.1.
- Option to delete linked experiments when deleting a project.
- Documentation:
- Add example on running an unsupervised experiment.
[1.10.4.post1] - 2022-11-01¶
Fixed¶
TypeError: __init__() got an unexpected keyword argument 'allowed_methods'
error.
[1.10.4] - 2022-10-25¶
Added¶
- Support for Driverless AI 1.10.4.
- Support for data recipes and individual recipes.
- H2O Feature Store connector support.
- Support for Python 3.10.
- Retry upon HTTP connection error.
- Documentation:
- Add example on using the Google Big Query connector to import a dataset
- Add example on using the H2O Drive connector to import a dataset
Fixed¶
- Better handling using unsupported connectors
- Documentation improvements
- Add validation for
bool
type parameters in experiment creation - Improve error handling connection problems
- Improve error handling on experiment prediction job failures
Deprecated¶
- Deprecate support for Python 3.6.
[1.10.3.1] - 2022-06-16¶
Added¶
- Support for Driverless AI 1.10.3.1.
[1.10.3] - 2022-05-03¶
Added¶
- Support for Driverless AI 1.10.3.
- H2O Drive connector support.
- Capability to change the description of a project.
[1.10.2] - 2022-02-22¶
Added¶
- Support for Driverless AI 1.10.2.
- Support for Driverless AI 1.9.3.3.
- Support for project sharing.
Fixed¶
- Experiments and datasets not importing from a shared project when H2O.ai Storage is used.
[1.10.1.3] - 2022-01-10¶
Added¶
- Support for Driverless AI 1.10.1.3.
[1.10.1.2] - 2021-12-23¶
Added¶
- Support for Driverless AI 1.10.1.2.
Fixed¶
- Setting name for
models
andtransformers
returned incorrectly by experimentsettings
attribute.
[1.10.1.1] - 2021-12-15¶
Added¶
- Support for Driverless AI 1.10.1.1.
[1.10.1.post1] - 2021-12-06¶
Added¶
- Support for Driverless AI 1.9.3.2.
Fixed¶
- Empty
username
server attribute when logging in with tokens.
[1.10.1] - 2021-11-12¶
Added¶
- Support for Driverless AI 1.10.1.
- Experiment export/import.
size
property forExperiment
objects.leaderboard
method for creating experiments.force
parameter toprojects
create
method.final_models_only
parameter toExperiment
object retrain methods.- Optionally specify location/region for GBQ connector.
Changed¶
- Project
create
method throws error if project with same name already exists. Can be overridden withforce
argument.
Fixed¶
Project
objectgui()
hyperlink.
[1.10.0] - 2021-09-30¶
Added¶
- Support for Driverless AI 1.10.0.
to_pandas
method for prediction objects.file_paths
property for artifacts.
Changed¶
- Improved documentation layout, search results, and examples.
Fixed¶
- When getting SHAP values with predictions, some characters would be decoded incorrectly.
- Return type of
PredictionJobs
result
method. get
methods not throwing exception if used to retrieve a non-existent entity.
[1.9.3.1] - 2021-08-12¶
Added¶
- Support for Driverless AI 1.9.3.1.
- Creation timestamp property for some server objects.
- Run duration property for some server objects.
- Ability to set 'image' dataset column logical type.
- Dataset type argument for link/unlink project methods can be plural.
Fixed¶
- Not being able to pass prediction, weight, or drop column arguments to MLI create methods.
[1.9.3] - 2021-06-02¶
Added¶
- Support for Driverless AI 1.9.3.
[1.9.2.1.post1] - 2021-05-05¶
Added¶
- Support for MLI explainer recipes.
- Misc. documentation updates.
Changed¶
- Disabled interpretation settings property for servers >= 1.9.1.
- References to
autoreport
changed toautodoc
. Code that usesautoreport
will still work.
Fixed¶
- Server objects referencing self after being deleted from the server.
- Error if non-numeric column has no histogram ticks.
[1.9.2.1] - 2021-04-05¶
Added¶
- Support for Driverless AI 1.9.2.1.
- Support for AutoViz.
- Detailed HTTP errors for
OAuth2TokenProvider
.
Changed¶
GUILink
class renamed toHyperlink
.
[1.9.1.1] - 2021-02-24¶
Added¶
- Support for Driverless AI 1.9.1.1.
- Support for Driverless AI 1.8.10.
- Ability to include columns when downloading experiment artifact predictions.
- Server object list methods display as a table.
- Server object list methods return a sequence that accepts multiple indices.
Changed¶
- Recipe list methods return
ServerObjectList
.
[1.9.1.post1] - 2021-01-25¶
Added¶
- Ability to pass FSSPEC based file system object to all download methods.
[1.9.1] - 2021-01-19¶
Added¶
- Support for Driverless AI 1.9.1.
- Experiment
predict
methods acceptenable_mojo
,include_shap_values_for_original_features
,include_shap_values_for_transformed_features
,use_fast_approx_for_shap_values
arguments (some of these arguments only apply for Driverless AI server versions >= 1.9.1, see documentation for more details). - Project
link_dataset
method acceptslink_associated_experiments
argument (only applies for Driverless AI server versions >= 1.9.1). gui
method for interpretations.
Changed¶
include_shap_values
parameter removed from experimentpredict
methods (useinclude_shap_values_for_transformed_features
instead).- If Driverless AI server license is invalid, client will error upon initialization.
[1.9.0.6] - 2021-01-06¶
Added¶
- Support for Driverless AI 1.9.0.6.
- MLI
create
methods accept dataset argument. - MLI settings can be passed to MLI
create
methods as keyword arguments. search_expert_settings
method for MLI.- MLI
create
methods check if interpretation name already exists on server.
[1.9.0.5] - 2020-12-21¶
Added¶
- Support for Driverless AI 1.9.0.5.
Changed¶
- All attributes converted to properties.
[1.9.0.4.post2] - 2020-12-02¶
Added¶
- Support for Projects.
Changed¶
- Download timeout increased to 30 seconds.
Fixed¶
object has no attribute '_ServerJob___info'
error.PermissionError: [Errno 13]
error for Pandas DataFrame uploads on Windows.- Pandas DataFrames created with Pandas > 1.0.5 fail to import.
- Server version detected as 8 when wrong address is used.
[1.9.0.4.post1] - 2020-11-16¶
Added¶
is_deprecated
property to experiment objects.file_size
attribute to dataset objects.- Support for exporting experiment artifacts.
- Support for exporting datasets.
- Speed up for initial view of server object
list()
method by lazy loading server objects. - Speed up for iterating over server object list by only initializing server object properties if they're accessed.
- Conda install instructions to documentation.
- Cleaned up return annotations for documentation.
Changed¶
- Server object
list()
methods now return immutable sequences. If you want the old behavior, you can use a list comprehension. For example:[ex for ex in dai.experiments.list()]
. - Some server object attributes are now properties.
Fixed¶
- Experiment summary not displaying for 1.9 servers.
" Model"
incorrectly being appended to custom model names.- Experiment preview error when using
force
orname
argument. HTTPError: 500 Server Error: Internal Server Error for url
when downloading dataset from server with address that contains a path (likehttps://enterprise-puddle.h2o.ai/dab455b5-409e-4e3a-8dc9-4512c0ad3669
)
[1.9.0.4] - 2020-10-23¶
Added¶
- Support for Driverless AI 1.9.0.4.
- Support for Driverless AI 1.8.9.
Fixed¶
IndexError: list index out of range
error when column had allNaN
values.- Missing
unavailable_at_prediction_time_columns
keyword parameter for experiment creation/preview.
[1.9.0.3] - 2020-10-09¶
Added¶
- Support for Driverless AI 1.9.0.3.
- Support for Driverless AI 1.8.8.
- Examples to docstrings/documentation for recipe methods.
- Misc. documentation updates.
file_paths
attribute for prediction objects.
Changed¶
toml
is now a requirement.
Fixed¶
- Unclear or missing information in docstrings based on user feedback.
- Passing a dictionary as an argument for an experiment expert setting would sometimes fail.
[1.9.0.2.post2] - 2020-09-28¶
Added¶
- Ability to name downloaded files with
dst_file
parameter in non-artifact download methods. - Predictions section to documentation.
Fixed¶
- Download failure when using token based authentication.
[1.9.0.2.post1] - 2020-09-24¶
Added¶
- Token based authentication for Driverless AI servers that support it (server versions >= 1.9.0).
- OAuth 2.0 token provider.
[1.9.0.2] - 2020-09-17¶
Added¶
- Support for Driverless AI 1.9.0.2.
- Support for MLI.
reproducible
parameter for experiment creation.
Fixed¶
- Unclear or missing information in docstrings based on user feedback.
- Behavior if user tried to download artifact that didn't exist on server.
- Rare
RemoteError: {'code': 0, 'message': "KeyError: 'Kind.DATASET.None'"}
when using experimentget
method.
[1.9.0.1] - 2020-08-14¶
Added¶
- Support for Driverless AI 1.9.0.1.
- Support for Driverless AI 1.9.0.
- Support for Driverless AI 1.8.7.3.
- Requirements to documentation.
- Examples to docstrings/documentation for core, server, and datasets methods.
Changed¶
- Experiment notifications are now a list of dictionaries instead of a list of strings.
Fixed¶
backend_version_override="latest"
argument forClient
initialization didn't work if client version number containedpost
.- Pandas DataFrame upload didn't throw error when dataset with same name already existed on the server.
- Job progress could show less than 100% when job was complete.
- When server was configured with
authentication_method=none
, server version detection failed.
[1.8.7.2.post1] - 2020-07-17¶
Fixed¶
- PyPI description.
[1.8.7.2] - 2020-07-17¶
Added¶
- Support for Driverless AI 1.8.7.2.
- Hive connector.
- More connector documentation.
- Misc. documentation updates.
Fixed¶
- Google Big Query connector.
- kdb connector for server versions 1.8.6 and up.
- Snowflake connector.
[1.8.7.1-beta-0] - 2020-07-06¶
Added¶
- Support for Driverless AI 1.8.7.1.
- Changelog to documentation.
Changed¶
create
methods throw error if dataset/experiment with same name already exists. Can be overridden withforce
argument.
Fixed¶
- JDBC connector for server versions 1.8.6 and up.
is_server_up
detection of http connection errors.- Misc. documentation improvements.
[1.8.7-beta-0] - 2020-06-23¶
Added¶
- Support for Driverless AI 1.8.7.
- Get SHAP/raw values from experiment predict method.
[1.8.6.3-beta-1] - 2020-06-08¶
Added¶
- Example notebooks can be downloaded from documentation.
- Misc. documentation updates.
Fixed¶
- Uploading Pandas DataFrames fails with Python 3.8.
- Experiment log head/tail charmap error on Windows.
[1.8.6.3-beta-0] - 2020-05-29¶
Added¶
- Support for Driverless AI 1.8.6, 1.8.6.1, 1.8.6.2, 1.8.6.3.
- Ability to upload Pandas DataFrame.
Changed¶
- Client constructor parameter
server_version
replaced withbackend_version_override
.backend_version_override
specifies the version of client backend to use and overrides Driverless AI server version detection. Specify"latest"
to get the most recent backend supported. In most cases the user should rely on Driverless AI server version detection and leave this as the defaultNone
.
Fixed¶
- Documentation typos.
- Column summary error for datetime columns.
- Server detection when connecting to a server that uses https.
[1.8.5.1-beta-0] - 2020-04-10¶
Added¶
- Support for Driverless AI 1.8.5 and 1.8.5.1.
- Import data with recipe.
modify_by_code
,modify_by_code_preview
,modify_by_recipe
methods to datasets objects.server
attribute to client to get information about the Driverless AI server connected to.silent
parameter to allresult
methods.
Changed¶
- Connect to Driverless AI server from
Client
constructor.connect
anddisconnect
methods are removed. is_server_up
moved todriverlessai
namespace.- Server object
__repr__
displays the object name last to make lists of objects easier to visually parse. gui
methods return aGUILink
object that behaves likestr
except displays as a clickable link in notebooks.
Fixed¶
- Wait for data splits to complete before displaying 'Complete'.
- All expert setting parameter names retrievable from
search_expert_settings
can be used to create experiments. - Misc. optimizations and fixes.
[1.8.4.1-beta-1] - 2020-03-17¶
Added¶
gui()
method for experiments page.- Concepts section added to documentation.
Changed¶
- All parameters that ended with
_col(s)
were renamed to end with_column(s)
.
Fixed¶
- Dataset list not showing all datasets when the server has inactive datasets on it.
- Compatibility with server version 1.8.2.
[1.8.4.1-beta-0] - 2020-03-06¶
Initial beta release.