Server Objects

Server objects represent an entity that exists on the Driverless AI server.

Dataset

Dataset objects correspond to existing datasets on a Driverless AI server. Dataset objects are retrievable using the Client.

API Reference:

class Dataset

Interact with a dataset on the Driverless AI server.

Examples:

# Import the iris dataset
ds = dai.datasets.create(
        data='s3://h2o-public-test-data/smalldata/iris/iris.csv',
        data_source='s3'
)

ds.columns
ds.data_source
ds.file_path
ds.file_size
ds.key
ds.name
ds.shape
column_summaries(columns: Optional[List[str]] = None) → DatasetColumnSummaryCollection

Returns a collection of column summaries.

The collection can be indexed by number or column name:

  • dataset.column_summaries()[0]
  • dataset.column_summaries()[0:3]
  • dataset.column_summaries()['C1']

A column summary has the following attributes:

  • count: count of non-missing values
  • data_type: raw data type detected by Driverless AI when the data was imported
  • datetime_format: user defined datetime format to be used by Driverless AI (see dataset.set_datetime_format())
  • freq: count of most frequent value
  • logical_types: list of user defined data types to be used by Driverless AI (overrides data_type, also see dataset.set_logical_types())
  • max: maximum value for numeric data
  • mean: mean of values for numeric data
  • min: minimum value for numeric data
  • missing: count of missing values
  • name: column name
  • sd: standard deviation of values for numeric data
  • unique: count of unique values

Printing the collection or an individual summary displays a histogram along with summary information, like so:

--- C1 ---

 4.3|███████
    |█████████████████
    |██████████
    |████████████████████
    |████████████
    |███████████████████
    |█████████████
    |████
    |████
 7.9|████

Data Type: real
Logical Types: ['categorical', 'numerical']
Datetime Format:
Count: 150
Missing: 0
Mean: 5.84
SD: 0.828
Min: 4.3
Max: 7.9
Unique: 35
Freq: 10
Parameters:columns (Optional[List[str]]) – list of column names to include in the collection

Examples:

# Import the iris dataset
ds = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/iris/iris.csv',
    data_source='s3'
)

# print column summary for the first three columns
print(ds.column_summaries()[0:3])
Return type:DatasetColumnSummaryCollection
property columns: List[str]

List of column names.

Return type:List[str]
property creation_timestamp: float

Creation timestamp in seconds since the epoch (POSIX timestamp).

Return type:float
property data_source: str

Original source of data.

Return type:str
delete() → None

Delete dataset on Driverless AI server.

Examples:

# Import the iris dataset
ds = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/iris/iris.csv',
    data_source='s3'
)

ds.delete()
Return type:None
download(dst_dir: str = '.', dst_file: Optional[str] = None, file_system: Optional[fsspec.spec.AbstractFileSystem] = None, overwrite: bool = False) → str

Download dataset from Driverless AI server as a csv.

Parameters:
  • dst_dir (str) – directory where csv will be saved
  • dst_file (Optional[str]) – name of csv file (overrides default file name)
  • file_system (Optional[ForwardRef]) – FSSPEC based file system to download to, instead of local file system
  • overwrite (bool) – overwrite existing file

Examples:

# Import the iris dataset
ds = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/iris/iris.csv',
    data_source='s3'
)

ds.download()
Return type:str
export(**kwargs: Any) → str

Export dataset csv from the Driverless AI server. Returns a relative path for the exported csv.

Note

Export location is configured on the Driverless AI server.

Return type:str
property file_path: str

Path to dataset bin file on the server.

Return type:str
property file_size: int

Size in bytes of dataset bin file on the server.

Return type:int
head(num_rows: int = 5) → Table

Return headers and first n rows of dataset in a Table.

Parameters:num_rows (int) – number of rows to show

Examples:

# Load in the iris dataset
ds = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/iris/iris.csv',
    data_source='s3'
)

# Print the headers and first 5 rows
print(ds.head(num_rows=5))
Return type:Table
property key: str

Universally unique identifier.

Return type:str
modify_by_code(code: str, names: Optional[List[str]] = None) → Dict[str, Dataset]

Create a dictionary of new datasets from original dataset modified by a Python code string, that is the body of a function where:

  • there is an input variable X that represents the original dataset in the form of a datatable frame (dt.Frame)
  • return type is one of dt.Frame, pd.DataFrame, np.ndarray or a list of those
Parameters:
  • code (str) – Python code that modifies X
  • names (Optional[List[str]]) – optional list of names for the new dataset(s)

Examples:

# Import the iris dataset
ds = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/iris/iris.csv',
    data_source='s3'
)

# Keep the first 4 columns
new_dataset = ds.modify_by_code(
    'return X[:, :4]', names=['new_dataset']
)

# Split on 4th column
new_datasets = ds.modify_by_code(
    'return [X[:, :4], X[:, 4:]]',
    names=['new_dataset_1', 'new_dataset_2']
)

The dictionary will map the dataset names to the returned element(s) from the Python code string.

Return type:Dict[str, Dataset]
modify_by_code_preview(code: str) → Table

Get a preview of the dataset modified by a Python code string, where:

  • there exists a variable X that represents the original dataset in the form of a datatable frame (dt.Frame)
  • return type is one of dt.Frame, pd.DataFrame, np.ndarray or a list of those (only first element of the list is shown in preview)
Parameters:code (str) – Python code that modifies X

Examples:

# Import the iris dataset
ds = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/iris/iris.csv',
    data_source='s3'
)

# Keep first 4 columns
ds.modify_by_code_preview('return X[:, :4]')
Return type:Table
modify_by_recipe(recipe: str, names: Optional[List[str]] = None) → Dict[str, Dataset]

Create a dictionary of new datasets from original dataset modified by a recipe.

The dictionary will map the dataset names to the returned element(s) from the recipe.

Parameters:
  • recipe (str) – path to recipe or url for recipe
  • names (Optional[List[str]]) – optional list of names for the new dataset(s)

Examples:

# Import the airlines dataset
ds = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/airlines/allyears2k_headers.zip',
    data_source='s3'
)

# Modify original dataset with a recipe
new_ds = ds.modify_by_recipe(
    recipe='https://github.com/h2oai/driverlessai-recipes/blob/rel-1.8.4/data/airlines_multiple.py',
    names=['new_airlines1', 'new_airlines2']
)
Return type:Dict[str, Dataset]
property name: str

Display name.

Return type:str
rename(name: str) → Dataset

Change dataset display name.

Parameters:name (str) – new display name

Examples:

# Import the iris dataset
ds = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/iris/iris.csv',
    data_source='s3'
)
ds.name

ds.rename(name='new-iris-name')
ds.name
Return type:Dataset
set_datetime_format(columns: Dict[str, str]) → None

Set datetime format of columns.

Parameters:columns (Dict[str, str]) – dictionary where the key is the column name and the value is a valid datetime format

Examples:

# Import the Eurodate dataset
date = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/jira/v-11-eurodate.csv',
    data_source='s3'
)

# Set the date time format for column ‘ds5'
date.set_datetime_format({'ds5': '%d-%m-%y %H:%M'})
Return type:None
set_logical_types(columns: Dict[str, Union[str, List[str]]]) → None

Designate columns to have the specified logical types. The logical type is mainly used to determine which transformers to try on the column’s data.

Possible logical types:

  • 'categorical'
  • 'date'
  • 'datetime'
  • 'id'
  • 'numerical'
  • 'text'
Parameters:columns (Dict[str, Union[str, List[str]]]) – dictionary where the key is the column name and the value is the logical type or a list of logical types for the column (to unset all logical types use a value of None)

Example:

# Import the prostate dataset
prostate = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/prostate/prostate.csv',
    data_source='s3'
)

# Set the logical types
prostate.set_logical_types(
    {'ID': 'id', 'AGE': ['categorical', 'numerical'], 'RACE': None}
)
Return type:None
property shape: Tuple[int, int]

Dimensions (rows, cols).

Return type:Tuple[int, int]
split_to_train_test(train_size: float = 0.5, train_name: Optional[str] = None, test_name: Optional[str] = None, target_column: Optional[str] = None, fold_column: Optional[str] = None, time_column: Optional[str] = None, seed: int = 1234) → Dict[str, Dataset]

Split a dataset into train/test sets on the Driverless AI server and return a dictionary of Dataset objects with the keys 'train_dataset' and 'test_dataset'.

Parameters:
  • train_size (float) – proportion of dataset rows to put in the train split
  • train_name (Optional[str]) – name for the train dataset
  • test_name (Optional[str]) – name for the test dataset
  • target_column (Optional[str]) – use stratified sampling to create splits
  • fold_column (Optional[str]) – keep rows belonging to the same group together
  • time_column (Optional[str]) – split rows such that the splits are sequential with respect to time
  • seed (int) – random seed

Note

Only one of target_column, fold_column, or time_column can be passed at a time.

Examples:

# Import the iris dataset
ds = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/iris/iris.csv',
    data_source='s3'
)

# Split the iris dataset into train/test sets
ds_split = ds.split_to_train_test(train_size=0.7)
Return type:Dict[str, Dataset]
split_to_train_test_async(train_size: float = 0.5, train_name: Optional[str] = None, test_name: Optional[str] = None, target_column: Optional[str] = None, fold_column: Optional[str] = None, time_column: Optional[str] = None, seed: int = 1234) → DatasetSplitJob

Launch creation of a dataset train/test split on the Driverless AI server and return a DatasetSplitJob object to track the status.

Parameters:
  • train_size (float) – proportion of dataset rows to put in the train split
  • train_name (Optional[str]) – name for the train dataset
  • test_name (Optional[str]) – name for the test dataset
  • target_column (Optional[str]) – use stratified sampling to create splits
  • fold_column (Optional[str]) – keep rows belonging to the same group together
  • time_column (Optional[str]) – split rows such that the splits are sequential with respect to time
  • seed (int) – random seed

Note

Only one of target_column, fold_column, or time_column can be passed at a time.

Examples:

# Import the iris dataset
ds = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/iris/iris.csv',
    data_source='s3'
)

# Launch the creation of a dataset train/test split on the DAI server
ds_split = ds.split_to_train_test_async(train_size=0.7)
Return type:DatasetSplitJob
tail(num_rows: int = 5) → Table

Return headers and last n rows of dataset in a Table.

Parameters:num_rows (int) – number of rows to show

Examples:

ds = dai.datasets.create(
    data='s3://h2o-public-test-data/smalldata/iris/iris.csv',
    data_source='s3'
)

# Print the headers and last 5 rows
print(ds.tail(num_rows=5))
Return type:Table
class DatasetJob

Monitor creation of a dataset on the Driverless AI server.

is_complete() → bool

Return True if job completed successfully.

Return type:bool
is_running() → bool

Return True if job is scheduled, running, or finishing.

Return type:bool
property key: str

Universally unique identifier.

Return type:str
property name: str

Display name.

Return type:str
result(silent: bool = False) → Dataset

Wait for job to complete, then return a Dataset object.

Parameters:silent (bool) – if True, don’t display status updates
Return type:Dataset
status(verbose: int = 0) → str

Return job status string.

Parameters:verbose (int) –
  • 0: short description
  • 1: short description with progress percentage
  • 2: detailed description with progress percentage
Return type:str

Experiment

Experiment objects correspond to existing experiments on a Driverless AI server. Experiment objects are retrievable using the Client.

API Reference:

class Experiment

Interact with an experiment on the Driverless AI server.

abort() → None

Terminate experiment immediately and only generate logs.

Return type:None
property artifacts: driverlessai._experiments.ExperimentArtifacts

Interact with artifacts that are created when the experiment completes.

Return type:ExperimentArtifacts
property creation_timestamp: float

Creation timestamp in seconds since the epoch (POSIX timestamp).

Return type:float
property datasets: Dict[str, Optional[driverlessai._datasets.Dataset]]

Dictionary of train_dataset, validation_dataset, and test_dataset used for the experiment.

Return type:Dict[str, Optional[Dataset]]
delete() → None

Permanently delete experiment from the Driverless AI server.

Return type:None
export_dai_file(dst_dir: str = '.', dst_file: Optional[str] = None, file_system: Optional[fsspec.spec.AbstractFileSystem] = None, overwrite: bool = False) → str

Export experiment from Driverless AI server in dai format.

Parameters:
  • dst_dir (str) – directory where dai file will be saved
  • dst_file (Optional[str]) – name of dai file (overrides default file name)
  • file_system (Optional[ForwardRef]) – FSSPEC based file system to download to, instead of local file system
  • overwrite (bool) – overwrite existing file
Return type:

str
finish() → None

Finish experiment by jumping to final pipeline training and generating experiment artifacts.

Return type:None
gui() → Hyperlink

Get full URL for the experiment’s page on the Driverless AI server.

Return type:Hyperlink
is_complete() → bool

Return True if job completed successfully.

Return type:bool
property is_deprecated: bool

True if experiment was created by an old version of Driverless AI and is no longer fully compatible with the current server version.

Return type:bool
is_running() → bool

Return True if job is scheduled, running, or finishing.

Return type:bool
property key: str

Universally unique identifier.

Return type:str
property log: driverlessai._experiments.ExperimentLog

Interact with experiment logs.

Return type:ExperimentLog
metrics() → Dict[str, Union[str, float]]

Return dictionary of experiment scorer metrics and AUC metrics, if available.

Return type:Dict[str, Union[str, float]]
property name: str

Display name.

Return type:str
notifications() → List[Dict[str, str]]

Return list of experiment notification dictionaries.

Return type:List[Dict[str, str]]
predict(dataset: Dataset, enable_mojo: bool = True, include_columns: Optional[List[str]] = None, include_labels: Optional[bool] = None, include_raw_outputs: Optional[bool] = None, include_shap_values_for_original_features: Optional[bool] = None, include_shap_values_for_transformed_features: Optional[bool] = None, use_fast_approx_for_shap_values: Optional[bool] = None) → Prediction

Predict on a dataset, then return a Prediction object.

Parameters:
  • dataset (Dataset) – a Dataset object corresonding to a dataset on the Driverless AI server
  • enable_mojo (bool) – use MOJO (if available) to make predictions (server versions >= 1.9.1)
  • include_columns (Optional[List[str]]) – list of columns from the dataset to append to the prediction csv
  • include_labels (Optional[bool]) – append labels in addition to probabilities for classification, ignored for regression (server versions >= 1.10)
  • include_raw_outputs (Optional[bool]) – append predictions as margins (in link space) to the prediction csv
  • include_shap_values_for_original_features (Optional[bool]) – append original feature contributions to the prediction csv (server versions >= 1.9.1)
  • include_shap_values_for_transformed_features (Optional[bool]) – append transformed feature contributions to the prediction csv
  • use_fast_approx_for_shap_values (Optional[bool]) – speed up prediction contributions with approximation (server versions >= 1.9.1)
Return type:

Prediction
predict_async(dataset: Dataset, enable_mojo: bool = True, include_columns: Optional[List[str]] = None, include_labels: Optional[bool] = None, include_raw_outputs: Optional[bool] = None, include_shap_values_for_original_features: Optional[bool] = None, include_shap_values_for_transformed_features: Optional[bool] = None, use_fast_approx_for_shap_values: Optional[bool] = None) → PredictionJobs

Launch prediction job on a dataset and return a PredictionJobs object to track the status.

Parameters:
  • dataset (Dataset) – a Dataset object corresonding to a dataset on the Driverless AI server
  • enable_mojo (bool) – use MOJO (if available) to make predictions (server versions >= 1.9.1)
  • include_columns (Optional[List[str]]) – list of columns from the dataset to append to the prediction csv
  • include_labels (Optional[bool]) – append labels in addition to probabilities for classification, ignored for regression (server versions >= 1.10)
  • include_raw_outputs (Optional[bool]) – append predictions as margins (in link space) to the prediction csv
  • include_shap_values_for_original_features (Optional[bool]) – append original feature contributions to the prediction csv (server versions >= 1.9.1)
  • include_shap_values_for_transformed_features (Optional[bool]) – append transformed feature contributions to the prediction csv
  • use_fast_approx_for_shap_values (Optional[bool]) – speed up prediction contributions with approximation (server versions >= 1.9.1)
Return type:

PredictionJobs
rename(name: str) → Experiment

Change experiment display name.

Parameters:name (str) – new display name
Return type:Experiment
result(silent: bool = False) → Experiment

Wait for training to complete, then return self.

Parameters:silent (bool) – if True, don’t display status updates
Return type:Experiment
retrain(use_smart_checkpoint: bool = False, final_pipeline_only: bool = False, final_models_only: bool = False, **kwargs: Any) → Experiment

Create a new experiment using the same datasets and settings. Through kwargs it’s possible to pass new datasets or overwrite settings.

Parameters:
  • use_smart_checkpoint (bool) – start experiment from last smart checkpoint
  • final_pipeline_only (bool) – trains final pipeline using smart checkpoint if available, otherwise uses default hyperparameters
  • final_models_only (bool) – trains final pipeline models (but not transformers) using smart checkpoint if available, otherwise uses default hyperparameters and transformers (overrides final_pipeline_only)
  • kwargs (Any) – datasets and experiment settings as defined in experiments.create()
Return type:

Experiment
retrain_async(use_smart_checkpoint: bool = False, final_pipeline_only: bool = False, final_models_only: bool = False, **kwargs: Any) → Experiment

Launch creation of a new experiment using the same datasets and settings. Through kwargs it’s possible to pass new datasets or overwrite settings.

Parameters:
  • use_smart_checkpoint (bool) – start experiment from last smart checkpoint
  • final_pipeline_only (bool) – trains final pipeline using smart checkpoint if available, otherwise uses default hyperparameters
  • final_models_only (bool) – trains final pipeline models (but not transformers) using smart checkpoint if available, otherwise uses default hyperparameters and transformers (overrides final_pipeline_only)
  • kwargs (Any) – datasets and experiment settings as defined in experiments.create()
Return type:

Experiment
property run_duration: Optional[float]

Run duration in seconds.

Return type:Optional[float]
property settings: Dict[str, Any]

Experiment settings.

Return type:Dict[str, Any]
property size: int

Size in bytes of all experiment’s files on the Driverless AI server.

Return type:int
status(verbose: int = 0) → str

Return job status string.

Parameters:verbose (int) –
  • 0: short description
  • 1: short description with progress percentage
  • 2: detailed description with progress percentage
Return type:str
summary() → None

Print experiment summary.

Return type:None
variable_importance() → Optional[Table]

Get variable importance in a Table.

Return type:Optional[Table]

Experiment Artifacts

Experiment artifacts include anything outputted after a successfully completed experiment. These artificats include the autoreport, scoring pipelines, prediction csvs, experiment summary, and logs.

API Reference:

class ExperimentArtifacts

Interact with files created by an experiment on the Driverless AI server.

create(artifact: str) → None

(Re)build certain artifacts, if possible.

(re)buildable artifacts:

  • 'autodoc'
  • 'mojo_pipeline'
  • 'python_pipeline'
Parameters:artifact (str) – name of artifact to (re)build
Return type:None
download(only: Union[str, List[str]] = None, dst_dir: str = '.', file_system: Optional[fsspec.spec.AbstractFileSystem] = None, include_columns: Optional[List[str]] = None, overwrite: bool = False) → Dict[str, str]

Download experiment artifacts from the Driverless AI server. Returns a dictionary of relative paths for the downloaded artifacts.

Parameters:
  • only (Union[str, List[str]]) – specify specific artifacts to download, use experiment.artifacts.list() to see the available artifacts on the Driverless AI server
  • dst_dir (str) – directory where experiment artifacts will be saved
  • file_system (Optional[ForwardRef]) – FSSPEC based file system to download to, instead of local file system
  • include_columns (Optional[List[str]]) – list of dataset columns to append to prediction csvs
  • overwrite (bool) – overwrite existing files
Return type:

Dict[str, str]
export(only: Optional[Union[str, List[str]]] = None, include_columns: Optional[List[str]] = None, **kwargs: Any) → Dict[str, str]

Export experiment artifacts from the Driverless AI server. Returns a dictionary of relative paths for the exported artifacts.

Parameters:
  • only (Union[str, List[str], None]) – specify specific artifacts to export, use ex.artifacts.list() to see the available artifacts on the Driverless AI server
  • include_columns (Optional[List[str]]) – list of dataset columns to append to prediction csvs

Note

Export location is configured on the Driverless AI server.

Return type:Dict[str, str]
property file_paths: Dict[str, str]

Paths to artifact files on the server.

Return type:Dict[str, str]
list() → List[str]

List of experiment artifacts that exist on the Driverless AI server.

Return type:List[str]

Experiment Logs

Experiment logs list the events recorded during an experiment.

API Reference:

class ExperimentLog

Interact with experiment logs.

download(archive: bool = True, dst_dir: str = '.', dst_file: Optional[str] = None, file_system: Optional[fsspec.spec.AbstractFileSystem] = None, overwrite: bool = False) → str

Download experiment logs from the Driverless AI server.

Parameters:
  • archive (bool) – if available, prefer downloading an archive that contains multiple log files and stack traces if any were created
  • dst_dir (str) – directory where logs will be saved
  • dst_file (Optional[str]) – name of log file (overrides default file name)
  • file_system (Optional[ForwardRef]) – FSSPEC based file system to download to, instead of local file system
  • overwrite (bool) – overwrite existing file
Return type:

str
head(num_lines: int = 50) → None

Print first n lines of experiment log.

Parameters:num_lines (int) – number of lines to print
Return type:None
tail(num_lines: int = 50) → None

Print last n lines of experiment log.

Parameters:num_lines (int) – number of lines to print
Return type:None

Predictions

Prediction objects are created when predicting on a new dataset.

API Reference:

class Prediction

Interact with predictions from the Driverless AI server.

download(dst_dir: str = '.', dst_file: Optional[str] = None, file_system: Optional[fsspec.spec.AbstractFileSystem] = None, overwrite: bool = False) → str

Download csv of predictions.

Parameters:
  • dst_dir (str) – directory where csv will be saved
  • dst_file (Optional[str]) – name of csv file (overrides default file name)
  • file_system (Optional[ForwardRef]) – FSSPEC based file system to download to, instead of local file system
  • overwrite (bool) – overwrite existing file
Return type:

str
property file_paths: List[str]

Paths to prediction csv files on the server.

Return type:List[str]
property included_dataset_columns: List[str]

Columns from dataset that are appended to predictions.

Return type:List[str]
property includes_labels: bool

Whether classification labels are appended to predictions.

Return type:bool
property includes_raw_outputs: bool

Whether predictions as margins (in link space) were appended to predictions.

Return type:bool
property includes_shap_values_for_original_features: bool

Whether original feature contributions are appended to predictions (server versions >= 1.9.1).

Return type:bool
property includes_shap_values_for_transformed_features: bool

Whether transformed feature contributions are appended to predictions.

Return type:bool
property keys: Dict[str, str]

Dictionary of unique IDs for entities related to the prediction: dataset: unique ID of dataset used to make predictions experiment: unique ID of experiment used to make predictions prediction: unique ID of predictions

Return type:Dict[str, str]
to_pandas() → pandas.DataFrame

Transfer predictions to a local Pandas DataFrame.

Return type:pandas.DataFrame
property used_fast_approx_for_shap_values: Optional[bool]

Whether approximation was used to calculate prediction contributions (server versions >= 1.9.1).

Return type:Optional[bool]
class PredictionJobs

Monitor creation of predictions on the Driverless AI server.

property included_dataset_columns: List[str]

Columns from dataset that are appended to predictions.

Return type:List[str]
property includes_labels: bool

Whether classification labels are appended to predictions.

Return type:bool
property includes_raw_outputs: bool

Whether predictions as margins (in link space) are appended to predictions.

Return type:bool
property includes_shap_values_for_original_features: bool

Whether original feature contributions are appended to predictions (server versions >= 1.9.1).

Return type:bool
property includes_shap_values_for_transformed_features: bool

Whether transformed feature contributions are appended to predictions.

Return type:bool
is_complete() → bool

Return True if all jobs completed successfully.

Return type:bool
is_running() → bool

Return True if one or more jobs is running or finishing.

Return type:bool
property jobs: Sequence[driverlessai._utils.ServerJob]

List of ServerJob objects.

Return type:Sequence[ServerJob]
property keys: Dict[str, str]

Dictionary of entity unique IDs: dataset: unique ID of dataset used to make predictions experiment: unique ID of experiment used to make predictions prediction: unique ID of predictions

Return type:Dict[str, str]
result(silent: bool = False) → Prediction

Wait for all jobs to complete.

Parameters:silent (bool) – if True, don’t display status updates
Return type:Prediction
status(verbose: int = 0) → List[str]

Returns list of job status strings.

Parameters:verbose (int) –
  • 0: short description
  • 1: short description with progress percentage
  • 2: detailed description with progress percentage
Return type:List[str]
property used_fast_approx_for_shap_values: Optional[bool]

Whether approximation was used to calculate prediction contributions (server versions >= 1.9.1).

Return type:Optional[bool]

Interpretation

Interpretation objects correspond to existing interpretations on a Driverless AI server. Interpretation objects are retrievable using the Client.

API Reference:

class Interpretation

Interact with a MLI interpretation on the Driverless AI server.

property artifacts: driverlessai._mli.InterpretationArtifacts

Interact with artifacts that are created when the interpretation completes.

Return type:InterpretationArtifacts
property creation_timestamp: float

Creation timestamp in seconds since the epoch (POSIX timestamp).

Return type:float
property dataset: Optional[driverlessai._datasets.Dataset]

Dataset for the interpretation.

Return type:Optional[Dataset]
delete() → None

Delete MLI interpretation on Driverless AI server.

Return type:None
property experiment: Optional[driverlessai._experiments.Experiment]

Experiment for the interpretation.

Return type:Optional[Experiment]
gui() → Hyperlink

Get full URL for the interpretation’s page on the Driverless AI server.

Return type:Hyperlink
is_complete() → bool

Return True if job completed successfully.

Return type:bool
is_running() → bool

Return True if job is scheduled, running, or finishing.

Return type:bool
property key: str

Universally unique identifier.

Return type:str
property name: str

Display name.

Return type:str
rename(name: str) → Interpretation

Change interpretation display name.

Parameters:name (str) – new display name
Return type:Interpretation
result(silent: bool = False) → Interpretation

Wait for job to complete, then return an Interpretation object.

Return type:Interpretation
property run_duration: Optional[float]

Run duration in seconds.

Return type:Optional[float]
property settings: Dict[str, Any]

Interpretation settings.

Return type:Dict[str, Any]
status(verbose: int = 0) → str

Return job status string.

Parameters:verbose (int) –
  • 0: short description
  • 1: short description with progress percentage
  • 2: detailed description with progress percentage
Return type:str

Interpretation Artifacts

Interpretation artifacts include anything available for download after a successfully completed interpretation.

API Reference:

class InterpretationArtifacts

Interact with files created by a MLI interpretation on the Driverless AI server.

download(only: Union[str, List[str]] = None, dst_dir: str = '.', file_system: Optional[fsspec.spec.AbstractFileSystem] = None, overwrite: bool = False) → Dict[str, str]

Download interpretation artifacts from the Driverless AI server. Returns a dictionary of relative paths for the downloaded artifacts.

Parameters:
  • only (Union[str, List[str]]) – specify specific artifacts to download, use interpretation.artifacts.list() to see the available artifacts on the Driverless AI server
  • dst_dir (str) – directory where interpretation artifacts will be saved
  • file_system (Optional[ForwardRef]) – FSSPEC based file system to download to, instead of local file system
  • overwrite (bool) – overwrite existing files
Return type:

Dict[str, str]
property file_paths: Dict[str, str]

Paths to artifact files on the server.

Return type:Dict[str, str]
list() → List[str]

List of interpretation artifacts that exist on the Driverless AI server.

Return type:List[str]

Project

Project objects correspond to existing projects on a Driverless AI server. Project objects are retrievable using the Client.

API Reference:

class Project

Interact with a project on the Driverless AI server.

property datasets: Dict[str, Sequence[driverlessai._datasets.Dataset]]

Datasets linked to the project.

Return type:Dict[str, Sequence[Dataset]]
delete(include_experiments: bool = False) → None

Permanently delete project from the Driverless AI server.

Parameters:include_experiments (bool) – unlink & delete experiments linked to this project.
Return type:None
property description: Optional[str]

Project description.

Return type:Optional[str]
property experiments: Sequence[driverlessai._experiments.Experiment]

Experiments linked to the project.

Return type:Sequence[Experiment]
gui() → Hyperlink

Get full URL for the project’s page on the Driverless AI server.

Return type:Hyperlink
property key: str

Universally unique identifier.

Return type:str

Link a dataset to the project.

Parameters:
  • dataset (Dataset) – a Dataset object corresonding to a dataset on the Driverless AI server
  • dataset_type (str) – can be one of: 'train_dataset(s)', 'validation_dataset(s)', or 'test_dataset(s)'
  • link_associated_experiments (bool) – also link experiments that used the dataset (server versions >= 1.9.1)
Return type:

Project

Link an experiment to the project.

Parameters:experiment (Experiment) – an Experiment object corresonding to a experiment on the Driverless AI server
Return type:Project
property name: str

Display name.

Return type:str
redescribe(description: str) → Project

Change project description. Requires server version >= 1.9.1.

Parameters:description (str) – new description
Return type:Project
rename(name: str) → Project

Change project display name.

Parameters:name (str) – new display name
Return type:Project
share(username: str, role: str = 'Default') → None

Share a project. Requires server versions >= 1.9.3 with H2O.ai Storage connected.

Parameters:
  • username (str) – Driverless AI username of user to share with
  • role (str) – one of “Default” or “Reader”
Return type:

None
property sharings: List[Dict[str, Optional[str]]]

Users the project is shared with. Requires server versions >= 1.9.3 with H2O.ai Storage connected.

Return type:List[Dict[str, Optional[str]]]

Unlink a dataset from the project.

Parameters:
  • dataset (Dataset) – a Dataset object corresonding to a dataset on the Driverless AI server
  • dataset_type (str) – can be one of: 'train_dataset(s)', 'validation_dataset(s)', or 'test_dataset(s)'
Return type:

Project

Unlink an experiment from the project.

Parameters:experiment (Experiment) – an Experiment object corresonding to a experiment on the Driverless AI server
Return type:Project
unshare(username: str) → None

Unshare a project. Requires server versions >= 1.9.3 with H2O.ai Storage connected.

Parameters:username (str) – Driverless AI username of user to unshare with
Return type:None

Recipe

Recipe objects correspond to existing recipes on a Driverless AI server. Recipe objects are retrievable using the Client.

API Reference:

class ExplainerRecipe

Interact with an explainer recipe on the Driverless AI server.

property for_binomial: bool

True if explainer works for binomial models.

Return type:bool
property for_iid: bool

True if explainer works for I.I.D. models.

Return type:bool
property for_multiclass: bool

True if explainer works for multiclass models.

Return type:bool
property for_regression: bool

True if explainer works for regression models.

Return type:bool
property for_timeseries: bool

True if explainer works for time series models.

Return type:bool
property id: str

Identifier.

Return type:str
property is_custom: bool

True if the recipe is custom.

Return type:bool
property key: str

Universally unique identifier.

Return type:str
property name: str

Display name.

Return type:str
search_settings(search_term: str, show_description: bool = False) → None

Search explainer settings and print results. Useful when looking for explainer kwargs (see explainer.with_settings()) to use when creating interpretations.

Parameters:
  • search_term (str) – term to search for (case insensitive)
  • show_description (bool) – include description in results
Return type:

None
property settings: Dict[str, Any]

Explainer settings set by user.

Return type:Dict[str, Any]
with_settings(**kwargs: Any) → ExplainerRecipe

Changes the explainer settings from defaults. Settings reset to defaults everytime this is called.

Note

To search possible explainer settings for your server version, use explainer.search_settings(search_term).

Return type:ExplainerRecipe
class ModelRecipe

Interact with a model recipe on the Driverless AI server.

property is_custom: bool

True if the recipe is custom.

Return type:bool
property is_unsupervised: bool

True if recipe doesn’t require a target column.

Return type:bool
property key: str

Universally unique identifier.

Return type:str
property name: str

Display name.

Return type:str
class ScorerRecipe

Interact with a scorer recipe on the Driverless AI server.

property description: str

Recipe description.

Return type:str
property for_binomial: bool

True if scorer works for binomial models.

Return type:bool
property for_multiclass: bool

True if scorer works for multiclass models.

Return type:bool
property for_regression: bool

True if scorer works for regression models.

Return type:bool
property is_custom: bool

True if the recipe is custom.

Return type:bool
property key: str

Universally unique identifier.

Return type:str
property name: str

Display name.

Return type:str
class TransformerRecipe

Interact with a transformer recipe on the Driverless AI server.

property is_custom: bool

True if the recipe is custom.

Return type:bool
property key: str

Universally unique identifier.

Return type:str
property name: str

Display name.

Return type:str

Utility

API Reference:

Renders clickable link in notebooks but otherwise behaves the same as str.

class Table

Table that pretty prints.

Parameters:
  • data (List[List[Any]]) – two-dimensional list
  • headers (List[str]) – column labels
property data: List[List[Any]]

Table data.

Return type:List[List[Any]]
property headers: List[str]

Table headers.

Return type:List[str]

Visualization

Visualization objects correspond to existing dataset visualizations on a Driverless AI server. Visualization objects are retrievable using the Client.

API Reference:

class Visualization

Interact with a dataset visualization on the Driverless AI server.

property dataset: driverlessai._datasets.Dataset

Dataset that was visualized.

Return type:Dataset
delete() → None

Permanently delete visualization from the Driverless AI server.

Return type:None
gui() → Hyperlink

Get full URL for the visualization’s page on the Driverless AI server.

Return type:Hyperlink
is_complete() → bool

Return True if job completed successfully.

Return type:bool
property is_deprecated: Optional[bool]

True if visualization was created by an old version of Driverless AI and is no longer fully compatible with the current server version.

Return type:Optional[bool]
is_running() → bool

Return True if job is scheduled, running, or finishing.

Return type:bool
property key: str

Universally unique identifier.

Return type:str
property name: str

Display name.

Return type:str
result(silent: bool = False) → Visualization

Wait for job to complete, then return self.

Parameters:silent (bool) – if True, don’t display status updates
Return type:Visualization
status(verbose: int = 0) → str

Return job status string.

Parameters:verbose (int) –
  • 0: short description
  • 1: short description with progress percentage
  • 2: detailed description with progress percentage
Return type:str