Prepare OpenML import

[LizzAlice]

Get Overview about data available via api. This will be documented here:

• dataset data openml.datasets.list_datasets(output_format="dataframe")
  • did: unique dataset ID
  • name: non unique
  • version: int, the combination of name and version seems to be unique in every case but one
  • uploader: int (maybe this is a user id??)
  • status: "active" for all of them
  • format: one of ARFF, SParse_ARFF, arff or sparse_arff
  • MajorityClassSize: number or NaN
  • MaxNominalAttDistinctValues: number or NaN
  • MinorityClassSize: number or NaN
  • NumberOfClasses: number or NaN
  • NumberOfFeatures: number or NaN
  • NumberOfInstances: number or NaN
  • NumberOfInstancesWithMissingValues: number or NaN
  • NumberOfMissingValues: number or NaN
  • NumberOfNumericFeatues: number or NaN
  • NumberOfSymbolicFeatures: number or NaN
• evaluations (have to give evaluation function)
  • run_id: run id
  • task_id: task id
  • setup_id: setup id
  • flow_id: flow id
  • flow_name: flow name
  • data_id: dataset id?
  • data_name: dataset name?
  • function: evaluation function
  • upload_time: time it was uploaded
  • uploader: uploader number
  • uploader_name: name string
  • value: int
  • values: always None?
  • array_data: always None?
• flows
  • id: unique id
  • full_name: name with number in parentheses
  • name: name of python class or function\
  • version: number
  • external_version: None or package versions with package name in the form 'openml==0.14.1,sklearn==1.3.0'
  • uploader: number
• runs
  • run_id: unique id
  • task_id: task id
  • setup_id: setup id
  • flow_id: flow id
  • uploader: number
  • task_type: instance of task type in the following form: TaskType.LEARNING_CURVE
  • upload_time: time in the format of 2014-04-06 23:30:40
  • error_message: string
• setups:
  • setup_id: unique id
  • flow_id: flow id
  • parameters: dict of things that are given as numbers; the dicts contain information such as flow information, data_type, default_value etc
• study openml.study.list_studies(output_format="dataframe") (a bit unclear, what this is, but there are only two... However, from the ids, it seems as if there were more)
  • id: unique id, only 123 and 226
  • main_entity_type: "run"
  • status: "active"
  • creation_date: time in the format of 2019-02-21 19:55:30
  • creator: number
  • alia: NaN or "amlb"
• tasks openml.tasks.list_tasks(output_format="dataframe")
  • tid: unique task id
  • ttid: String with task type in the form of TaskType.TASK_TYPE_NAME
  • did: dataset id
  • name: should be the task name, but actually looks like the dataset name
  • task_type: task type as in ttid, but in words
  • status: "active" for all of them
  • estimation_procedure: string
  • evaluation_measures: string or NaN
  • source_data: seems to be the same as did
  • target_feature: string
  • MajorityClassSize: number or NaN (is this the value from the dataset?)
  • MaxNominalAttDistinctValues: number or NaN (is this the value from the dataset?)
  • MinorityClassSize: number or NaN (is this the value from the dataset?)
  • NumberOfClasses: number or NaN (is this the value from the dataset?)
  • NumberOfFeatures: number or NaN (is this the value from the dataset?)
  • NumberOfInstances: number or NaN (is this the value from the dataset?)
  • NumberOfInstancesWithMissingValues: number or NaN (is this the value from the dataset?)
  • NumberOfMissingValues: number or NaN (is this the value from the dataset?)
  • NumberOfNumericFeatures: number or NaN (is this the value from the dataset?)
  • NumberOfSymbolicFeatures: number or NaN (is this the value from the dataset?)
  • number_samples: number or NaN
  • cost_matrix: NaN or matrix in list of lists format or string or number
  • source_data_labeled: NaN or '1227' or '1451'
  • target_feature_event: NaN, or 'event' or 'OS_event'
  • target_feature_left: NaN
  • target_feature_right: NaN or "time" or "OS_years"
  • quality_measure: NaN or string
  • target_value: NaN or string

Dependencies: Task on Dataset; Run on Task, Setup and Flow; Setup on Flow, Evaluation on Run, Task, Setup, Flow, Dataset

[LizzAlice]

Questions:

• Dataset:
  • why is dataset name not unique? --> just how it works
  • uploader: is this ID unique? -->yes
  • what does status=active mean and why are they all active? always active, can be ignored
• Evaluation:
  • is data_id the dataset_id? yes
  • where do I find a list of evaluation functions? list_evaluation_measures
  • what is values and when is it not None?
  • what is array_data and when is it not None?
• Study:
  • what are studies, why are there only two but the ids seem as if there are more, why are they not linked to the other --> seems to be a bug
• Task:
  • name here is not the task name, but the dataset name, or what? they dont have a name
  • what does source_data_labeled mean?
  • target_feature_event: what is the difference between event and OS_event?
  • task type classification and regression only important?

[LizzAlice]

• excluded fields for Dataset:
  • MaxNominalAttDistinctValues: this one is the number of distinct attributes overall, i.e. over several columns; doesn't make sense to show

[LizzAlice]

potential changes to prototype:

• what about a field for the quality? --> verified/not
• extra field with just text from "cites work"
• rename cites work to sth that makes it clear that it is an item and it has a doi?
• are all I get back from the api active? --> yes!