"""
Load preprocessed datasets
"""
import json
import warnings
from urllib.request import HTTPError, urlopen # type: ignore
from getml.data import DataFrame
VERSION = "v1"
BUCKET = "https://static.getml.com/datasets"
"""S3 bucket containing the CSV files"""
ASSETS = {
"air_pollution": ["population"],
"atherosclerosis": ["population", "contr"],
"biodegradability": [
"molecule_train",
"molecule_test",
"molecule_validation",
"atom",
"bond",
"gmember",
"group",
],
"consumer_expenditures": [
"population_training",
"population_testing",
"population_validation",
"expd",
"fmld",
"memd",
],
"interstate94": ["traffic"],
"loans": [
"population_train",
"population_test",
"order",
"trans",
"meta",
],
"loans_new": [
"account",
"card",
"client",
"disp",
"district",
"loan",
"order",
"trans",
],
"occupancy": ["population_train", "population_test", "population_validation"],
}
def _load_dataset(
ds_name, assets=None, roles=False, units=False, as_pandas=False, as_dict=False
):
"""Helper function to load a dataset
Args:
ds_name (str):
name of the dataset
assets (list):
CSV files to be loaded from the S3 bucket
roles (bool):
Return getml.DataFrame with roles set
units (bool):
Return getml.DataFrame with units set
as_pandas (bool):
Return data as `pandas.DataFrame` s
as_dict (bool):
Return data as dict with `df.name`s as keys and
`df` as values.
Returns:
tuple:
Tuple containing (sorted alphabetically by `df.name`s) the data as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True) or
dict:
dict containing the with (`df.name`s as keys and `df`s as values) as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True) if `as_dict` is `True`.
"""
base = f"{BUCKET}/{VERSION}/{ds_name}/preprocessed"
if assets is None:
assets = ASSETS[ds_name]
if roles:
filename = base + "/" + ds_name + "_" + "roles.json"
try:
response = urlopen(filename)
roles = json.loads(response.read())
except HTTPError:
raise Exception("Information on roles could not be retrieved.")
else:
roles = {}
dfs = {}
for asset in assets:
filename = base + "/" + ds_name + "_" + asset + ".csv"
with warnings.catch_warnings():
warnings.simplefilter("ignore")
print()
print(f"Loading {asset}...")
dfs[asset] = DataFrame.from_csv(
name=asset,
fnames=filename,
roles=roles.get(asset),
verbose=False,
)
if units:
filename = base + "/" + ds_name + "_" + "units.json"
try:
response = urlopen(filename)
units = json.loads(response.read())
except HTTPError:
raise Exception("Information on units could not be retrieved.")
for key, df in dfs.items():
for unit, columns in units.get(key, {}).items():
if unit:
df.set_unit(columns, unit)
if as_pandas:
dfs = {key: df.to_pandas() for key, df in dfs.items()}
if as_dict:
return dfs
return tuple(dfs.values())
[docs]def load_air_pollution(roles=True, as_pandas=False):
"""
Regression dataset on air pollution in Beijing, China
The dataset consits of a single table split into train and test sets
arround 2014-01-01.
The orgininal publication is:
Liang, X., Zou, T., Guo, B., Li, S., Zhang, H., Zhang, S., Huang, H. and
Chen, S. X. (2015). Assessing Beijing's PM2.5 pollution: severity, weather
impact, APEC and winter heating. Proceedings of the Royal Society A, 471,
20150257.
Args:
as_pandas (bool):
Return data as `pandas.DataFrame` s
roles (bool):
Return data with roles set
Returns:
getml.data.DataFrame:
A DataFrame holding the data described above.
The following DataFrames are returned:
* air_pollution
Examples:
>>> air_pollution = getml.datasets.load_air_pollution()
>>> type(air_pollution)
... getml.data.data_frame.DataFrame
For an full analysis of the atherosclerosis dataset including all necessary
preprocessing steps please refer to `getml-examples
<https://github.com/getml/getml-examples/tree/master/atherosclerosis>`_.
Note:
Roles can be set ad-hoc by supplying the respective flag. If
`roles` is `False`, all columns in the returned
:class:`~getml.DataFrames` s have roles
:const:`~getml.data.roles.unused_string` or
:const:`~getml.data.roles.unused_float`. This dataset contains no units.
Before using them in an analysis, a data model needs to be constructed
using :class:`~getml.data.Placeholder` s.
"""
ds_name = "air_pollution"
return _load_dataset(
ds_name=ds_name,
roles=roles,
as_pandas=as_pandas,
)[0]
[docs]def load_atherosclerosis(roles=True, as_pandas=False, as_dict=False):
"""
Binary classification dataset on the lethality of atherosclerosis
The atherosclerosis dataset is a medical dataset from the the `CTU Prague
Relational Learning Repository
<https://relational.fit.cvut.cz/dataset/Atherosclerosis>`_. It contains
information from an longitudal study on 1417 middle-aged men obeserved over
the course of 20 years. After preprocessing, it consists of 2 tables with 76
and 66 columns:
* `population`: Data on the study's participants
* `contr`: Data on control dates
The population table is split into a training (70%), a testing (15%) set and a
validation (15%) set.
Args:
as_pandas (bool):
Return data as `pandas.DataFrame` s
roles (bool):
Return data with roles set
as_dict (bool):
Return data as dict with `df.name` s as keys and
`df` s as values.
Returns:
tuple:
Tuple containing (sorted alphabetically by `df.name`s) the data as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True) or
dict:
if `as_dict` is `True`: Dictionary containing the data as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True). The keys correspond to the name of the DataFrame on the
:mod:`~getml.engine`.
The following DataFrames are returned:
* population
* contr
Examples:
>>> population, contr = getml.datasets.load_atherosclerosis()
>>> type(population)
... getml.data.data_frame.DataFrame
For an full analysis of the atherosclerosis dataset including all necessary
preprocessing steps please refer to `getml-examples
<https://github.com/getml/getml-examples/tree/master/atherosclerosis>`_.
Note:
Roles can be set ad-hoc by supplying the respective flag. If
`roles` is `False`, all columns in the returned
:class:`~getml.DataFrames` s have roles
:const:`~getml.data.roles.unused_string` or
:const:`~getml.data.roles.unused_float`. This dataset contains no units.
Before using them in an analysis, a data model needs to be constructed
using :class:`~getml.data.Placeholder` s.
"""
ds_name = "atherosclerosis"
return _load_dataset(
ds_name=ds_name,
roles=roles,
as_pandas=as_pandas,
as_dict=as_dict,
)
[docs]def load_biodegradability(roles=True, as_pandas=False, as_dict=False):
"""
Regression dataset on molecule weight prediction
The QSAR biodegradation dataset was built in the Milano Chemometrics and
QSAR Research Group (Universita degli Studi Milano-Bicocca, Milano, Italy).
The data have been used to develop QSAR (Quantitative Structure Activity
Relationships) models for the study of the relationships between chemical
structure and biodegradation of molecules. Biodegradation experimental
values of 1055 chemicals were collected from the webpage of the National
Institute of Technology and Evaluation of Japan (NITE).
The orgininal publication is:
Mansouri, K., Ringsted, T., Ballabio, D., Todeschini, R., Consonni, V.
(2013). Quantitative Structure - Activity Relationship models for ready
biodegradability of chemicals. Journal of Chemical Information and Modeling,
53, 867-878
The dataset was collected through the `UCI Machine Learning Repository`
<https://archive.ics.uci.edu/ml/datasets/QSAR+biodegradation>
It contains information on 1309 molecules with 6166 bonds. It consists of 5
tables.
The population table is split into a training (50 %) and a testing (25%) and
validition (25%) sets.
Args:
as_pandas (bool):
Return data as `pandas.DataFrame` s
roles (bool):
Return data with roles set
as_dict (bool):
Return data as dict with `df.name` s as keys and
`df` s as values.
Returns:
tuple:
Tuple containing (sorted alphabetically by `df.name`s) the data as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True) or
dict:
if `as_dict` is `True`: Dictionary containing the data as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True). The keys correspond to the name of the DataFrame on the
:mod:`~getml.engine`.
The following DataFrames are returned:
* molecule
* atom
* bond
* gmember
* group
Examples:
>>> biodegradability = getml.datasets.load_biodegradability(as_dict=True)
>>> type(biodegradability["molecule_train"])
... getml.data.data_frame.DataFrame
For an full analysis of the biodegradability dataset including all necessary
preprocessing steps please refer to getml-examples (forthcoming).
Note:
Roles can be set ad-hoc by supplying the respective flag. If
`roles` is `False`, all columns in the returned
:class:`~getml.DataFrames` s have roles
:const:`~getml.data.roles.unused_string` or
:const:`~getml.data.roles.unused_float`. This dataset contains no units.
Before using them in an analysis, a data model needs to be constructed
using :class:`~getml.data.Placeholder` s.
"""
ds_name = "biodegradability"
return _load_dataset(
ds_name=ds_name,
roles=roles,
as_pandas=as_pandas,
as_dict=as_dict,
)
# TODO: Remove split
# TODO: Update notebook
[docs]def load_consumer_expenditures(roles=True, units=True, as_pandas=False, as_dict=False):
"""
Binary classification dataset on consumer expenditures
The Consumer Expenditure Data Set is a public domain data set provided by
the American Bureau of Labor Statistics (https://www.bls.gov/cex/pumd.htm).
It includes the diary entries, where American consumers are asked to keep
diaries of the products they have purchased each month,
We use this dataset to classify wether an item was pruchased as a gift.
Args:
roles (bool):
Return data with roles set
units (bool):
Return data with units set
as_pandas (bool):
Return data as `pandas.DataFrame` s
as_dict (bool):
Return data as dict with `df.name` s as keys and
`df` s as values.
Returns:
tuple:
Tuple containing (sorted alphabetically by `df.name`s) the data as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True) or
dict:
if `as_dict` is `True`: Dictionary containing the data as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True). The keys correspond to the name of the DataFrame on the
:mod:`~getml.engine`.
The following DataFrames are returned:
* population
* expd
* fmld
* memd
Examples:
>>> ce = getml.datasets.load_consumer_expenditures(as_dict=True)
>>> type(ce["expd"])
... getml.data.data_frame.DataFrame
For an full analysis of the occupancy dataset including all necessary
preprocessing steps please refer to `getml-examples
<https://github.com/getml/getml-examples/tree/master/consumer_expenditures>`_.
Note:
Roles and units can be set ad-hoc by supplying the respective flag. If
`roles` is `False`, all columns in the returned
:class:`~getml.DataFrames` s have roles
:const:`~getml.data.roles.unused_string` or
:const:`~getml.data.roles.unused_float`.`
Before using them in an analysis, a data model needs to be constructed
using :class:`~getml.data.Placeholder` s.
"""
ds_name = "consumer_expenditures"
return _load_dataset(
ds_name=ds_name,
roles=roles,
units=units,
as_pandas=as_pandas,
as_dict=as_dict,
)
[docs]def load_interstate94(roles=True, units=True, as_pandas=False):
"""
Regression dataset on traffic volume predicition
The interstate94 dataset is a multivariate time series containing the
hourly traffic volume on I-94 westbound from Minneapolis-St Paul. It is
based on data provided by the `MN Department of Transportation
<https://www.dot.state.mn.us/>`_. Some additional data preparation done by
`John Hogue <https://github.com/dreyco676/Anomaly_Detection_A_to_Z/>`_. The
dataset features some particular interesting characteristics common for
time series, which classical models may struggle to appropriately deal
with. Such characteristics are:
* High frequency (hourly)
* Dependence on irregular events (holidays)
* Strong and overlapping cycles (daily, weekly)
* Annomalies
* Multiple seasonalities
Args:
roles (bool):
Return data with roles set
units (bool):
Return data with units set
as_pandas (bool):
Return data as `pandas.DataFrame` s
Returns:
getml.data.DataFrame:
A DataFrame holding the data described above.
The following DataFrames are returned:
* traffic
Examples:
>>> traffic = getml.datasets.load_interstate94()
>>> type(traffic)
... getml.data.data_frame.DataFrame
For an full analysis of the interstate94 dataset including all necessary
preprocessing steps please refer to `getml-examples
<https://github.com/getml/getml-examples/tree/master/interstate94>`_.
Note:
Roles and units can be set ad-hoc by supplying the respective flags. If
`roles` is `False`, all columns in the returned
:class:`~getml.DataFrames` s have roles
:const:`~getml.data.roles.unused_string` or
:const:`~getml.data.roles.unused_float`. Before using them in an
analysis, a data model needs to be constructed using
:class:`~getml.data.Placeholder` s.
"""
ds_name = "interstate94"
return _load_dataset(
ds_name=ds_name,
roles=roles,
units=units,
as_pandas=as_pandas,
)[0]
[docs]def load_loans(roles=True, units=True, as_pandas=False, as_dict=False):
"""
Binary classification dataset on loan default
The loans dataset is based on financial dataset from the the `CTU Prague
Relational Learning Repository
<https://relational.fit.cvut.cz/dataset/Financial>`_.
The original publication is:
Berka, Petr (1999). Workshop notes on Discovery Challange PKDD'99.
The dataset contains information on 606 successful and 76 unsuccessful
loans. After some preprocessing it contains 5 tables
* `account`: Information about the borrower(s) of a given loan.
* `loan`: Information about the loans themselves, such as the date of creation, the amount, and the planned duration of the loan. The target variable is the status of the loan (default/no default)
* `meta`: Meta information about the obligor, such as gender and geo-information
* `order`: Information about permanent orders, debited payments and account balances.
* `trans`: Information about transactions and accounts balances.
The population table is split into a training and a testing set at 80% of the main population.
Args:
roles (bool):
Return data with roles set
units (bool):
Return data with units set
as_pandas (bool):
Return data as `pandas.DataFrame` s
as_dict (bool):
Return data as dict with `df.name` s as keys and
`df` s as values.
Returns:
tuple:
Tuple containing (sorted alphabetically by `df.name`s) the data as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True) or
dict:
if `as_dict` is `True`: Dictionary containing the data as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True). The keys correspond to the name of the DataFrame on the
:mod:`~getml.engine`.
The following DataFrames are returned:
* account
* loan
* meta
* order
* trans
Examples:
>>> loans = getml.datasets.load_loans(as_dict=True)
>>> type(loans["population_train"])
... getml.data.data_frame.DataFrame
For an full analysis of the loans dataset including all necessary
preprocessing steps please refer to `getml-examples
<https://github.com/getml/getml-examples/tree/master/loans>`_.
Note:
Roles and units can be set ad-hoc by supplying the respective flags. If
`roles` is `False`, all columns in the returned
:class:`~getml.DataFrames` s have roles
:const:`~getml.data.roles.unused_string` or
:const:`~getml.data.roles.unused_float`. Before using them in an
analysis, a data model needs to be constructed using
:class:`~getml.data.Placeholder` s.
"""
ds_name = "loans"
return _load_dataset(
ds_name=ds_name,
roles=roles,
units=units,
as_pandas=as_pandas,
as_dict=as_dict,
)
# TODO: Update notebook
[docs]def load_occupancy(roles=True, as_pandas=False, as_dict=False):
"""
Binary classification dataset on occupancy detection
The occupancy detection data set is a very simple multivariate time series
from the `UCI Machine Learning Repository
<https://archive.ics.uci.edu/ml/datasets/Occupancy+Detection+>`_. It is a
binary classification problem. The task is to predict room occupancy
from Temperature, Humidity, Light and CO2.
The original publication is:
Candanedo, L. M., & Feldheim, V. (2016). Accurate occupancy detection of an
office room from light, temperature, humidity and CO2 measurements using
statistical learning models. Energy and Buildings, 112, 28-39.
Args:
roles (bool):
Return data with roles set
as_pandas (bool):
Return data as `pandas.DataFrame` s
as_dict (bool):
Return data as dict with `df.name` s as keys and
`df` s as values.
Returns:
tuple:
Tuple containing (sorted alphabetically by `df.name`s) the data as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True) or
dict:
if `as_dict` is `True`: Dictionary containing the data as
:class:`~getml.DataFrame` s or :class:`pandas.DataFrame` s (if `as_pandas`
is True). The keys correspond to the name of the DataFrame on the
:mod:`~getml.engine`.
The following DataFrames are returned:
* population_train
* population_test
* population_validation
Examples:
>>> population_train, population_test, _ = getml.datasets.load_occupancy()
>>> type(occupancy_train)
... getml.data.data_frame.DataFrame
For an full analysis of the occupancy dataset including all necessary
preprocessing steps please refer to `getml-examples
<https://github.com/getml/getml-examples/tree/master/occupancy>`_.
Note:
Roles can be set ad-hoc by supplying the respective flag. If
`roles` is `False`, all columns in the returned
:class:`~getml.DataFrames` s have roles
:const:`~getml.data.roles.unused_string` or
:const:`~getml.data.roles.unused_float`. This dataset contains no units.
Before using them in an analysis, a data model needs to be constructed
using :class:`~getml.data.Placeholder` s.
"""
ds_name = "occupancy"
return _load_dataset(
ds_name=ds_name,
roles=roles,
as_pandas=as_pandas,
as_dict=as_dict,
)