Data Sources

Input Data Types for Transforms and Trainers

The transforms and trainers in nimbusml support additional types of data sources as inputs, besides arrays and matrices. The supported data sources are:

  • list - for dense data

  • numpy.ndarray and numpy.array - for dense data

  • scipy.sparse_csr - for sparse data

  • pandas.DataFrame and pandas.Series - for dense data with a schema

  • FileDataStream - for dense data with a schema.

Data in Lists

A list is a natural way to represent data layout.

Most trainers accept a list of values for X and y, as shown in the example below. The values should be valid. If NaNs are present, they need to be imputed or filtered before feeding to a learner. Additionally, the dimensions of X and y should be congruent, and the number of elements in the examples should be identical and of the same type.

For transforms, the data dimension and composition requirements are less rigid, and depend entirely on the transform.

Example:

from nimbusml.linear_model import LogisticRegressionBinaryClassifier
X = [[.1, .2],[.2, .3]]
y = [0,1]
LogisticRegressionBinaryClassifier().fit(X,y).predict(X)

Data in Numpy Arrays

The data source can also be a numpy.array or numpy.ndarray object. The dimension and data composition requirements are similar to the ones for lists.

Example:

import numpy as np
from nimbusml.linear_model import LogisticRegressionBinaryClassifier

X = np.array([[.1, .2],[.2,.3]])
y = np.array([0,1])
LogisticRegressionBinaryClassifier().fit(X,y).predict(X)

Data in DataFrames and Series

Data in pandas.DataFrame and pandas.Series classes may also be used with trainers and transforms. One advantage of dataframes is that column names can be user defined and used to specify which columns should be transformed (see Column Operations for Transforms).

Example:

import pandas as pd
from nimbusml.linear_model import LogisticRegressionBinaryClassifier
X = pd.DataFrame(data=dict(
        Sepal_Length=[2.5, 2.6],
        Sepal_Width=[.75, .9],
        Petal_Length=[2.5, 2.5],
        Petal_Width=[.8, .7]))

y = pd.DataFrame([0,1])
LogisticRegressionBinaryClassifier().fit(X,y).predict(X)

Data from a FileDataStream

Data in a file can be processed directly without preloading into memory. The data can be streamed efficiently using nimbusml.FileDataStream class, which replaces the X and y arguments of the fit() and predict() methods of trainers. Users can create a nimbusml.FileDataStream class using the FileDataStream.read_csv() function or based on a DataSchema. More details about constructing a nimbusml.DataSchema is discussed in Schema.

Example:

from nimbusml.datasets import get_dataset
from nimbusml import Pipeline, FileDataStream, DataSchema
from nimbusml.ensemble import LightGbmClassifier

path = get_dataset('infert').as_filepath()

schema = DataSchema.read_schema(path, sep=',')
ds = FileDataStream(path, schema = schema)

#Equivalent to
#ds = FileDataStream.read_csv(path, sep=',')

pipeline = Pipeline([
    LightGbmClassifier(feature=['age', 'parity', 'induced'], label='case')
    ])

pipeline.fit(ds)
pipeline.predict(ds)

Output Data Types of Transforms

When used inside a sklearn.pipeline.Pipeline, the return type of all of the transforms is a pandas.DataFrame.

When used individually or inside a nimbusml.Pipeline that contains only transforms, the default output is a pandas.DataFrame. To instead output an IDataView, pass as_binary_data_stream=True to either transform() or fit_transform(). To output a sparse CSR matrix, pass as_csr=True. See nimbusml.Pipeline for more information.

Note, when used inside a nimbusml.Pipeline, the outputs are often stored in a more optimized VectorDataViewType, which minimizes data conversion to dataframes. When several transforms are combined inside an nimbusml.Pipeline, the intermediate transforms will store the data in the optimized format and only the last transform will return a pandas.DataFrame (or IDataView/CSR; see above).