Run scikit-learn Functions using Teradata Introduced Arguments - Run scikit-learn Functions using Teradata Introduced Arguments - Teradata Package for Python

An alternative to the legacy scikit-learn arguments, teradataml OpenSourceML introduces new arguments data, feature_columns, label_columns and group_columns to run functions of scikit-learn classes. The functionality is similar to what was described in the previous example, but avoids additional step of creating X and y DataFrames from select() API.

Description of Teradata Introduced Arguments

• data: teradataml DataFrame containing columns specified in the following three arguments
• feature_columns: Column name or list of column names (as str) containing features to be trained on or tested for
• label_columns: Column name or list of column names (as str) containing labels which are used for training
• group_columns: Column name or list of column names (as str) containing group columns needed for classes in model selection module

The column names provided in the feature_columns, label_columns and group_columns arguments should be present in the teradataml DataFrame specified by data.

Example using Teradata introduced arguments

• Generate data.

  df_train = DataFrame("test_classification")

  df_train
  

                 col1                    col2                  col3                   col4    label
  -1.1305820619922704     -0.0202959251414216   -0.7102336334648424    -1.4409910829920618        0
  -0.2869200001717422     -0.7169529842687833   -0.9865850877151031     -0.848214734984639        0
  -2.5604297516143286      0.4022323367243113   -1.1007419820939435    -2.9595882598466674        0
   0.4223414406917685     -2.0391144030275625    -2.053215806414584    -0.8491230457662061        0
   0.7216694959200303     -1.1215566442946217   -0.8318398647044646     0.1507420965953343        0
  -0.9861325665504175      1.7105310292848412    1.3382818041204743    -0.0853410902974293        1
  -0.5097927128625588      0.4926589443964751    0.2482067293662461    -0.3095907315896897        1
   0.1833246820582146      -0.774610353732039    -0.766054694735782    -0.2936686329125327        0
  -0.4032571038523639      2.0061840569850093    2.0275124771199318     0.8508919440196763        1
  -0.0715602561938739      0.2295539000122874   0.21654344712218576     0.0652739792167357        1
  

  feature_columns = ["col1", "col2", "col3", "col4"]

  label_columns = "label"

• Create an instance of scikit-learn LinearSVC object through teradataml open-source machine learning function 'td_sklearn'.

  from teradataml import td_sklearn as osml
  

  linear_svc = osml.LinearSVC(loss="hinge", tol=0.01)
  

  linear_svc
  

  LinearSVC(loss='hinge', tol=0.01)

• Train the model.

  linear_svc.fit(data=df_train, feature_columns=feature_columns, label_columns=label_columns)
  

  LinearSVC(loss='hinge', tol=0.01)

• Get predictions on test data.

  teradataml OpenSourceML returns teradataml DataFrame with both features and labels for predict and similar functions.

  linear_svc.predict(data=df_train, feature_columns=feature_columns)

                col1                  col2                 col3                 col4    linearsvc_predict_1
    1.23195055037206     -1.53949525926716    -0.99510531686895    0.511600970144431                    0.0
    1.26780439921386     -1.80170792990881    -1.27034986297172    0.379112827728592                    0.0
  -0.869536951900537      1.99896877100815     1.73590334857413    0.257374908024379                    1.0
    1.43370121321312     -1.75423983622451    -1.11573423222268    0.620716743476382                    0.0
   -1.05286597780779    -0.641515112432539    -1.36672011108273    -1.76399738946526                    0.0
  -0.345538051487565     -2.29672333669221    -2.81180710379968     -1.9931134219738                    0.0
    -1.2573206891836     -2.14861012008993    -3.19826339415065    -3.04373306805433                    0.0
  -0.205721671526727      1.75895320535307     1.86752027575658    0.932664558487293                    1.0
   -3.58754622394712      0.29181935785016    -1.85016852734401    -4.33105451025007                    0.0
   -2.52159550020822      2.47822554412282     1.27458363813847    -1.50328319686837                    1.0
  

• Access attributes.

  linear_svc.intercept_

  array([0.55058172])