tdextensions 1.0.0rc1

Teradata Consulting Python Client Extensions

Teradata ML Extensions

Extensions to the core teradataml library by Teradata Consulting to aid in field development work around BYOM, STO, RTO and AnalyticOps solutions.

Installation

You can install via pip.

pip install tdextension

Usage

You must use the same version of python on your client side as is used in Teradata (3.6+ at the time of writing). The reason for this is due to differences in serialization between versions of python (e.g. between 3.5 and 3.6).

from teradataml.dataframe.dataframe import DataFrame
from tdextensions.distributed import DistDataFrame, DistMode
from teradataml import create_context
import pandas as pd
import numpy as np

pd.options.display.max_colwidth = 250

engine = create_context(host="localhost", username="ivsm_user", password="ivsm_user")

A simple map row example where we multiple the value of two columns on a row by row basis

def my_fun(row):
    return np.array([row.idx, row.sepal_length * row.sepal_width])

df = DistDataFrame("iris_train", dist_mode=DistMode.STO, sto_id="my_dumb_map")
df = df.map(lambda row: my_fun(row), 
            returns=[["idx", "INTEGER"], ["my_derived_col", "INTEGER"]])

df.head()

A more advanced example where we train a different model for each partition of a dataset

from sklearn.ensemble import RandomForestClassifier
import base64
import dill

def train(partition):
    X = partition[['sepal_length', 'sepal_width', 'petal_length', 'petal_width']]
    y = partition[['species']]

    clf = RandomForestClassifier()
    clf.fit(X, y.values.ravel())

    return np.array([[partition.species.iloc[0], "my_model_id", base64.b64encode(dill.dumps(clf))]])

df = DistDataFrame("iris_train", dist_mode=DistMode.STO, sto_id="my_model_train")
df = df.map_partition(lambda partition: train(partition), 
                      partition_by="species", 
                      returns=[["partition_id", "VARCHAR(255)"], 
                               ["model_id", "VARCHAR(255)"],
                               ["model_artefact", "CLOB"]])
df.to_pandas().head()

Permissions

SET SESSION SEARCHUIFDBPATH = <database>;
GRANT EXECUTE procedure on <db> to <user>;
GRANT EXECUTE procedure on SYSUIF to <user>;
GRANT CREATE external procedure on <db> to <user>;
GRANT EXECUTE FUNCTION ON TD_SYSFNLIB.SCRIPT to <user>;
GRANT EXECUTE ON  SYSUIF.DEFAULT_AUTH TO <user>;