#!/usr/bin/env python3
# -*- coding: utf-8 -*-
#-----------------------------------------------------------------------------
# Copyright (c) 2023, IBM Corp.
# All rights reserved.
#
# Distributed under the terms of the BSD Simplified License.
#
# The full license is in the LICENSE file, distributed with this software.
#-----------------------------------------------------------------------------
"""
In many classification applications it may be required or desirable not
only to accurately classify instances, but also to inspect the model.
The inspection makes it possible to explain its decisions, modify it,
or combine with some existing background knowledge. In such applications,
where both the high classification accuracy and human-readability of the
model are required, the method of choice is typically going to be
decision trees.
A decision tree is a hierarchical structure that represents a classification
model using a "divide and conquer" approach. Internal tree nodes represent
splits applied to decompose the data set into subsets, and terminal nodes,
also referred to as leaves, assign class labels to sufficiently small
or uniform subsets. Splits are specified by logical conditions based on
selected single attributes, with a separate outgoing branch corresponding
to each possible outcome.
The concept of decision tree construction is to select splits that decrease
the impurity of class distribution in the resulting subsets of instances,
and increase the domination of one or more classes over the others.
The goal is to find a subset containing only or mostly instances of one
class after a small number of splits, so that a leaf with that class
label is created. This approach promotes simple trees, which typically
generalize better.
"""
from typing import List
from nzpyida.frame import IdaDataFrame
from nzpyida.base import IdaDataBase
from nzpyida.analytics.predictive.classification import Classification
from nzpyida.analytics.utils import q
[docs]
class DecisionTreeClassifier(Classification):
"""
Decision tree based classifier.
"""
def __init__(self, idadb: IdaDataBase, model_name: str):
"""
Creates the classifier class.
Parameters
----------
idada : IdaDataBase
database connector
model_name : str
model name - if it exists in the database, it will be used, otherwise
it must be trained using fit() function before prediction or scoring is called.
"""
super().__init__(idadb, model_name)
self.fit_proc = 'DECTREE'
self.predict_proc = 'PREDICT_DECTREE'
self.has_print_proc = True
[docs]
def fit(self, in_df: IdaDataFrame, target_column: str, id_column: str=None,
in_columns: List[str]=None, col_def_type: str=None, col_def_role: str=None,
col_properties_table: str=None, weights: str=None, eval_measure: str=None,
min_improve: float=0.02, min_split: int=50, max_depth: int=10, val_table: str=None,
val_weights: str=None, qmeasure: str=None, statistics: str=None):
"""
Grows the decision tree and stores its model in the database.
Parameters
----------
in_df : IdaDataFrame
the input data frame
target_column : str
the input table column representing the class
id_column : str, optional
the input table column identifying a unique instance id - if skipped,
the input data frame indexer must be set and will be used as an instance id
in_columns : str, optional
the list of input table columns with special properties.
Each column is followed by one or several of the following properties:
its type: ':nom' (for nominal), ':cont' (for continuous).
Per default, all numerical types are continuous, other types are nominal.
its role: ':id', ':target', ':input', ':ignore'.
(Remark: ':objweight' is unsupported, i.e. ':objweight' same as ':ignore').
(Remark: ':colweight(<wgt>)' is unsupported, i.e. ':colweight(<wgt>)' same as
':col-weight(1)' same as ':input').
If the parameter is undefined, all columns of the input table have default properties.
col_def_type : str, optional
default type of the input table columns. Allowed values are 'nom' and 'cont'.
If the parameter is undefined, all numeric columns are considered continuous,
other columns nominal.
col_def_role : str, optional
default role of the input table columns. Allowed values are 'input' and 'ignore'.
If the parameter is undefined, all columns are considered 'input' columns.
col_properties_table : str, optional
the input table where column properties for the input table columns are stored.
The format of this table is the output format of stored procedure
nza..COLUMN_PROPERTIES(). If the parameter is undefined, the input table column
properties will be detected automatically.
(Remark: colPropertiesTable with "COLROLE" column with value 'objweight' is unsupported,
i.e. same as 'ignore')
(Remark: colPropertiesTable with "COLWEIGHT" column with value '<wgt>' is unsupported,
i.e. same as '1')
weights : str, optional
the input table containing optional instance or class weights for the input
table columns.
If the parameter is undefined, we assume that the weights are uniformly equal to 1.
The <weights> table contains following columns:
weight: a numeric column containing the instance or class weight,
id: a column to be joined with the <id> column of <intable>, defining
instance weights,
class: a column to be joined with the <target> column of <intable>, defining
class weights.
The id or class column can be missing, at least one of them must be present.
For instances or classes not occurring in this table, weights of 1 are assumed.
eval_measure : str, optional
the class impurity measure used for split evaluation.
Allowed values are 'entropy' and 'gini'
min_improve : float, optional
the minimum improvement of the split evaluation measure required
min_split : int, optional
the minimum number of instances per tree node that can be split
max_depth : int, optional
the maximum number of tree levels (including leaves)
val_table : str, optional
the input table containing the validation dataset.
If this parameter is undefined, no pruning will be performed.
val_weights : str, optional
the input table containing optional instance or class weights for the
validation dataset.
It is similar to the <weights> table.
qmeasure : str, optional
the quality measure for pruning.
Allowed values are Acc or wAcc.
statistics : str, optional
flags indicating which statistics to collect.
Allowed values are: none, columns, values:n, all.
If statistics=none, no statistics are collected.
If statistics=columns, statistics on the input table columns like mean
value are collected.
If statistics=values:n with n a positive number, statistics about the
columns and the column
values are collected. Up to <n> column value statistics are collected:
If a nominal column contains more than <n> values, only the <n> most
frequent column statistics are kept.
If a numeric column contains more than <n> values, the values will be
discretized and the statistics will be collected on the discretized values.
Indicating statistics=all is equal to statistics=values:100.
"""
params = {
'id': q(id_column),
'target': q(target_column),
'incolumn': q(in_columns),
'coldeftype': col_def_type,
'coldefrole': col_def_role,
'colpropertiestable': col_properties_table,
'weights': weights,
'eval': eval_measure,
'minimprove': min_improve,
'minsplit': min_split,
'maxdepth': max_depth,
'valtable': val_table,
'valweights': val_weights,
'qmeasure': qmeasure,
'statistics': statistics
}
self._fit(in_df=in_df, params=params)
[docs]
def predict(self, in_df: IdaDataFrame, out_table: str=None,
id_column: str=None, prob: bool=False, out_table_prob: str=None) -> IdaDataFrame:
"""
Makes predictions based on this model. The model must exist.
Parameters
----------
in_df : IdaDataFrame
the input data frame
out_table : str, optional
the output table where the predictions will be stored
id_column : str, optional
the input table column identifying a unique instance id
Default: id column used to build the model
prob : bool, optional
the flag indicating whether the probability of the predicted class should be included
into the output table or not
out_table_prob : str, optional
if specified, the probability output table where class probability predictions
will be stored
Returns
-------
IdaDataFrame
the data frame containing row identifiers and predicted target values
"""
params = {
'id': q(id_column),
'prob': prob,
'outtableprob': out_table_prob
}
return self._predict(in_df=in_df, params=params, out_table=out_table)