spkit.ml.DecisionTree

class spkit.ml.DecisionTree(min_samples_split=2, min_impurity=1e-07, max_depth=inf, thresholdFromMean=False)

Super class of RegressionTree and ClassificationTree.

Optimizing depth:

• In this version, a very large value can be used to build a tree and later can be shrink to lower depth (d),

• using “.updateTree(shrink=True, max_depth=d)”. The optimal value of depth for given data can be choosing by analysing learning curve using “.getLcurve” method and/or “plotTree(Measures=True)”

Parameters:

max_depth: int:>0,

• maximum depth to go for tree, default is Inf, which leads to overfit

• decrease the max depth to reduce the overfitting.

min_samples_split: int

• minimum number of samples to split further

min_impurity: float:

• minimum impurity (or gain) to split

thresholdFromMean: bool, default = False.

• if threshold is selcted from mean of two

• concecutive unique values of selected a an unique value of feaure.

  Only applicable to float or int type features, not to catogorical type.

Methods

DictDepth(DT[, n])	Get the maximum depth of dictionary
fit(X, y)	Building a tree and saving in an dictionary at self.tree
getLcurve([Xt, yt, Xs, ys, measure])	Getting Learning Curve By supplying Trainign and Testing Data, compute the given measure for each level of depth
getTreeDepth([DT])	Get the maximum depth of the tree
get_tree()	Extract built tree
plotLcurve([ax, title])	Plotting Learning Curve
plotTree([scale, show, showtitle, ...])	Plot Decision Tree
plotTreePath(path[, ax, fig])	Plotting the path of the tree
predict(X[, max_depth, treePath])	Predicting labels
predict_proba(X[, label_counts, max_depth, ...])	Predicting probabilties of labels
pruneTree(DT)	Prunning the tree
set_featureNames([feature_names])	Set Feature names
set_xyNode(DT[, lxy, xy, rxy, ldiff])	Setting the xy location of each node
showTree(DT[, DiffBranchColor, ...])	Helper function fot plotTree
shrinkTree(DT[, max_depth])	Shinking the tree
treeDepth(DT[, mx])	Compute the maximum depth of the tree
updateTree([DT, shrink, max_depth])	Updating Tree with Shrinking and Pruning

DictDepth(DT, n=0)

Get the maximum depth of dictionary

fit(X, y)

Building a tree and saving in an dictionary at self.tree

Parameters:

X: ndarray

• (number of sample, number of features)

y: list of 1D array

• labels

verbose: int (default=0)

• 0 no progress or tree

• 1 - show progress

• 2 - show tree

getLcurve(Xt=None, yt=None, Xs=None, ys=None, measure='acc')

Getting Learning Curve By supplying Trainign and Testing Data, compute the given measure for each level of depth

Parameters:

Xt, yt: training data

Xs, ys: testing data

measure: str, metric

default=’acc’ for accuracy

Returns:

Lcurve: dict,

• as training and testing measures at each level of depth

getTreeDepth(DT=None)

Get the maximum depth of the tree

get_tree()

Extract built tree

plotLcurve(ax=None, title=True)

Plotting Learning Curve

After computing Learning Curve using getLcurve, Learnign curve can be plotted

Parameters:

ax: axis to plot

title: bool, if to show title

plotTree(scale=True, show=True, showtitle=True, showDirection=False, DiffBranchColor=True, legend=True, showNodevalues=True, showThreshold=True, hlines=False, Measures=False, dcml=0, leaf_labels=None)

Plot Decision Tree

Parameters:

scale:

show:

showtitle:

showDirection:

DiffBranchColor:

legend:

showNodevalues:

showThreshold:

hlines:

Measures:

dcml:

leaf_labels:

plotTreePath(path, ax=None, fig=None)

Plotting the path of the tree

predict(X, max_depth=inf, treePath=False)

Predicting labels

USE “max_depth” to limit the depth of tree for expected values

Compute expected value for each sample in X, up till depth=max_depth of the tree

For classification: Expected value is a label with maximum probabilty among the samples at leaf nodes. For probability and count of each labels at leaf node, use “.predict_proba” method

For Regression: Expected value is a mean value of smaples at the leaf node. For Standard Deviation and number of samples at leaf node, use “.predict_proba” method

Parameters:

X: ndarray

• (number of sample, number of features)

max_depth: int, default=np.inf

• maximum depth of the tree to use for the prediction

treePath: bool, Default=False

• if True, path of the tree is also returned, such as ‘TTFTFT…’

Returns:

y: list of 1D array

• labels

• and paths, if treePath is True

predict_proba(X, label_counts=False, max_depth=inf, treePath=False)

Predicting probabilties of labels

USE “max_depth” to limit the depth of tree for expected values Compute probabilty/SD for labeles at the leaf till max_depth level, for each sample in X

For classification: Returns the Probability of samples one by one and return the set of labels label_counts=True: Includes in the return, the counts of labels at the leaf

For Regression: Returns the standard deviation of values at the leaf node. Mean value is returened with “.predice()” method label_counts=True: Includes in the return, the number of samples at the leaf

treePath=True: Includes the path of tree for each sample as string

Parameters:

X: ndarray

• (number of sample, number of features)

max_depth: int, default=np.inf

• maximum depth of the tree to use for the prediction

label_counts: bool, default=False

• If true

  • count of each class labels are returned (classification)

  • count of samples at the leaf returned (regression)

treePath: bool, Default=False

• if True, path of the tree is also returned, such as ‘TTFTFT…’

Returns:

y_prob: 1D/2D array

• labels

y_counts: count of labels

y_paths: path of each samples

pruneTree(DT)

Prunning the tree

Parameters:

DT: A decision Tree

Returns:

DT: Prunned Tree

set_featureNames(feature_names=None)

Set Feature names

set_xyNode(DT, lxy=[0, 1], xy=[1, 1], rxy=[2, 1], ldiff=1)

Setting the xy location of each node

in the tree

showTree(DT, DiffBranchColor=False, showNodevalues=True, showThreshold=True)

Helper function fot plotTree

shrinkTree(DT, max_depth=inf)

Shinking the tree

Parameters:

DT: A decision Tree

max_depth: int

• depth, to which tree is to be shrinked

Returns:

DT: Shrinked Tree

treeDepth(DT, mx=0)

Compute the maximum depth of the tree

updateTree(DT=None, shrink=False, max_depth=inf)

Updating Tree with Shrinking and Pruning

Parameters:

DT: A decision Tree,

• if None, root tree self.tree is used

max_depth: int

• depth, to which tree is to be shrinked

shrink: bool,

• if to shrink the tree

Returns:

DT: Updated Tree