qiaomaiya/hmc

Python Module for decision tree based hierarchical multi-classification

hmc

Decision Tree Hierachical Multi-Classifier

A thin wrapper for sklearn DecisionTreeClassifier for hierarchical classes, implementing HSC-Clus [1].

Define a class hierarchy. Class hierarchies are constructed with the root node. Nodes are added with child parent pairs.

import hmc
ch = hmc.ClassHierarchy("colors")
ch.add_node("light", "colors")
ch.add_node("dark", "colors")
ch.add_node("white", "light")
ch.add_node("black", "dark")
ch.add_node("gray", "dark")
ch.add_node("slate", "gray")
ch.add_node("ash", "gray")

Pretty print it.

>>> ch.print_()
└─colors
  ├─dark
  │ ├─black
  │ └─gray
  │   ├─ash
  │   └─slate
  └─light
    └─white

Load some data from the included functions and split for training. The class hierarchy itself can also be loaded from the module.

ch = hmc.load_shades_class_hierachy()
X, y = hmc.load_shades_data()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.50)

Each parent node, of at least one child, will generate a decision tree classification stage. Stages are assigned depth first, ascending alpha.

>>> dt.stages
[{'classes': ['dark', 'light', 'colors'],
  'depth': 0,
  'labels': ['dark', 'light'],
  'stage': 'colors'},
{'classes': ['black', 'gray', 'dark'],
  'depth': 1,
  'labels': ['black', 'gray'],
  'stage': 'dark'},
{'classes': ['ash', 'slate', 'gray'],
  'depth': 2,
  'labels': ['ash', 'slate'],
  'stage': 'gray'},
{'classes': ['white', 'light'],
  'depth': 1,
  'labels': ['white'],
  'stage': 'light'}]

The hmc.DecisionTreeHierarchicalClassifier is idiomatic to the sklearn tree.DecisionTreeClassifier. Fit, predict and score the same way. Traditional multi-classification average accuracy is comparable.

from sklearn import tree
dt = tree.DecisionTreeClassifier()
dt = dt.fit(X_train, y_train)
dt_predicted = dt.predict(X_test)
dt_accuracy = dt.score(X_test, y_test)

dth = hmc.DecisionTreeHierarchicalClassifier(ch)
dth = dth.fit(X_train, y_train)
dth_predicted = dth.predict(X_test)
dth_accuracy = dth.score(X_test, y_test)

>>> dt_accuracy
0.48526522593320237
>>> dth_accuracy
0.45776031434184677

Additional hierarchical multi-classification specific metrics [2] are provided.

import hmc.metrics as metrics

>>> metrics.accuracy_score(ch, y_test, dth_predicted)
0.45776031434184677
>>> metrics.precision_score_ancestors(ch, y_test, dth_predicted)
0.8
>>> metrics.recall_score_ancestors(ch, y_test, dth_predicted)
0.8052190121155638
>>> metrics.f1_score_ancestors(ch, y_test, dth_predicted)
0.8026010218300047
>>> metrics.precision_score_descendants(ch, y_test, dth_predicted)
0.647191011235955
>>> metrics.recall_score_descendants(ch, y_test, dth_predicted)
0.6260869565217392
>>> metrics.f1_score_descendants(ch, y_test, dth_predicted)
0.63646408839779

Ancestor and Descendant precision and recall scores are calculated as the fraction of shared ancestor or descendant classes over the sum of either the predicted or true class for precision and recall respectively [3].

true = ['dark', 'white', 'gray']

pred_sibling = ['dark', 'white', 'black']

>>> metrics.accuracy_score(ch, true, pred_sibling)
0.66666666666666663
>>> metrics.precision_score_ancestors(ch, true, pred_sibling)
0.8
>>> metrics.precision_score_descendants(ch, true, pred_sibling)
0.8571428571428571

pred_narrower = ['dark', 'white', 'ash']

>>> metrics.accuracy_score(ch, true, pred_narrower)
0.66666666666666663
>>> metrics.precision_score_ancestors(ch, true, pred_narrower)
0.8333333333333334
>>> metrics.precision_score_descendants(ch, true, pred_narrower)
1.0

pred_broader = ['dark', 'white', 'dark']

>>> metrics.accuracy_score(ch, true, pred_broader)
0.66666666666666663
>>> metrics.precision_score_ancestors(ch, true, pred_broader)
1.0
>>> metrics.precision_score_descendants(ch, true, pred_broader)
0.8181818181818182

1. Vens, C., Struyf, J., Schietgat, L., Džeroski, S., & Blockeel, H. (2008). Decision trees for hierarchical multi-label classification. Mach Learn Machine Learning, 73(2), 185-214.
2. Sokolova, M., & Lapalme, G. (2009). A systematic analysis of performance measures for classification tasks. Information Processing & Management, 45(4), 427-437. doi:10.1016/j.ipm.2009.03.002
3. Costa, E., Lorena, A., Carvalho, A., & Freitas, A. (2007). A review of performance evaluation measures for hierarchical classifiers. In Proceedings of the AAAI 2007 workshop "Evaluation methods for machine learning" (pp. 1–6).