README.md

CPT

CPT is a cython open-source implementation of the Compact Prediction Tree algorithm using multithreading.

This is an implementation of the following research papers

http://www.philippe-fournier-viger.com/ADMA2013_Compact_Prediction_trees.pdf

http://www.philippe-fournier-viger.com/spmf/PAKDD2015_Compact_Prediction_tree+.pdf

Simple example

You can test the model with the following code

from cpt.cpt import Cpt
model = Cpt()

model.fit([['hello', 'world'],
           ['hello', 'this', 'is', 'me'],
           ['hello', 'me']
          ])

model.predict([['hello'], ['hello', 'this']])
# Output: ['me', 'is']

Sklearn Example

CPT is compatible with sklearn, you can, for instance, use GridSearch on it.

from sklearn.base import BaseEstimator
from cpt.cpt import Cpt

class SKCpt(Cpt, BaseEstimator):
    def __init__(self, split_index=0, noise_ratio=0, MBR=0):
        super().__init__(split_index)
        self.noise_ratio = noise_ratio
        self.MBR = MBR

    def predict(self, sequences):
        return super().predict(sequences, self.noise_ratio, self.MBR)

    def score(self, X):
        predictions = self.predict(list(map(lambda x: x[self.split_index:-1], X)))
        score = sum([predictions[i] == X[i][-1] for i in range(len(X))]) / len(X) * 100
        return score

data = [['hello', 'world'], ['hello', 'cpt']]


from sklearn.grid_search import GridSearchCV

tuned_params = {'MBR': [0, 5], 'split_index': [0, 1, 5]}

gs = GridSearchCV(SKCpt(), tuned_params)

gs.fit(data)

You can test it with more data to have more relevant tuning.

Features

Train

The model can be trained with the train method.

If needed the model can be retrained with the same methods. It adds new sequences to the model and do not remove the old ones.

Multithreading

The predictions are launched by default with multithreading with OpenMP.

The predictions can also be launched in a single thread with the option multithread=False in the predict method.

By default the number of threads equals the number of cores, you can control it by setting the following environment variable OMP_NUM_THREADS.

Pickling

You can pickle the model to save it, and load it later via pickle library.

from cpt.cpt import Cpt
import pickle


model = Cpt()
model.fit([['hello', 'world']])

dumped = pickle.dumps(model)

unpickled_model = pickle.loads(dumped)

print(model == unpickled_model)