This is one of those.
It can train on data like this:
documents = [
{'words':['yada', 'yada', 'yada'], 'class':'yada'},
{'words':['bla', 'bla', 'bla'], 'class':'bla'}
]
nbc = NBC(documents)
nbc.save('trainedclassifier.pickle')and classify new documents:
document = ['yada', 'yada', 'bla']
nbc = NBC.load('trainedclassifier.pickle')
result = nbc.classify(document)and give reports on:
- Accuracy
- Precision
- Recall
- F1 measure
and print during training:
Started training on 12928 documents...
0.277s 75% left...
0.594s 50% left...
0.931s 25% left...
1.277s 0% left...
Calculating probabilities...
done.
and also train again and again and again
nbc = NBC()
for documents in documents_collection:
nbc.train(documents)
# wait some time ...