Closed ahmedmazari-dhatim closed 6 years ago
Confusion matrix play a more important role in classification. For outputs like text, other metrics, editing distance for example, are more meaningful.
@meijieru thank you for your answer. We can say that sequence prediction is a classification problem, isn't it ? Because we have an input which is an image that contains a sequence of character and its label is that sequence. Can't we use confusion matrix in this case ? otherwise, what do you propose ?
@ahmedmazari-dhatim I went through a paper of alex graves, he gave this formulae 100 * (1- (insertion+deletions+subsitutions)/total number of words)) Here, words with insertion, deletion & substitution are counted and accuracy is obtained.
Hi @rremani ,
Do you mind sharing with us this article ?
l'm not sure to understand the relationship between confusion matrix and the the formula 100 * (1- (insertion+deletions+subsitutions)/total number of words)) ?
Thank you
sure, It's a chapter from a book, 'Neural Networks for Handwriting Recognition'. DOI number: doi.org/10.1007/978-3-642-24049-2_2 You can get this through sci-hub ;) cheers
http://doi.org.sci-hub.io/10.1007/978-3-642-24049-2_2 is not working
Thank you
Hello,
l'm thinking about making confusion matrix for CRNN. However the output classes are variable sequences and not as standard output classes as 0 1 2 .. or A B C . Any idea of how making confusion matrix for CRNN.
Thank you