Hi, in order to improve the model on a specific task, I have updated the vocabulary and the language model but I still get mis-prediction on few words. For instance I often get 'udr' instead of 'edr'. In my case I will never have to use the word 'udr' and I was wondering if it was possible to delete this word from the dictionnary to avoid this type of misprediction.
Do I simply need to remove it from the fr.dic and fr.vocab an then recompile the graph or is there another way to do this.
Hi, in order to improve the model on a specific task, I have updated the vocabulary and the language model but I still get mis-prediction on few words. For instance I often get 'udr' instead of 'edr'. In my case I will never have to use the word 'udr' and I was wondering if it was possible to delete this word from the dictionnary to avoid this type of misprediction. Do I simply need to remove it from the fr.dic and fr.vocab an then recompile the graph or is there another way to do this.