I had some trouble trying to train POS tagger models on my own data sets using one of the 5 other algorithms.
The quickstart guide contains the following options under optimizer:
algorithm: perceptron, softmax, adagrad, agagrad-mini-batch, agadelta-mini-batch, agagrad-regression.
When editing the config file the following of those options are incorrect and need to be replaced with:
softmax - softmax-regression
agadelta-mini-batch - adadelta-mini-batch
agagrad-mini-batch - adagrad-mini-batch
agagrad-regression - adagrad-regression
I also noticed that ffnn-softmax is a possible option listed in the source code although not specified in the quickstart guide - when testing it it fails reporting a null pointer exception - will this option be available in future or is this a possible error?
FFNN was developed for our internal experiments and it is functional but we didn't make it publicly available due to a lack of wrapper APIs. Thanks for those correction!
Hi,
I had some trouble trying to train POS tagger models on my own data sets using one of the 5 other algorithms.
The quickstart guide contains the following options under optimizer: algorithm: perceptron, softmax, adagrad, agagrad-mini-batch, agadelta-mini-batch, agagrad-regression.
When editing the config file the following of those options are incorrect and need to be replaced with:
I also noticed that ffnn-softmax is a possible option listed in the source code although not specified in the quickstart guide - when testing it it fails reporting a null pointer exception - will this option be available in future or is this a possible error?