lanyihuai

we name the model class as RobertaGen, not RoBERTaGen,so you setup --model=RobertaGen to find the model. the framework is not supporting finding model through different uppercase or lowercase letters yet.

TRNN has two stages to generate. if the equation is ```[n0 + n1 - n2 ]``` first generate template, like ```[n0 n1 n2]```, `````` is unkown operator token. the label...

you can view this page https://mwptoolkit.readthedocs.io/en/latest/mwptoolkit/mwptoolkit.utils.preprocess_tool.number_transfer.html we name constants as generate numbers. these numbers will be added to decoder vocabulary automatically. you could see 'min_generate_keep', a parameter of function number_transfer,...

I'm not sure the specific reason. But you could check the value of these variables below. ``` dataset.copy_nums dataset.num_start dataset.generate_list dataset.out_idx2symbol ``` they are all about the decoder's vocabulary of...

Code for building number stack MWPToolkit/mwptoolkit/data/dataset/abstactdataset.py line 192 ``` def _build_num_stack(self, equation, num_list): num_stack = [] for word in equation: temp_num = [] flag_not = True if word not in...

Yes, i'm working on this, I will update if I test its correctness and it significantly improves speed.

is it the newest version? @CeMarzie