how is the data organized？

[QQToyota]

hello，I read your code and don‘t find the description of data format，Can you add it？

[yuboxie]

Hello,

I just uploaded the training data (tokenized). Please check the README file.

Briefly speaking, the format is:

• enc_input.npy: tokenized dialog context, with shape H * N * L
• enc_input_len.npy: utterance lengths of dialog context, with shape H * N
• enc_input_e.npy: emotion labels of the dialog context, with shape N * H * E
• dec_input.npy: tokenized decoder input, with shape N * (L+1)
• dec_input_len.npy: lengths of decoder input, with shape N
• target.npy: tokenized target utterance, with shape N * (L+1)
• hist_len.npy: lengths of dialog context, with shape N

where

• N is the number of data points
• H is the maximum dialog context length
• L is the maximum utterance length
• E is the number of emotion categories