classification task error

[ntmc-community]

classification task error

[pl8787]

We have fixed this bug in the newest version. You can download the newest version in the master branch.

[yangliuy]

@Chunlinx you can take a look at this file https://github.com/faneshion/MatchZoo/blob/master/data/example/readme.md

to understand the input data format of MatchZoo toolkit.

[yangliuy]

@Chunlinx Explanations about the parameters in config files can be found here : https://github.com/faneshion/MatchZoo/blob/master/matchzoo/models/readme.txt

[Chunlinx]

Hi, when i tried to run matchzoo on the the msrf dataset , I got an error indices[1,1]=10821 is not in the range(0,3273), I think this is due to the configure file. Could you please explain the parameters below :
"inputs": {
"share": {
"text1_corpus": "../data/example/classification/corpus_preprocessed.txt",
"text2_corpus": "../data/example/classification/corpus_preprocessed.txt",
"use_dpool": true,
"fill_word": 3272,
"embed_size": 100,
"train_embed": true,
"vocab_size": 3273,
"target_mode": "classification",
"class_num": 2,
"text1_maxlen": 20,
"text2_maxlen": 20
},
what are fill_word embed_size, vocab_size mean? and how to determine these figures when applying to a certain dataset
Thank you !

[yangliuy]

fill_word: the max word id, which is vocab_size-1.
embed_size: the dimension of word embeddings.
vocab_size: the size of vocabulary.

A suitable configure is decided by the property of your data. For hyper-parameters in your model, you need to tune them with the validation set.