thunlp / SE-WRL-SAT

Revised Version of SAT Model in "Improved Word Representation Learning with Sememes"
MIT License
50 stars 8 forks source link

result about evaluation #2

Closed embraceth closed 5 years ago

embraceth commented 5 years ago

您好!我拿您提供的词向量结果和提供的wordsim-240和297脚本进行了测试,为啥结果跟您的不一样,我的240得分是58.6比您展示的要低,但是297是65.3比您展示的要高? 望指教!感谢!

Fanchao-Qi commented 5 years ago

你好,我刚刚测试了一下,发现结果仍然是61和64没变。请问你是用了这个repo的EvalWordSim.py评测的cloud上的word-vec.zh文件吗?

embraceth commented 5 years ago

是的,用的就是您发的这俩文件! 这是盘出来的结果: Word embeddings reading completes and total number of words is: 10144 WordSim-240 Score: 0.5868977927526543 TestPair: 183 SkipPair: 57 WordSim-297 Score: 0.6533767056765682 TestPair: 191 SkipPair: 106

Fanchao-Qi commented 5 years ago

不对呀,词表有40多万词,你读出来才10000多。。

embraceth commented 5 years ago

找到原因了,在那个云盘下文件的时候,老是下到一半自己就停了

embraceth commented 5 years ago

您好!问题解决了,非常感谢!