Reproduce Experiments on CoNLL2003 English NER

[HAWLYQ]

A clear and concise description of what you want to know.
Hi, @alanakbik, I ran the reproduce code showed in https://github.com/zalandoresearch/flair/blob/master/resources/docs/EXPERIMENTS.md 5 times and the mean f1 is 92.76, which is lower than reported 93.07. Could you give me some suggestions? The language models I used are lm-news-english-backward-v0.2rc.pt and lm-news-english-forward-v0.2rc.pt. This is one of my training log files.
conll_flair_reproduce5.log
(The patience you set in your code is ‘’3‘’, which is inconsistent with the ‘’5‘’ written in paper, but I think this is not the main problem)

[alanakbik]

Hi @HAWLYQ thanks for reporting this - since a lot of things have changed since the original Flair release (e.g. bug fixes #184 #390 etc., changes #48 #270 etc.), I am currently re-doing all experiments to compute updated numbers and updated instructions on how to reproduce (old instructions only apply for Flair 0.2). I think everything will be done next week, so we can push everything with the 0.4.1 release!

[stefan-it]

I did also two runs for CoNLL2003 using the latest version of flair. I could only get 0.9264 (patience 5) and 0.9266 (patience 3).

[HAWLYQ]

@alanakbik ， thanks for your kind reply and work. I am looking forward to the 0.4.1 release!

[yahshibu]

Hello, thank you for your contributions to the NLP community!

I was expecting that the 0.4.1 would be released last week. Will it take more time?
I'm thinking of using Flair embeddings for my work... (although it's self-centered...)

[alanakbik]

Flair 0.4.1 is out, we also updated the experiments description!