跑实验记录一

1.使用tagger&wikipedia-pubmed-and-PMC-w2v词向量

Loading pretrained embeddings from ../.local/lib/python3.5/site-packages/neuroner/data/word_vectors/wikipedia-pubmed-and-PMC-w2v.txt...
WARNING: 5443657 invalid lines
Loaded 0 pretrained embeddings.
0 / 18309 (0.0000%) words have been initialized with pretrained embeddings.
0 found directly, 0 after lowercasing, 0 after lowercasing + zero.
Compiling...

词向量无效的问题。

2.使用tagger&PMC-w2v词向量

Loading pretrained embeddings from ./dataset/PMC-w2v.txt...
WARNING: 2515687 invalid lines
Loaded 0 pretrained embeddings.
0 / 18141 (0.0000%) words have been initialized with pretrained embeddings.
0 found directly, 0 after lowercasing, 0 after lowercasing + zero.
Compiling...

 依旧是词向量不能加载的问题。

解决：找到原因了，因为词向量中的维度和默认维度不同，需要指定默认维度啊，--word_dim 200。即可：

Found 10407 unique words (115614 in total)
Loading pretrained embeddings from ./dataset/PMC-w2v.txt...
Found 80 unique characters
Found 9 unique named entity tags
4595 / 4598 / 4840 sentences in train / dev / test.
Saving the mappings to disk...

Loading pretrained embeddings from ./dataset/PMC-w2v.txt...
WARNING: 1 invalid lines
Loaded 2515686 pretrained embeddings.
17963 / 18141 (99.0188%) words have been initialized with pretrained embeddings.
17876 found directly, 46 after lowercasing, 41 after lowercasing + zero.
Compiling...

目前使用的是Att中的CDR数据集进行训练的。

3.使用tagger和chemdner_pubmed_drug.word2vec_model_token4_d50词向量