We directly modify the following parts of the py_entitymatching, such modified package can be found on GitHub: py_entitymatching.

1. On py_entitymatching, in feature/tokenizer.py. We change the function "**_make_tok_delim**", add the case that "d" is a list.
2. On py_entitymatching, in blocker/rule_based_blocker.py. We change the function "**block_tables**", add the case to remove "PARSE_EXP" in "l_proj_attrs" as well as "r_proj_attrs".
3. On py_entitymatching, in feature/autofeaturegem.py. We change the function "**_get_feat_lkp_tbl**", modify the default generated features for all types of attributes.
4. On py_entitymatching, in feature/autofeaturegen.py. We change the function "**conv_fn_str_to_obj**", add the case to check if "f" in "fn_tup" is None.

Also for py_stringmatching:

1. On py_stringmatching, in tokenizer/qgram_tokenizer. We change the function "**tokenize**" under the class "**QgramTokenizer**" to eliminate the q-gram tokens which are not alphanumeric

At last, it is worth to mention that we installed Megallen packages on serval different Linux meachines and found it may fail due to following reasons:

1. py_stringmatching needs numpy < 0.2.0 due to a new recent release.
2. py_stringsimjoin can only be installed by building source code since the source release on PyPI is missing cythonize.py (not sure if it is fixed now, please try pip install first).