NLIDB Progress

Looks like I can beat state of the art or at least come close to it. This is probably going to be one of the best unsupervised NLIDB. Take it in your ass my family, barclays, corrupt and useless politicians,  and UK.

NLIDB problem is mainly of sequence modelling and word embeddings. Programming can be stressful if you are not focused.

Wrote the first version for identifying x,y,z.First draft of syntaxnet done. Moving on to sling. Will move on to other rules and maybe some supervised learning afterwards.

Need to refine it a lot. Too much work needed. It can work for simple queries. Need to refine it for complex queries. A little bit of help from a linguist will be useful.

I dare anyone to build a pure deep learning NLIDB. Quora haters can suck my dick after trying. I will allow them to swallow the sweet nectar of my dick. I know that the handcrafted features can be learnt by the deep learning model. Still I dont think end to end systems can be built; especially open domain systems. Sling is a case in point. 

Figured out the scoring module. Also built a phrase extractor.

They have resorted to hacking to demoralize me. Useless fucks.

Can a linguist consult with me for an hour? Just to make the queries better. 

Working on implicit entities using graph traversal. It is too much fucking work.

Too much hacking. Will work when hacking stops. Seems like hacking has stopped. Will have to work even if it is a Sunday. Who knows when hacking will resume. I am aiming for more than 85 percent precision. I dont give a fuck about recall. Should at least come close to state of the art.

This is definitely going to be the worlds best unsupervised NLIDB. I checked a few deep learning NLIDBs. Dont look promising. 70 percent precision on easy queries that too on column level and not on query level. Fuck google and quora.

my approach has been justified by the results of this paper. https://www.cs.utexas.edu/~nyaghma/sqlizer.pdf

Have started consolidation. It needs a lot of additional queries and some refinements. I am definitely going to beat the state of the art.

Did some refining. It is shaping up nicely. Will combine graph traversal with ER graph to figure out links and implicit entities. The query can also be rewritten to fit the slots. Graph traversal can be used to convert it into a semamtic parser.

I am going to practice it on atis dataset. Motherfuckers have resorted to hacking to demoralize me and slow down my work.

With ER graph it should give good results. I will definetely beat google.