Mike Kroutikov

Version 1.3.0 crashes with IndexOutOfBounsException. To reproduce: 1. Download PDF file: `wget https://www.sec.gov/files/formcustody.pdf` 2. Run tabula: ``` java -Dfile.encoding=UTF8 -jar tabula-1.0.3-jar-with-dependencies.jar \ --pages 7 --area 70.847,72.698,178.03,564.261 \ --stream --format JSON...

@mkudinov please review. This is somewhat theoretical fix. 1. removed initial cleanup of padding in char embedding matrix from train.py. This should not affect training as each training step will...

makes model a tiny bit smaller with no change in the expressive power

Reproduced BPC of 1.077 using model with one attention layer

3

``` | epoch 13 | 11210/10986 batches | lr 0.00100 | ms/batch 188.01 | loss 0.71 | ppl 2.04 | bpc 1.029 | epoch 13 | 11220/10986 batches | lr...

Error when serializing model

2

When running example from README.md I get this message: ``` WARNING:tensorflow:Error encountered when serializing layer_variables/seq2seq_model. Type is unsupported, or the types of the items don't match field type in CollectionDef....

Problem I am solving: skip Pipeline step if file already exists in Google Storage bucket. The best solution I see so far is to have Google Storage Plugin define an...