Tupa

* Support for CoNLL-U and SDP formats.
* Identify format per passage without requiring the `-f` option.
* Fixes to AMR parsing; add `Label` as a separate transition to label nodes (rather than doing it immediately upon creation).
* Support for multi-task training, with a shared BiLSTM input encoding but separate MLP (and labels) per format.
* Save model hyper-parameters to a .json file rather than Pickle. Models trained with TUPA v1.1 are no longer compatible.
* Save perceptron weights to a separate Pickle file with a .data suffix.

This release introduces a number of changes from the version used in the ACL 2017 paper, [v1.0](https://github.com/huji-nlp/tupa/releases/tag/v1.0).

* Upgrade to [DyNet v2.0](https://github.com/clab/dynet/releases/tag/v2.0): this entails a different format for model files, so models trained with TUPA v1.0 are no longer compatible.
* Easier installation by a simple `pip install tupa` or `python setup.py install`.
* Preliminary support for AMR parsing, by specifying the `-f amr` option.
* Functioning demo server (available [online](http://bit.ly/tupademo)).
* Various bug fixes.

This is the version used for the experiments in [the following paper](http://aclweb.org/anthology/P17-1104):

InProceedings{hershcovich2017a,
author = {Hershcovich, Daniel and Abend, Omri and Rappoport, Ari},
title = {A Transition-Based Directed Acyclic Graph Parser for {UCCA}},
booktitle = {Proc. of ACL},
year = {2017},
pages = {1127--1138},
url = {http://aclweb.org/anthology/P17-1104}
}

(Models repaired with `sed -i '1s/::archive 14/::archive 12/' models/*.model`, see https://stackoverflow.com/a/47194301/223267)