Deepchem

1462 Typo fix predict_proba
1385, 1418 Build Fixes
1423 Yapf updates
1408 indentation cleanup
1344 Python 3.6 updates
1330: Docs updates
1337 Large screens tutorial
1338 Colab notebook version
1437 Python3 fixes
1454 Make RDKit a soft requirement
1455 Make simdna a soft requirement
1456, 1484 Make six a soft requirement
1458 Add tutorial section
1420 genomics code grouping into single file
1535, 1485, 1487, 1371, 1421, 1479, 1480 Test Improvements and Fixes

This minor release swaps the backend metadata storage for `DiskDataset` to hd5 from pickle files. This change was necessary since pickle is a brittle file format. (A recurring issue was that py2 pickles and py3 pickles were not compatible). This minor release introduces a transition implementation: loading any `DiskDataset` on disk with the `dc.data.DiskDataset` implementation in this release will automatically swap from pickle to hd5 format on disk with no extra input needed from users.

The next DeepChem release is going to be a 2.0 release which will consolidate on the hd5 format, so we encourage users to upgrade to 1.3.1 and transition any data stored on disk in preparation.

Thanks as always to our core developers and special thanks to our first-time contributors!

Detailed Changes:
- Pickle to hd5 transition for `DiskDataset` (899)
- TensorGraph Improvements (895, 901, 917)
- Dataset fixes (914, 916)
- MoleculeNet Jenkins build (910)
- Major refactoring/improvements to RDKIT grid featurizer (879)
- Installation/Testing/Docs improvements (892, 904, 907, 909, 920, 923)

This major version release consolidates around `TensorGraph` as DeepChem's high-level deep learning API of choice. Lots of improvements and bugfixes have been made to the core `TensorGraph` library, and many new layers and models have been added. In particular, DeepChem now features GANs, Seq2Seq models, Model Agnostic Meta Learning and more! Many improvements to tutorials, examples, website, and installation have been added as well.

Our thanks to all the developers who contributed to this release, with a particular shout-out to those who made their first PRs to DeepChem!

Detailed Changes:
- TensorGraph Improvements (693, 705, 723, 730, 731, 746, 751, 753, 758, 760, 763, 766, 774, 778, 782, 783, 788, 791, 794, 799, 811, 817, 822, 824, 826, 833, 847, 850, 853, 854, 860, 871)
- RL Improvements: Hindsight Experience Replay, Proximal Policy Optimization, API cleanup (686, 688, 697, 713, 719, 729, 740)
- IPython Notebook Improvements (703, 706, 709, 711, 717, 721, 727, 745, 750, 829)
- Cleaning up examples (755, 816, 819, 830, 840)
- MoleculeNet Improvements (696, 718, 772, 854, 880, 738)
- Rehaul Website (800, 801, 806, 807)
- Improvements and extensions to `dc.splits` (690, 765, 784)
- Improvements to rdkit-grid-featurizer (868, 873, 883)
- Miscellaneous cleanups/fixes (701, 712, 724, 732, 735, 739, 796, 848, 885,
- DeepChem installation/import improvements (737, 793, 802, 803, 804, 813, 814, 815, 852, 857, 859, 885, 888)

Detailed listing of new models added:

- GANs (855, 866)
- Model Agnostic Meta Learning (759)
- Seq2seq models (828)
- ANI-1 Models (823, 839)
- Spatial Filtering Graph Convs (851)
- Message Passing Neural Networks (710)
- TextCNNs (874)
- Sluice Networks (805)

The major new change in this release is a new reinforcement learning framework. There have also been many upgrades and bugfixes to TensorGraph, large upgrades to MoleculeNet, and significant effort spent cleaning up and solidifying our test suite, documentation, and community standards.

Detailed Changes:
- MoleculeNet Upgrades (556, 578, 589, 592, 594, 606, 628, 629, 633, 661, 665, 667 )
- Reinforcement Learning Support (557, 618, 640, 652, 656)
- Cleaning up tests and making robust (560, 561, 563, 566, 582, 584, 595, 611, 662)
- GPU Docker support (574)
- Documentation Improvements (575, 597, 609, 615, 641, 670, 674)
- TensorGraph Refinements and Debugging (552, 567, 569, 577, 603, 608, 636, 637, 638, 642, 655, 666)
- Added code of conduct (580)
- PyTorch Model Upgrades (646, 652)

This minor release version adds `TensorGraph`, a new backend for DeepChem models inspired by Keras's functional API. `TensorGraph` should now be ready for early users to start experimenting with. Over the next few releases, we will deprecate non-`TensorGraph` models in favor of the newer implementations. This release also contains a number of major improvements to MoleculeNet, with new models, datasets, and metrics.

Detailed Changes:

- Added a `contrib/` folder to allow users to contribute models/code more easily.
- PyTorch multitask models merged into `contrib/` (481)
- XGBoost models added (483)
- A number of new datasets integrated with `dc.molnet` (484)
- Basic queue support added . Allows for higher GPU utilization to be achieved (488)
- Introduction of `TensorGraph`, a new backend for DeepChem inspired by Keras's function API (493, 505, 517, 520, 523, 544, 546 )
- Adding support for weight-sharing between layers in `TensorGraph` (521)
- Porting of multitask classifiers/regressors into `TensorGraph` and adding Dropout support (522)
- Implementation of molecular DAG models (495)
- Implementation of weave convolutions (496)
- Adding atomic convolutions (509, 537 )
- MPNNs added to `contrib/` (512)
- Made `dc.splits` work with `NumpyDataset` (513)
- Adding auPRC metric (531)

Made home-page point to git.