Full benchmark Updates

[lilleswing]

qm7b, toxcast, and kaggle are currently broken and I cannot run them through end to end. I will file issues for each one with the stack trace.

We also have multiple results not hitting desired expectations. Below is the full list.

Dataset	Model	Split	Metric	Retrieved Value	Desired Value
muv	Multitask network	Index splitting	test_score	0.749443916693	0.764
muv	logistic regression	Index splitting	test_score	0.756238252709	0.766
muv	graph convolution	Index splitting	test_score	0.791876998348	0.823
pcba	Multitask network	Index splitting	train_score	0.814176263345	0.826
pcba	Multitask network	Index splitting	test_score	0.788107828923	0.802
clintox	Multitask network	Index splitting	test_score	0.819248826291	0.83
clintox	robust MT-NN	Index splitting	test_score	0.81514084507	0.827
nci	graphconv regression	Index splitting	train_score	0.112239244388	0.123
nci	graphconv regression	Index splitting	test_score	0.0464834859959	0.048
qm7b	NN regression	Index splitting	train_score	0.876963119457	0.931
qm7b	NN regression	Index splitting	test_score	0.790498748877	0.803
qm7b	NN regression	Index splitting	train_score	0.878199577192	0.931
qm7b	NN regression	Index splitting	test_score	0.791941826351	0.803
qm7b	NN regression	Index splitting	train_score	0.879837090375	0.931
qm7b	NN regression	Index splitting	test_score	0.792008993439	0.803
qm7	NN regression	Index splitting	test_score	0.97433488412	0.986
sampl	NN regression	Index splitting	test_score	0.755635996	0.764
sampl	NN regression	Random splitting	test_score	0.761702572858	0.83
sampl	NN regression	Scaffold splitting	train_score	0.880153572737	0.891
nci	graphconv regression	Index splitting	train_score	0.117258262877	0.123
nci	graphconv regression	Index splitting	train_score	0.115851838028	0.123
nci	graphconv regression	Index splitting	test_score	0.0463009131706	0.048
qm7	NN regression	Index splitting	test_score	0.958073353377	0.986
nci	graphconv regression	Random splitting	train_score	0.106167884328	0.117
nci	graphconv regression	Random splitting	test_score	0.0706557598031	0.076

[coveralls]

Coverage remained the same at 85.447% when pulling ce95dbc on lilleswing:full-benchmark-cr into 10b6cd8 on deepchem:master.

[coveralls]

Coverage decreased (-0.1%) to 85.326% when pulling 255d43a on lilleswing:full-benchmark-cr into 10b6cd8 on deepchem:master.

[rbharath]

@lilleswing Thanks for the PR! We will need to find a systematic way of ensuring that the benchmarks here and MoleculeNet numbers match to within 1%. Could you open an issue with the table of outstanding failures? I'm going to go ahead and merge this PR in.