Fix markdown link typo (#137)

multitask/multitask_tutorial.py

@@ -6,7 +6,7 @@
 # Multi-task learning, or training a single model on multiple tasks, is becoming a standard tool for the modern ML practioner (see Ruder's [survey](http://ruder.io/multi-task/) from 2017 for a nice overview).
 # It often leads to computational gains (one model performing many tasks takes up less memory and storage) as well as performance gains (learning to do well on a related _auxiliary_ task can improve the model's ability on the _primary_ task).
 #
-# While the primary purpose of the Snorkel project is to support training data creation and management, it also comes with a PyTorch-based modeling framework intended to support flexible multi-task learning (e.g. [slice-aware models]https://snorkel.org/use-cases/03-spam-data-slicing-tutorial).
+# While the primary purpose of the Snorkel project is to support training data creation and management, it also comes with a PyTorch-based modeling framework intended to support flexible multi-task learning (e.g. [slice-aware models](https://snorkel.org/use-cases/03-spam-data-slicing-tutorial).
 # Using this particular framework (as opposed to other excellent third party libraries) is entirely optional, but we have found it helpful in our own work and so provide it here.
 # In particular, because MTL in general often requires easily *adding new datasets, tasks, and metrics* (and just as easily removing them), each of these concepts has been decoupled in the snorkel MTL classifier.