Obtaining large annotated datasets is critical for training successful\nmachine learning models and it is often a bottleneck in practice. Weak\nsupervision offers a promising alternative for producing labeled datasets\nwithout ground truth annotations by generating probabilistic labels using\nmultiple noisy heuristics. This process can scale to large datasets and has\ndemonstrated state of the art performance in diverse domains such as healthcare\nand e-commerce. One practical issue with learning from user-generated\nheuristics is that their creation requires creativity, foresight, and domain\nexpertise from those who hand-craft them, a process which can be tedious and\nsubjective. We develop the first framework for interactive weak supervision in\nwhich a method proposes heuristics and learns from user feedback given on each\nproposed heuristic. Our experiments demonstrate that only a small number of\nfeedback iterations are needed to train models that achieve highly competitive\ntest set performance without access to ground truth training labels. We conduct\nuser studies, which show that users are able to effectively provide feedback on\nheuristics and that test set results track the performance of simulated\noracles.\n