Bill.com Chanllenge

Inspiration

What it does

With the training data given, this model can predict if any two new node has connections with the accuracy of 85%

How we built it

Since each nodes has clear type and features and large amount of data is given, we train a hidden Markov chain to learn the among types and features. The models can predict the possibility of two nodes to have connections by the relation between their features and types, and generate a edge based on that possibility.

Challenges we ran into

1. We are facing over five hundred thousand pieces of data and each node can have different number and kind of features and different kinds, so it is hard to find a clear relation among these variables. To find the relation among all these data, we visualize the connections among nodes with different parameters(degree, number of features, types) and found that the kind of type is one of the fundamental
2. There are many signs that two nodes have relation such as they are of the same type or their features has some relation(the relation is found by machine learning), so it is hard to give a clear and solid conclusion from these signs especially when the conclusions of signs are confliciting with each others, such as they are of the same type but their features does not match with each other. Therefore, we came up with different enhance functions to magnify the effect of a certain sign. We gave each different parameter different weight and assign different enhance function to interpret these signs. ## Accomplishments that we're proud of

What we learned

What's next for Bill.com Chanllenge

Built With

• python