I've continued working on the correlation/prediction algorithm. I found a relatively major bug that was messing up certain calculations for making predictions. The "bugged" version was mostly just not doing what I intended theoretically but still does perform at the level I discussed in my Week 24 issue. The "fixed" version is not yet performing at the same level as the "bugged" version, so there is a possibility I will revert back to the "bugged" version. I do believe the "fixed" version, however, has potential to substantially outperform the "bugged" version. At the moment, based on my simulation, the algorithm can predict 44% of stressful events, after viewing 100 events, given a 0.3 probability of randomness. That is to say a stressful factor only has a 70% chance of actually leading to a stressful event, and events without any stressful factors still have a 30% chance of being stressful. The randomness element is there to account for the fact that not all information about a person's current environment can be know by the application. Naturally the less randomness, the better the algorithm performs. 44% prediction is much higher than the previous 25%, but only 70% of the predicted stressful events have stress factors - as opposed to the 80% in the "bugged" version. To fix this, there are a number of things I'm working on accounting for, but I haven't yet figured out the best way to approach those issues. Notable files again include Node.java, CorrelationGraph.java, and EventGraph.java
@poorvi-vora
I've continued working on the correlation/prediction algorithm. I found a relatively major bug that was messing up certain calculations for making predictions. The "bugged" version was mostly just not doing what I intended theoretically but still does perform at the level I discussed in my Week 24 issue. The "fixed" version is not yet performing at the same level as the "bugged" version, so there is a possibility I will revert back to the "bugged" version. I do believe the "fixed" version, however, has potential to substantially outperform the "bugged" version. At the moment, based on my simulation, the algorithm can predict 44% of stressful events, after viewing 100 events, given a 0.3 probability of randomness. That is to say a stressful factor only has a 70% chance of actually leading to a stressful event, and events without any stressful factors still have a 30% chance of being stressful. The randomness element is there to account for the fact that not all information about a person's current environment can be know by the application. Naturally the less randomness, the better the algorithm performs. 44% prediction is much higher than the previous 25%, but only 70% of the predicted stressful events have stress factors - as opposed to the 80% in the "bugged" version. To fix this, there are a number of things I'm working on accounting for, but I haven't yet figured out the best way to approach those issues. Notable files again include Node.java, CorrelationGraph.java, and EventGraph.java