Debugging Bad Results

The process of building a good model may involve some iterations of trial and error. If you are not getting good results in a model it’s good to check the following things to improve your results.

• Collect more data - If you are working with a small dataset, it’s likely you do not have enough data to find a good match. The Data Studio makes it easy to quickly iterate and improve your model by using Knowledge Pack results to re-train your dataset with new labels. See more about this feature in the Data Studio Documentation

• Check your sensor placement - Is your sensor oriented consistently throughout your training data? A bad sensor placement can cause very unpredictable results

• Evaluate your labels are consistent and accurate - Are your labels consistent throughout your project? Check spacing and start/end placement to verify you are labeling your events in a consistent way

• Evaluate different algorithms, seeds, segmenters - You may need to try out different parameters in the model building step above or move to the SensiML Python SDK to get better results if your events are more complex