C5W3A2 Trigger word Detection - Why more positives?

Course Q&A Deep Learning Specialization Sequence Models
1

Hello,

What is the motivation behind inserting 0-4 activate words into training sequences and only 0-2 negative words? It seems that ignoring random loud words is just as important as recognizing the trigger word. And even if one should be given more weight, what is the intuition behind picking a ratio?

My naive approach would be to insert say 1 activate and 4 negative words into each segment, or perhaps make them appear equally often, but certainly not having more activates than others.

Thank you!
Andrey

2

I guess the point of the assignment is to introduce the concept and a training method that yields acceptable performance.

You are welcome to try different data preparation techniques after submitting the assignment.

Happy learning