100 Days of ML — Day 9 — What Can Customer Service Teach Us About Overfitting?
Note: Tomorrow I’ll have code!
I moved up from food service to customer service in Pittsburgh, PA in 2011. Unemployment at that level was still high, so employers could pretty much treat their employees however they wanted. When I was let go of that position for incompetence and apathy, I was still pretty shocked. I bounced back.
When I returned to Raleigh, my three weeks of experience in the Steel City landed me a gig in a similar role, with a similar deal. Know your place or get fired.
Things were okay for the first few years. I very quickly picked up on customer needs and anticipated further questions without them asking. Since I was taking 100 calls a day, I didn’t want to be bored, so I improvised a lot, creating voices and characters and bits, especially for repeat callers.
The company was up for sale and that time came to a swift end. We were then instructed to learn about 100 scripts. All the same intros. All the same outros. Just a bunch of scripts for each situation and transitioning into a new situation.
My soul died.
I gave it the best run I could, but these scripts did not perfectly match every scenario and anything said off-script would result in a write-up, the natural outcome being that reps spoke robotically to frustrated clients.
This real life lack of generalization is a good cautionary tale for those in data science and artificial intelligence. We want to be able to predict things with 100% accuracy 100% of the time and, thus, it is so tempting to feel elated when the training set gives us 100% accuracy.
That should scare you.
It doesn’t matter how much data you have, there is too much randomness, too much noise to where nothing will ever be 100% accurate. Simply put, overfitting makes your data as useless as a customer service rep trying to log people in to their accounts.
Wiggle room in your neural networks is what will give it the power to be a good model. Aim for that. You’ll hit success every time.
Jimmy Murray is a Florida based comedian who studied Marketing and Film before finding himself homeless. Resourceful, he taught himself coding, which led to a ton of opportunities in many fields, the most recent of which is coding away his podcast editing. His entrepreneurial skills and love of automation have led to a sheer love of all things related to AI.
#100DaysOfML
#ArtificialIntelligence
#MachineLearning
#DeepLearning
