My favorite NN debugging test
If you are implementing some fancy architecture, I advice you to implement a following test: Take batch of examples and run it through the network Take one (let...
Author: Dawid Smoleń
3 Kinds Of Indicators For Improving Your ML System
You should never guess how to improve your classifier. I’ve seen spectacular improvements and accuracy jumps in Neural Networks that were achieved by acci...
Pairing lovers story
Your partner (for the fluent writings let’s say it’s a woman), the psychology student, got the grant for the research about the relationships. She created a pre...
How to recognize digits: Yet another tutorial nobody needs
I don’t know if I’m any better than the people I’m gonna criticize today. Nevertheless, I can’t believe in the scale with which wast...
Sanity checks are your 13th month of work
For a moment I was strongly influenced by Test Driven Development (TDD). It took me a while to relate it to Machine Learning in general. I hope to write a p...
Casting the biggest enemy: Overfitting vs Silent failures
For many years I had a strong conviction that the worst thing I could do is to overfit the training data. I loved to watch these legendary disasters fro...
The holy line of your ML project
Many clients, many servers, dataset versions, database migrations, artificial data, deprecated data, preparing for new functionality. A casual day. Leak...
Clothing recommendation story
A chain of shoe stores asked you to create a clothing recommendation system based on their own database. They have the pictures of shoes of people buying the sh...
Neural Networks common mistakes story
Training Recurrent Neural Networks is a hard task, especially for an inexperienced engineer, that can be surprised by some phenomena specific to that type o...
Adding buildings to Google Maps story
Your goal is to create a classifier that transcripts numbers of houses from pictures. You have 30,000,000 annotated images, and billions of images to tag by...