Deep learning is a pretty hot topic. Unfortunately, the vast majority of engineers and data scientists don’t have practical experience. However, you’re in luck!
I wrote a guide on solving a problem from start to finish using neural networks, the same way I would approach and solve the problem.
Who am I, you ask?
I’m a programmer and self proclaimed data wizard.
More importantly, I have been working with neural networks for years (credentials below)!
Yet Another Neural Network Guide
The goal of this guide is different than many guides out there.
Applying neural networks to problem(s), from start to finish.
From dataset curation, to model comparisons, and even hyperparameter tuning.
With no math and functional code.
I would really appreciate comments, PRs to the github repo, larger labeled datasets, etc.
- Acquiring & formatting data for deep learning applications
- Word embedding and data splitting
- Bag-of-words to classify sentence types (Dictionary)
- Classify sentences via a multilayer perceptron (MLP)
- Classify sentences via a recurrent neural network (LSTM)
- Convolutional neural networks to classify sentences (CNN)
- FastText for sentence classification (FastText)
- Hyperparameter tuning for sentence classification
Good Guides, Start with a Real Problem
I always find it easier to follow guides that have stories. As such, this guide to applied neural networks in production will be covering an actual problem we faced at Metacortex.
Above all, we needed our NLP engine and bots to understand when a question was being asked or a command given. This enabled our system to capture information (from statements) or respond to questions or commands. The goal being, to create an intuitive query interface for an organizations institutional knowledge.
Thus, this guide will cover sentence [type] classification, for examples.
- Developed & deployed neural networks to production to at fortune 100 companies
- Founder: Metacortex
- Developer: hnprofile.com, projectpiglet.com, lettergram.net, easy-a.net
- Issued patent list on the USPTO
- Why You Don’t Necessarily Need Data for Data Science (hint, we used deep learning)
- I host trainings on deep learning regularly, with groups of forty or more
Feel free to reach out any time for questions or suggestions!
I want this guide to help people and plan to continue to improve it with any suggestions I receive.