Applying embeddings on categorical data

Chetan06 · June 16, 2020, 8:20am

I am working on a dataset which comprises of both image and tabular data. So I wanted to train them both together in a single model. For images I am using pretrained model from torchvision and for tabular data fastai’s TabularModel() but I was a little confused about the arguments taken in by TabularModel. Can someone explain it by an example

muellerzr · June 16, 2020, 11:18am

See here for an example:

github.com

muellerzr/Practical-Deep-Learning-for-Coders-2.0/blob/master/Tabular Notebooks/01_Adults.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 01 - Adults\n",
    "\n",
    "Now let's do some `fastai2` tabular! As usual, we'll want to install the library:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Run only once per session\n",
    "!pip install fastai2 -q"
   ]

This file has been truncated. show original

While this is in v2, the idea in v1 (And parameters) is still the same. Calculate the size of your embeddings, then you’ll need your number of continuous variables and your number of class output (1 for regression).