Labeling test list of ItemsLists with label_from_df

takotab · January 29, 2019, 6:56am

Hi all,

I’m having trouble using the Itemslists. When I want to add labels to the test list I’m getting errors. I replicated my problem on the planet dataset:

gist.github.com

https://gist.github.com/takotab/6351f85827a1e77bb380d020c545ca0f

fastai_trouble.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "from fastai.vision import *"

This file has been truncated. show original

The specific error I get:

TypeError: process() got an unexpected keyword argument ‘name’

Ubuntu 18.04
python 3.7.2
fastai 1.0.42

What do I do wrong? Is there another way to surpass these strange errors.

Just replicating the notebook would already be helpfull.

sgugger · January 29, 2019, 2:22pm

You’re not following the steps of the data block API in the right order, that’s why you get this error. You are supposed to split before you label (which is why you shouldn’t define an ImageItemList for your validation set), and then after labelling you can add the test set with add_test.

takotab · January 29, 2019, 3:23pm

Thanks for your reaction. Stupid. I should have followed the steps that are clearly marked on the Data Block API page.

lazy link: https://docs.fast.ai/data_block.html