Getting Started with Autolabel¶
This page will walk you through your very first labeling task using Refuel Autolabel. Specifically, it'll go over:
- Installation
- Overview of a dataset to label
- Labeling the dataset using Autolabel
Installation¶
Autolabel is available on PyPI and can be installed by running:
Separate from the Autolabel library, you'll also need to install an integration with your favorite LLM provider. In the example below, we'll be using OpenAI, so you'll need to install the OpenAI SDK and set your API key as an environment variable:
To use a different LLM provider, follow the documentation here.
Goal: Sentiment Analysis on a Movie Review Dataset¶
Let's say we wanted to run sentiment analysis on a dataset of movie reviews. We want to train our own ML model, but first, we need to label some data for training.
Now, we could label a few hundred examples by hand which would take us a few hours. Instead, let's use Autolabel to get a clean, labeled dataset in a few minutes.
A dataset1 containing 200 unlabeled movie reviews is available here, and a couple of examples (with labels) are shown below:
| text | label |
|---|---|
| I was very excited about seeing this film, anticipating a visual excursus on the relation of artistic beauty and nature, containing the kinds of wisdom the likes of "Rivers and Tides." However, that's not what I received. Instead, I get a fairly uninspired film about how human industry is bad for nature. Which is clearly a quite unorthodox claim. The photographer seems conflicted about the aesthetic qualities of his images and the supposed "ethical" duty he has to the workers occasionally peopling the images, along the periphery. And frankly, the images were not generally that impressive. And according to this "artist," scale is the basis for what makes something beautiful. In all respects, a stupid film. For people who'd like to feel better about their environmental consciousness ... but not for any one who would like to think about the complexities of the issues surrounding it. |
negative |
| I loved this movie. I knew it would be chocked full of camp and silliness like the original series. I found it very heart warming to see Adam West, Burt Ward, Frank Gorshin, and Julie Newmar all back together once again. Anyone who loved the Batman series from the 60's should have enjoyed Return to the Batcave. You could tell the actors had a lot of fun making this film, especially Adam West. And I'll bet he would have gladly jumped back into his Batman costume had the script required him to do so. I told a number of friends about this movie who chose not to view it... now they wished they had. I have all of the original 120 episodes on VHS. Now this movie will join my collection. Thank You for the reunion Adam and Burt. | positive |
Our goal is to label the full 200 examples using Autolabel.
Labeling with AutoLabel¶
Autolabel provides a simple 3-step process for labeling data:
- Specify the configuration of your labeling task as a JSON
- Preview the labeling task against your dataset
- Label your data!
Specify the labeling task via configuration¶
First, create a JSON file that specifies:
- Task:
task_nameisMovieSentimentReviewand thetask_typeisclassification - LLM: Choice of LLM provider and model - here we are using
gpt-3.5-turbofrom OpenAI - Instructions: These are the labeling guidelines provided to the LLM for labeling
config = {
"task_name": "MovieSentimentReview",
"task_type": "classification",
"dataset": {
"label_column": "label"
},
"model": {
"provider": "openai",
"name": "gpt-3.5-turbo"
},
"prompt": {
"task_guidelines": "You are an expert at analyzing the sentiment of movie reviews. Your job is to classify the provided movie review into one of the following labels: {labels}",
"labels": [
"positive",
"negative",
"neutral"
],
"few_shot_examples": [
{
"example": "I got a fairly uninspired stupid film about how human industry is bad for nature.",
"label": "negative"
},
{
"example": "I loved this movie. I found it very heart warming to see Adam West, Burt Ward, Frank Gorshin, and Julie Newmar together again.",
"label": "positive"
},
{
"example": "This movie will be played next week at the Chinese theater.",
"label": "neutral"
}
],
"example_template": "Example: {example}\nLabel: {label}"
}
}
To create a custom configuration, you can use the CLI or write your own.
Preview the labeling against your dataset¶
First import autolabel, create a LabelingAgent object and then run the plan command against the dataset (available here) and can be downloaded through the autolabel.get_data function):
from autolabel import LabelingAgent, AutolabelDataset, get_data
get_data('movie_reviews')
agent = LabelingAgent(config)
ds = AutolabelDataset('data/movie_reviews/test.csv', config = config)
agent.plan(ds)
This produces:
Computing embeddings... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100/100 0:00:00 0:00:00
┌──────────────────────────┬─────────┐
│ Total Estimated Cost │ $0.538 │
│ Number of Examples │ 200 │
│ Average cost per example │ 0.00269 │
└──────────────────────────┴─────────┘
───────────────────────────────────────────── Prompt Example ─────────────────────────────────────────────
You are an expert at analyzing the sentiment of moview reviews. Your job is to classify the provided movie review as positive or negative.
You will return the answer with just one element: "the correct label"
Now I want you to label the following example:
Input: I was very excited about seeing this film, anticipating a visual excursus on the relation of artistic beauty and nature, containing the kinds of wisdom the likes of "Rivers and Tides." However, that's not what I received. Instead, I get a fairly uninspired film about how human industry is bad for nature. Which is clearly a quite unorthodox claim.<br /><br />The photographer seems conflicted about the aesthetic qualities of his images and the supposed "ethical" duty he has to the workers occasionally peopling the images, along the periphery. And frankly, the images were not generally that impressive. And according to this "artist," scale is the basis for what makes something beautiful.<br /><br />In all respects, a stupid film. For people who'd like to feel better about their environmental consciousness ... but not for any one who would like to think about the complexities of the issues surrounding it.
Output:
──────────────────────────────────────────────────────────────────────────────────────────────────────────
This shows you:
- Number of examples to be labeled in the dataset:
200 - Estimated cost of running this labeling task:
<$1 - Exact prompt being sent to the LLM
Having previewed the labeling, we are ready to start labeling.
Label your dataset¶
Now, you can use the run command to label:
This takes just a few minutes to run, and returns the labeled data as an Autolabel Dataset. We can explore this by running:
ds.df.head()
>
text ... MovieSentimentReview_llm_label
0 I was very excited about seeing this film, ant... ... negative
1 Serum is about a crazy doctor that finds a ser... ... negative
2 This movie was so very badly written. The char... ... negative
3 Hmmmm, want a little romance with your mystery... ... negative
4 I loved this movie. I knew it would be chocked... ... positive
[5 rows x 4 columns]
At this point, we have a labeled dataset ready, and we can begin training our ML models.
Using Hugging Face datasets¶
If you want to use a Hugging Face dataset directly, you can pass it into agent.plan and agent.run as you would a file path or pandas.DataFrame.
dataset = load_dataset(DATASET_NAME)
agent = LabelingAgent(config)
agent.plan(test_dataset)
agent.run(test_dataset)
Summary¶
In this simple walkthrough, we have installed autolabel, gone over an example dataset to label (sentiment analysis for moview reviews) and used autolabel to label this dataset in just a few minutes.
We hope that this gives you a glimpse of what you can do with Refuel. There are many other labeling tasks available within Autolabel, and if you have any questions, join our community here or open an issue on Github.
-
Andrew L. Maas, Raymond E. Daly, Peter T. Pham, Dan Huang, Andrew Y. Ng, and Christopher Potts. (2011). Learning Word Vectors for Sentiment Analysis. The 49th Annual Meeting of the Association for Computational Linguistics (ACL 2011). ↩