steamship-plugins/prompt-generation-default

Steamship Tagger Plugin

This project implements a basic Steamship Tagger that you can customize and deploy for your own use.

In Steamship, Taggers add annotations to text that can be queried and composed later. Note that a file must have first been blockified in order to be tagged.

This sample project adds paragraph and sentence tags for a sample text document, but the implementation you create might:

• Use entity recognition to tag named entities in text block
• Use sentiment analysis to tag positive and negative sections of a transcribed conversation

Once a Tagger has generated data in Steamship, that data is ready for use by the rest of the ecosystem. For example, you could perform a query over the sentences or embed each sentence.

First Time Setup

We recommend using Python virtual environments for development. To set one up, run the following command from this directory:

python3 -m venv .venv

Activate your virtual environment by running:

source .venv/bin/activate

Your first time, install the required dependencies with:

python -m pip install -r requirements.dev.txt
python -m pip install -r requirements.txt

Developing

All the code for this plugin is located in the src/api.py file:

• The TaggerPlugin class
• The run method that is invoked via the File.tag call

Testing

Tests are located in the test/test_api.py file. You can run them with:

pytest

We have provided sample data in the test_data/ folder.

Deploying

Deploy your tagger to Steamship by running:

ship deploy --register-plugin

That will deploy your plugin to Steamship and register it as a plugin for use.

Using

Once deployed, your Tagger Plugin can be referenced by the handle in your steamship.json file.

from steamship import Steamship, Block, File, MimeTypes, Tag

MY_PLUGIN_HANDLE = ".. fill this out .."

client = Steamship()
tagger = client.use_plugin(plugin_handle=MY_PLUGIN_HANDLE, plugin_instance="unique-instance-id")

with open("./test_data/king_speech.txt", "r") as text:
    # here, we add an initial block, as tagging requires files have been blockified.
    content = text.read()
    file = File.create(client, content=content, mime_type=MimeTypes.TXT, blocks=[Block.CreateRequest(text=content)])
    
file.tag(tagger.handle).wait()

# now that our file has been tagged, we can access the new tags by refreshing the file.
file = file.refresh()
for block in file.blocks:
    print(block.text)
    print(block.tags)

# we can also query for tags in the file (here, finding sentences)
print("\n".join([content[t.start_idx:t.end_idx] for t in Tag.query(client, f'file_id "{file.id}" and kind "sentence"').tags]))

Sharing

Please share what you've built with [email protected]!

We would love take a look, hear your suggestions, help where we can, and share what you've made with the community.