Animius is an open source software library for creating deep-learning-powered virtual assistants. It provides an intuitive workflow that extracts data from existing media (such as anime and TV shows) and trains on them to provide a personalized AI. The flexible architecture enables you to add custom functionality to your virtual assistant.
Animius also ships with a high-level API animius.Console that allows users without programming
experience to use Animius.
Install the current release from PyPi:
pip install animius
Then, install Tensorflow (recommended version 1.14. Animius does not support TF 2.0). We recommend using the GPU package (tensorlfow-gpu)
if you are going to train your own virtual assistant. Read more on Tensorflow installation here.
See Installing Animius for detailed instructions and Docker installation guide.
For more information, check out our quick start guides.
Here's a quick overview of how to create an intent-ner model with Animius:
# create a new model config for intent ner and set default values
myIntentModelConfig = am.ModelConfig(cls='IntentNER')
myIntentModelConfig.apply_defaults()
# we can set & view the values of the config before creating the model
print(myModelIntentConfig.model_structure)
myIntentModelConfig.hyperparameters['learning_rate'] = 0.01
# load a pretrained word embedding from disk
myGloveEmbedding = am.WordEmbedding()
myGloveEmbedding.create_embedding('\\resources\\glove.twitter.27B.50d.txt', vocab_size=40000)
# create a data object for your training set
myIntentData = am.IntentNERData()
myIntentData.add_embedding_class(myGloveEmbedding)
myIntentData.set_intent_folder('resources\\IntentNER Data')
# create a model
myIntentModel = am.IntentNER.IntentNERModel()
# build a TF graph based on the config and data
myIntentModel.build_graph(myModelConfig, myData)
# start tensorflow
myIntentModel.init_tensorflow()
# train!
myIntentModel.train(epochs=10)After training the intent-ner and chatbot models, we can create a waifu - a virtual assistant. The code is a bit long, so I will skip a lot of it. Feel free to check out our demo and our quick start guide for more information.
Anyways, you can simply interact with a working waifu like this:
myWaifu.predict('Hey!')
myWaifu.predict('What is the time in New York right now?')
myWaifu.predict('Play me some music')Also, feel free to check out our other projects surrounding Animius:
- Animius Link - A middleware for extensive assistance functions and socketio support
- Animius Anywhere - A Vue Native mobile GUI client for Animius.
- Animius Live - A WIP Electron desktop client.