Overview
This NLP project provides functions to analyze emotions in video or audio files. It offers a comprehensive set of tools to detect and analyze emotions at a sentence level, producing valuable insights into the emotional content of multimedia sources. It includes a package with all necessary functions, two pipelines – one for training the NLP model and one for inference, and Sphinx documentation.
Installation
To install the package and its dependencies, use the following pip command:
pip install emotion_detective
Dependencies
The emotion_detective package requires the following dependencies: - python ^3.9 - pandas ^2.0.0 - nltk: - Version ^3.6.3 for win32 platforms - Version 3.8.1 for linux platforms - regex 2024.5.10 - moviepy 1.0.3 - pydub 0.25.1 - imageio 2.34.1 - imageio-ffmpeg 0.4.9 - scikit-learn ^1.0 - transformers ^4.30.0 - gensim ^4.3.2 - torch ^2.1.2 - numpy 1.23.2 - torchvision ^0.16.2 - setuptools-rust 1.9.0 - scipy 1.12 - pyspellchecker ^0.8.1 - accelerate >=0.20.1 - fastapi >=0.111.0 - uvicorn ^0.22.0 - typer >=0.12.3 - tokenizers: - Version 0.19 for win32 platforms - Version 0.11.1 for linux platforms - safetensors 0.4.2 - wordninja ^2.0.0 - textblob ^0.18.0.post0 - azureml-core ^1.56.0 - azureml-fsspec ^1.3.1 - sphinx-rtd-theme ^2.0.0 - h5py ^3.11.0 - matplotlib ^3.9.0 - mlflow ^2.13.2 - colorama ^0.4.6 - datasets ^2.20.0 - azureml-mlflow ^1.56.0 - azure-ai-ml ^1.17.0
Additional Dependencies
The package also requires to have the following dependencies installed on your system. To install the additional dependencies, refer to the installation documentation linked below:
Using the API
The Emotion Detective package also offers an API for the training and inference pipeline.
To use the API, create a simple POST request containing the arguments for the specific pipeline.
Here is an example using the python library ‘requests’:
import requests
import json
url = ""194.171.191.226:3950/train"
payload = json.dumps({
"train_data_path": "./data/CSV/5000_sampled_emotions_neutral.csv",
"test_data_path": "./data/CSV/val_set_5000.csv",
"text_column": "text",
"emotion_column": "label",
"num_epochs": 1,
"model_type": "rnn",
"model_dir": "./models",
"model_name": "rnn_test_api",
"learning_rate": 0.001,
"train_batch_size": 4,
"eval_batch_size": 8,
"cloud_logging": False
})
headers = {
'Content-Type': 'application/json'
}
response = requests.request("POST", url, headers=headers, data=payload)
print(response.text)
import requests
import json
url = "194.171.191.226:3950/inference"
payload = json.dumps({
"input_media_path": "./data/EmotionsforKids.mp3",
"model_path": "./models/rnn.pth",
"model_type": "rnn",
"emotion_mapping_path": "./models/mapping_emotions_rnn.json"
})
headers = {
'Content-Type': 'application/json'
}
response = requests.request("POST", url, headers=headers, data=payload)
print(response.text)