Skip to main content

​
Prerequisites

Before installing the microservice, ensure you have:
  • Python 3.7+ - The service uses modern Python features and type hints
  • pip - Python package installer (usually bundled with Python)
  • Git (optional) - For cloning the repository
  • 4GB+ RAM - Required for loading PyTorch models and NLTK data
The PyTorch models and dependencies require significant disk space (~2GB). Ensure you have adequate storage available.

​
Installation steps

1

Create virtual environment

Set up an isolated Python environment to avoid dependency conflicts:
python3 -m venv text_prediction
This creates a new directory text_prediction/ containing the virtual environment.Activate the environment:
source text_prediction/bin/activate
You should see (text_prediction) prefix in your terminal prompt.
2

Install dependencies

Install all required Python packages from requirements.txt:
pip3 install -r requirements.txt
This installs 89 packages including:
PackageVersionPurpose
torch1.13.0Neural network framework for emotion classification
Flask2.2.2Web framework for the REST API endpoint
nltk3.7Natural language processing and text preprocessing
scikit-learn1.0.1Traditional ML classifier for sentiment analysis
pycountry22.3.5Country name extraction from text
numpy1.23.5Numerical computing for tensor operations
pandas1.5.2Data manipulation utilities
The installation downloads PyTorch binaries which can take several minutes depending on your connection speed.
3

Download NLTK data

The microservice automatically downloads required NLTK resources on first run:
nltk.download('omw-1.4')      # Open Multilingual Wordnet
nltk.download('wordnet')       # WordNet lexical database
These downloads happen in microservice.py:91-92 when you start the server. No manual action needed.
4

Verify model files

Ensure the pre-trained model files exist in the models/ directory:
models/
├── model_26_87.12.pth           # PyTorch neural network weights
├── word_dict.json                # Word to index mapping for embeddings
├── emotion_classifier.model      # Scikit-learn sentiment classifier
└── vectorizer2.pickle            # TF-IDF vectorizer
These files are loaded during initialization (microservice.py:36-42):
vectorizer = pickle.load(open(f'models/vectorizer2.pickle', 'rb'))
emotion_classifier = pickle.load(open('models/emotion_classifier.model', 'rb'))

with open('models/word_dict.json') as json_file:
    word_dict = json.load(json_file)
labels_wigths = 'models/model_26_87.12.pth'
The microservice will fail to start if any model files are missing. Ensure all four files are present before running.
5

Run the microservice

Start the Flask development server:
python3 microservice.py
Expected output:
[nltk_data] Downloading package omw-1.4 to /home/user/nltk_data...
[nltk_data] Downloading package wordnet to /home/user/nltk_data...
Microserver running in port 3200
 * Serving Flask app 'microservice'
 * Debug mode: off
 * Running on http://127.0.0.1:3200
The server runs on:
  • Host: 127.0.0.1 (localhost only, not accessible from network)
  • Port: 3200
Configuration is defined in microservice.py:29-30:
PORT = 3200
HOST = "127.0.0.1"
6

Test the installation

Verify the service is working by sending a test request:
curl "http://127.0.0.1:3200/textbased_emotion?text=Hello%20world"
You should receive an HTML response with prediction results. If you see an HTML page, the installation is successful.Alternatively, visit the home page in your browser:
http://127.0.0.1:3200/

​
Configuration options

​
Change server port

To run the service on a different port, modify PORT in microservice.py:29: 
PORT = 5000  # Change from default 3200
HOST = "127.0.0.1"

​
Enable network access

To accept requests from other machines, change the host: 
PORT = 3200
HOST = "0.0.0.0"  # Listen on all network interfaces
Exposing the service to the network without authentication may pose security risks. Only use 0.0.0.0 in trusted environments.

​
Adjust toxicity threshold

The default threshold for toxicity classification is 0.29 (microservice.py:206-211). To modify sensitivity: 
# Lower threshold = more sensitive (more content flagged as toxic)
# Higher threshold = less sensitive (fewer false positives)
toxic_result = "Yes" if res[0][0].item() > 0.35 else "No"  # Less sensitive
severe_toxic_result = "Yes" if res[0][1].item() > 0.35 else "No"

​
Troubleshooting

​
Missing model files

Error: FileNotFoundError: [Errno 2] No such file or directory: 'models/word_dict.json' Solution: Ensure all four model files exist in the models/ directory before starting the server.

​
NLTK data download fails

Error: Resource omw-1.4 not found Solution: Manually download NLTK data: 
import nltk
nltk.download('omw-1.4')
nltk.download('wordnet')

​
Port already in use

Error: OSError: [Errno 48] Address already in use Solution: Either stop the process using port 3200 or change the PORT variable in microservice.py.

​
PyTorch CPU warnings

Warning: UserWarning: CUDA not available, using CPU Solution: This is expected behavior. The model loads with map_location=torch.device('cpu') (microservice.py:73) for CPU inference.

​
Next steps

Quickstart guide

Make your first prediction request and understand the response format

Model architecture

Learn how the neural network and classifiers work together

Build docs developers (and LLMs) love

Get started for freeTalk to us