lancetnic 2.0.2

A tool for working with text data

LANCETNIC 2.0.2

LANCETNIC is a library with built-in neural network models for working with text and numeric data. Lancetnic provides convenient tools for:

• Data preparation and vectorization
• Training classification models
• Visualization of metrics
• Forecasting on new data

The library allows you to work with purely textual data, as well as with a combination of textual and numerical features, trend and price analysis. Usage examples: text classification, identification of spam, fraudulent messages, working with numerical series and time signs.

🚀 Installing:

Install with CUDA

To work with the GPU, it is recommended to install PyTorch with CUDA support (OPTIONAL):

pip install torch==2.5.1+cu124 torchaudio==2.5.1+cu124 torchvision==0.20.1+cu124 --index-url https://download.pytorch.org/whl/cu124

Then install lancetnic:

pip install lancetnic

👥 Autors

• Lancet52

📄 Documentation

Документация на русском

Documentation in English

Quick start

Text classification example

from lancetnic.models import LancetMC
from lancetnic import TextClass

text_model = TextClass(
            text_column='description',  # Column name containing text data
            label_column='category',    # Column name containing labels
            split_ratio=0.2,            # Train/validation split ratio (if no val_path)
            random_state=42             # Random seed for reproducibility
            )

text_model.train(model_name=LancetMC,   # Model architecture for text classification
                train_path="train.csv", # Path to training data (CSV format)
                val_path="val.csv",     # Path to validation data (None for auto-split)
                num_epochs=50,          # Total training epochs
                hidden_size=256,        # Size of hidden layers
                num_layers=1,           # Number of hidden layers
                batch_size=256,         # Batch size for training
                learning_rate=0.001,    # Learning rate for optimizer
                dropout=0,              # Dropout rate (0-1)
                optim_name='Adam',      # Optimizer ('Adam', 'SGD', 'RAdam', etc.)
                crit_name='CELoss'      # Loss function ('CELoss' or 'BCELoss')
                )
           

Making predictions

from lancetnic import TextClass

text_model = TextClass()
text_pred = text_model.predict(
                model_path="model.pth", # Path to saved model
                text="Sample text to classify" # Text input for prediction
                )

Combined text and numeric features example

from lancetnic.models import LancetMC
from lancetnic import TextScalarClass

mixed_model = TextScalarClass(
                text_column='description',  # Text column name (None if only numeric)
                data_column=['feat1', 'feat2'], # List of numeric feature columns
                label_column='target',     # Target variable column
                split_ratio=0.2,            # Train/val split ratio
                random_state=42             # Random seed
                )

mixed_model.train(model_name=LancetMC,   # Model architecture for text classification
                train_path="train.csv", # Path to training data (CSV format)
                val_path="val.csv",     # Path to validation data (None for auto-split)
                num_epochs=50,          # Total training epochs
                hidden_size=256,        # Size of hidden layers
                num_layers=1,           # Number of hidden layers
                batch_size=256,         # Batch size for training
                learning_rate=0.001,    # Learning rate for optimizer
                dropout=0,              # Dropout rate (0-1)
                optim_name='Adam',      # Optimizer ('Adam', 'SGD', 'RAdam', etc.)
                crit_name='CELoss'      # Loss function ('CELoss' or 'BCELoss')
                )

Making predictions

from lancetnic import TextScalarClass

mixed_model = TextScalarClass()
mixed_pred = mixed_model.predict(
                model_path="mixed_model.pth", # Path to saved model
                text="Product description",  # Text input (None if only numeric)
                numeric=[0.5, 1.2]            # Numeric features as list
                )

There are two classes of basic models in LANCETNIC:

• LancetMC

from lancetnic.models import LancetMC

• LancetMCA

from lancetnic.models import LancetMC

Key Differences Between Models	LancetMC	LancetMCA
Feature
Core Architecture	Basic LSTM	LSTM + Attention
Complexity	Lower	Higher
Computational Cost	Less resource-intensive	More resource-intensive
Best For	Pure text classification	Mixed data or complex patterns
Interpretability	Standard	Provides attention weights
Sequence Handling	Good	Excellent for long sequences