fastText

import fasttext

# Train word embeddings on text file
model = fasttext.train_unsupervised(
    input='text_corpus.txt',
    model='skipgram',  # or 'cbow'
    dim=100,           # vector dimensions
    epoch=5,           # training epochs
    lr=0.05,          # learning rate
    minCount=5,       # minimum word frequency
    minn=3,           # min character n-gram length
    maxn=6,           # max character n-gram length
    thread=4          # number of threads
)

# Get word vector
vector = model.get_word_vector('hello')
print(f"Vector shape: {vector.shape}")

# Find similar words
similar_words = model.get_nearest_neighbors('king', k=10)
for score, word in similar_words:
    print(f"{word}: {score:.4f}")

# Handle out-of-vocabulary words
oov_vector = model.get_word_vector('unknownword123')
print("OOV word vector computed from subwords")

# Prepare training data (format: __label__class_name text)
# Example file content:
# __label__positive This movie is great!
# __label__negative This movie is terrible.

# Train classification model
model = fasttext.train_supervised(
    input='train.txt',
    epoch=25,
    lr=1.0,
    wordNgrams=2,     # use bigrams
    dim=100,          # vector dimensions
    loss='softmax',   # or 'hs' for hierarchical softmax
    thread=4
)

# Test the model
test_results = model.test('test.txt')
print(f"Accuracy: {test_results[1]:.4f}")
print(f"Precision: {test_results[0]:.4f}")

# Make predictions
predictions = model.predict(['This is a great product!'], k=3)
for label, score in zip(predictions[0][0], predictions[1][0]):
    print(f"{label}: {score:.4f}")

# Quantize model for smaller size
model.quantize(input='train.txt', retrain=True)
model.save_model('model_quantized.ftz')

# Automatic hyperparameter tuning
model = fasttext.train_supervised(
    input='train.txt',
    autotuneValidationFile='validation.txt',  # validation set for tuning
    autotuneDuration=600,  # tuning time in seconds
    autotuneModelSize='2M'  # target model size
)

# Manual hyperparameter search
import itertools

def grid_search():
    learning_rates = [0.1, 0.5, 1.0]
    epochs = [5, 10, 25]
    word_ngrams = [1, 2, 3]
    
    best_accuracy = 0
    best_params = {}
    
    for lr, epoch, ngram in itertools.product(learning_rates, epochs, word_ngrams):
        model = fasttext.train_supervised(
            input='train.txt',
            lr=lr,
            epoch=epoch,
            wordNgrams=ngram,
            dim=100,
            verbose=0
        )
        
        results = model.test('validation.txt')
        accuracy = results[1]
        
        if accuracy > best_accuracy:
            best_accuracy = accuracy
            best_params = {'lr': lr, 'epoch': epoch, 'wordNgrams': ngram}
            
        model = None  # Free memory
    
    return best_params, best_accuracy

# Run grid search
best_params, best_acc = grid_search()
print(f"Best params: {best_params}")
print(f"Best accuracy: {best_acc:.4f}")