evaluate.py

import numpy as np

embedding_table = np.load("trained_embeddings.npy")
word_to_idx = np.load("word_to_idx.npy", allow_pickle=True).item()
idx_to_word = {i: w for w, i in word_to_idx.items()}

def cosine_similarity(v1, v2):
    return np.dot(v1, v2) / (np.linalg.norm(v1) * np.linalg.norm(v2))

# Check specific word pairs
print("Cosine similarity AFTER training:")
print(f"crime vs punishment: {cosine_similarity(embedding_table[word_to_idx['crime']], embedding_table[word_to_idx['punishment']]):.4f}")
print(f"crime vs murder:     {cosine_similarity(embedding_table[word_to_idx['crime']], embedding_table[word_to_idx['murder']]):.4f}")
print(f"crime vs guilt:      {cosine_similarity(embedding_table[word_to_idx['crime']], embedding_table[word_to_idx['guilt']]):.4f}")
print(f"crime vs police:     {cosine_similarity(embedding_table[word_to_idx['crime']], embedding_table[word_to_idx['police']]):.4f}")
print(f"crime vs love:       {cosine_similarity(embedding_table[word_to_idx['crime']], embedding_table[word_to_idx['love']]):.4f}")
print(f"crime vs happy:      {cosine_similarity(embedding_table[word_to_idx['crime']], embedding_table[word_to_idx['happy']]):.4f}")

# Top 10 most similar words to crime
print("\nTop 10 words most similar to 'crime':")
crime_vec = embedding_table[word_to_idx['crime']]
similarities = []

for word, idx in word_to_idx.items():
    sim = cosine_similarity(crime_vec, embedding_table[idx])
    similarities.append((word, sim))

similarities.sort(key=lambda x: x[1], reverse=True)
for word, sim in similarities[:10]:
    print(f"  {word}: {sim:.4f}")