src/app.py

import streamlit as st
import tensorflow as tf
import os
import numpy as np
from scipy.spatial.distance import cosine

from two_tower_model import TwoTowerModel

TWO_TOWER_MODEL_SAVE_PATH = os.path.join("src","two_tower_model.keras")
TWO_TOWER_VOCAB_PATH = os.path.join("src","two_tower_vocabulary.json")
TWO_TOWER_EMBEDDING_PATH = os.path.join("src","two_tower_all_embeddings.npy")


st.set_page_config(page_title="Kelime Benzerliği Uygulaması", layout="wide")

st.title("Kelime Benzerliği Arama")
st.write("Girdiğiniz kelimeye en yakın kelimeleri bulmak için modelleri kullanın.")

def load_vocabulary(vocab_path: str) -> list[str]:
    import json
    with open(vocab_path, 'r', encoding='utf-8') as f:
        return json.load(f)

@st.cache_resource
def load_models_and_vocab():
    
    os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' 

    two_tower_model_loaded = None
    two_tower_vocab = []
    all_vocab_embeddings = None

    try:
        two_tower_vocab = load_vocabulary(TWO_TOWER_VOCAB_PATH)
    except Exception as e:
        st.error(f"Two-Tower kelime dağarcığı yüklenirken hata oluştu: {e}.")
        two_tower_vocab = []

    try:
        all_vocab_embeddings = np.load(TWO_TOWER_EMBEDDING_PATH)
    except Exception as e:
        st.error(f"Kelime embeddingleri yüklenirken hata oluştu: {e}.")
        all_vocab_embeddings = None

    try:
        custom_objects = {"TwoTowerModel": TwoTowerModel} 
        two_tower_model_loaded = tf.keras.models.load_model(TWO_TOWER_MODEL_SAVE_PATH, custom_objects=custom_objects)

        
    except Exception as e:
        st.error(f"Two-Tower modeli yüklenirken hata oluştu: {e}.")

    return two_tower_model_loaded, two_tower_vocab, all_vocab_embeddings

two_tower_model, vocabulary, all_vocab_embeddings = load_models_and_vocab()

def find_similar_words_two_tower(query_word, two_tower_model, vocabulary, all_vocab_embeddings, top_n=10):
    if two_tower_model is None or not vocabulary or all_vocab_embeddings is None:
        st.warning("Two-Tower modeli veya kelime dağarcığı/gömmeler yüklenemedi.")
        return []
    
    if query_word not in vocabulary:
        st.warning(f"''{query_word}'' kelimesi kelime dağarcığında bulunmuyor. Benzerlikler doğru olmayabilir.")
        query_embedding = two_tower_model.get_embedding(tf.constant([vocabulary[0]]))
    else:
        query_embedding = two_tower_model.get_embedding(tf.constant([query_word]))

    similarities = []
    for i, word_embedding in enumerate(all_vocab_embeddings):
        similarity = 1 - cosine(query_embedding.numpy().flatten(), word_embedding.flatten())
        similarities.append((vocabulary[i], similarity))
    
    similarities = sorted(similarities, key=lambda x: x[1], reverse=True)
    similarities = [s for s in similarities if s[0] != query_word]
    return similarities[:top_n]

query_word_input = st.selectbox("Kelime Seçin", options=vocabulary)
if query_word_input:
    st.subheader(f"''{query_word_input}'' kelimesi için Benzerlik Sonuçları")

    st.markdown("#### Two-Tower Model Sonuçları")
    two_tower_results = find_similar_words_two_tower(query_word_input, two_tower_model, vocabulary, all_vocab_embeddings)
    if two_tower_results:
        for word, score in two_tower_results:
            st.write(f"{word}: {score:.4f}")
    else:
        st.write("Two-Tower modeli için benzer kelime bulunamadı veya model/kelime dağarcığı/embeddingler yüklenemedi.")