Adding mypy.

.github/workflows/test.yaml

@@ -48,6 +48,11 @@ jobs:
           uv run ruff check ./src ./app ./tests
           uv run ruff format --check ./src ./app ./tests
 
+      # Run mypy type checking
+      - name: Run mypy type checking
+        run: |
+          uv run mypy src/ app/ tests/ --ignore-missing-imports
+
       # Run pytest (excludes audio-dependent modules like speech_to_text)
       - name: Run tests with pytest
         run: |

.gitignore

@@ -195,3 +195,4 @@ src/models/*.json
 
 
 app_simple.py
+mypy_output.txt

.pre-commit-config.yaml

@@ -25,6 +25,14 @@ repos:
 
   - repo: local
     hooks:
+      - id: mypy
+        name: mypy
+        entry: uv run mypy
+        language: system
+        types: [python]
+        files: ^(src/|app/|tests/).*\.py$
+        args: [--ignore-missing-imports]
+
       - id: pytest
         name: pytest
         entry: uv run python -m pytest

Makefile

@@ -1,5 +1,5 @@
 # Makefile for Sentiment Analysis
-.PHONY: help install test lint format clean run
+.PHONY: help install test lint format clean run type-check
 
 help: ## Show available commands
 	@grep -E '^[a-zA-Z_-]+:.*?## .*$$' $(MAKEFILE_LIST) | sort | awk 'BEGIN {FS = ":.*?## "}; {printf "\033[36m%-15s\033[0m %s\n", $$1, $$2}'
@@ -20,6 +20,12 @@ lint: ## Check and fix code quality
 	uv run ruff check --fix ./src ./app ./tests
 	uv run ruff format ./src ./app ./tests
 
+type-check: ## Run mypy type checking
+	uv run mypy src/ app/ tests/
+
+type-check-strict: ## Run mypy with strict mode
+	uv run mypy --strict src/ app/
+
 format: ## Format code only
 	uv run ruff format ./src ./app ./tests
 

README.md

@@ -1,10 +1,21 @@
 [![Linting: Ruff](https://img.shields.io/badge/linting-ruff-yellowgreen)](https://github.com/charliermarsh/ruff)
+[![Type Checking: mypy](https://img.shields.io/badge/type%20checking-mypy-blue)](http://mypy-lang.org/)
 [![CI: Passed](https://img.shields.io/badge/CI-Passed-brightgreen)](https://github.com/jvachier/Sentiment_Analysis/actions/workflows/test.yaml)
+[![Tests: pytest](https://img.shields.io/badge/tests-pytest-orange)](https://docs.pytest.org/)
+[![Pre-commit](https://img.shields.io/badge/pre--commit-enabled-brightgreen?logo=pre-commit)](https://github.com/pre-commit/pre-commit)
 [![Deep Learning](https://img.shields.io/badge/Deep%20Learning-TensorFlow-orange)](https://www.tensorflow.org/)
 [![Keras](https://img.shields.io/badge/Keras-red)](https://keras.io/)
 [![TensorFlow](https://img.shields.io/badge/TensorFlow-2.0%2B-orange)](https://www.tensorflow.org/)
 [![Python](https://img.shields.io/badge/Python-3.11%2B-blue)](https://www.python.org/)
 [![uv](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/uv/main/assets/badge/v0.json)](https://github.com/astral-sh/uv)
+[![NLP](https://img.shields.io/badge/NLP-Natural%20Language%20Processing-green)](https://en.wikipedia.org/wiki/Natural_language_processing)
+[![Transformers](https://img.shields.io/badge/Transformers-From%20Scratch-blueviolet)](https://arxiv.org/abs/1706.03762)
+[![Neural Machine Translation](https://img.shields.io/badge/Neural-Machine%20Translation-purple)](https://en.wikipedia.org/wiki/Neural_machine_translation)
+[![Sentiment Analysis](https://img.shields.io/badge/Sentiment-Analysis-pink)](https://en.wikipedia.org/wiki/Sentiment_analysis)
+[![Speech Recognition](https://img.shields.io/badge/Speech-Recognition-cyan)](https://en.wikipedia.org/wiki/Speech_recognition)
+[![Gradio](https://img.shields.io/badge/UI-Gradio-ff7c00)](https://gradio.app/)
+[![Optuna](https://img.shields.io/badge/Hyperparameter-Optuna-lightblue)](https://optuna.org/)
+[![Dash](https://img.shields.io/badge/Dashboard-Dash-blue)](https://dash.plotly.com/)
 [![License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
 
 # Sentiment Analysis and Translation
@@ -292,6 +303,17 @@ Sentiment_Analysis/
 
 This Kaggle notebook provides a detailed tutorial on the transformer architecture implemented in this repository.
 
+---
+
+## Live Demo
+
+** HuggingFace Space: English-to-French Translator**
+- Try the enhanced Transformer model live in your browser
+- Real-time translation with greedy and beam search decoding
+- No installation required - instant access
+- [Launch Demo on HuggingFace](https://huggingface.co/spaces/Jvachier/transformer-nmt-en-fr)
+
+
 ---
 
 ## Customization

gradio_apps/README.md

@@ -18,6 +18,16 @@ tags:
   - tensorflow
   - keras
   - from-scratch
+  - nlp
+  - seq2seq
+  - attention-mechanism
+  - encoder-decoder
+  - deep-learning
+  - machine-translation
+  - multilingual
+  - text-generation
+  - custom-model
+  - educational
 ---
 
 # English to French Enhanced Transformer

gradio_apps/requirements.txt

@@ -1,3 +1,4 @@
-gradio==4.0.0
-tensorflow==2.19.0
-numpy==1.26.0
+huggingface-hub==0.25.1
+tensorflow==2.20.0
+numpy>=1.26.2
+audioop-lts

pyproject.toml

@@ -48,6 +48,8 @@ dev = [
     "ruff>=0.2.1",
     "scikit-optimize>=0.9.0",
     "pre-commit>=4.0.0",
+    "mypy>=1.8.0",
+    "types-tensorflow>=2.16.0",
 ]
 macos = [
     "tensorflow-io-gcs-filesystem<0.35.0",
@@ -69,3 +71,41 @@ build-backend = "hatchling.build"
 
 [tool.hatch.build.targets.wheel]
 packages = ["src", "app"]
+
+[tool.mypy]
+python_version = "3.11"
+warn_return_any = false
+warn_unused_configs = true
+disallow_untyped_defs = false
+disallow_incomplete_defs = false
+check_untyped_defs = true
+disallow_untyped_calls = false
+warn_redundant_casts = true
+warn_unused_ignores = true
+strict_optional = true
+no_implicit_optional = true
+ignore_missing_imports = true
+show_error_codes = true
+pretty = true
+
+[[tool.mypy.overrides]]
+module = [
+    "tensorflow.*",
+    "keras.*",
+    "nltk.*",
+    "vosk.*",
+    "dash.*",
+    "optuna.*",
+    "transformers.*",
+    "polars.*",
+]
+ignore_missing_imports = true
+ignore_errors = true
+
+[[tool.mypy.overrides]]
+module = [
+    "src.translation_french_english",
+    "src.modules.optuna_transformer",
+    "tests.test_transformer_model",
+]
+disable_error_code = ["dict-item", "misc"]

src/modules/data_preprocess_nltk.py

@@ -78,7 +78,7 @@ def encode(self, text_tensor, label):
         """
         text = text_tensor.numpy().decode("utf-8")
         text = self.preprocess_text(text)
-        encoded_text = self.tokenizer.texts_to_sequences([text])[0]
+        encoded_text = self.tokenizer.texts_to_sequences([text])[0]  # type: ignore[union-attr]
         return encoded_text, label
 
     def fit_tokenizer(self, ds_raw):
@@ -163,7 +163,7 @@ def encode(self, text_tensor):
         """
         text = text_tensor.numpy().decode("utf-8")
         text = self.preprocess_text(text)
-        encoded_text = self.tokenizer.texts_to_sequences([text])[0]
+        encoded_text = self.tokenizer.texts_to_sequences([text])[0]  # type: ignore[union-attr]
         return encoded_text
 
     def fit_tokenizer(self, ds_raw):

src/modules/data_processor.py

@@ -2,7 +2,7 @@
 import tensorflow as tf
 import string
 import re
-from typing import Tuple, Dict
+from typing import Tuple, Dict, Optional
 
 
 class DatasetProcessor:
@@ -20,14 +20,20 @@ def __init__(self, file_path: str, delimiters: str = r"|"):
         """
         self.file_path = file_path
         self.delimiters = delimiters
-        self.split_df = None
+        self.split_df: Optional[pl.DataFrame] = None
+        self.df: Optional[pl.DataFrame] = None
 
     def load_data(self) -> None:
         """Load the Parquet file using Polars."""
         self.df = pl.read_parquet(self.file_path)
 
     def process_data(self) -> None:
         """Process the dataset by splitting, cleaning, and tokenizing."""
+        if self.df is None:
+            raise ValueError(
+                "Data must be loaded before processing. Call load_data() first."
+            )
+
         # Split the 'en' column into rows based on delimiters
         if "en" in self.df.columns:
             en_split = self.df.select(pl.col("en").str.split(self.delimiters)).explode(
@@ -76,6 +82,9 @@ def shuffle_and_split(
         """
 
         # Calculate the number of samples for validation and test sets
+        if self.split_df is None:
+            raise ValueError("Data must be processed before splitting")
+
         num_val_samples = int(val_split * len(self.split_df))
         num_train_samples = len(self.split_df) - 2 * num_val_samples
 

src/modules/load_data.py

@@ -1,6 +1,6 @@
 import pandas as pd
 import tensorflow as tf
-from pydantic import BaseModel, FilePath, Field, ValidationError
+from pydantic import BaseModel, Field, ValidationError
 from src.modules.utils import DatasetPaths
 
 
@@ -9,7 +9,7 @@ class DataLoaderConfig(BaseModel):
     Configuration for the DataLoader class.
     """
 
-    data_path: FilePath = Field(
+    data_path: str = Field(
         default=DatasetPaths.RAW_DATA.value,
         description="Path to the CSV file containing the dataset.",
     )

src/modules/model_bert_other.py

@@ -141,7 +141,9 @@ def build_model(self, num_classes):
         dropout = tf.keras.layers.Dropout(0.3)(cls_token)
         output = tf.keras.layers.Dense(1, activation="sigmoid")(dropout)
 
-        model = tf.keras.Model(inputs=[input_ids, attention_mask], outputs=output)
+        model: tf.keras.Model = tf.keras.Model(
+            inputs=[input_ids, attention_mask], outputs=output
+        )
         model.compile(
             optimizer=tf.keras.optimizers.legacy.RMSprop(
                 learning_rate=self.learning_rate

src/modules/model_sentiment_analysis.py

@@ -62,7 +62,10 @@ def train_and_evaluate(
                 callbacks=[early_stopping_callback, callbacks_model],
             )
             test_results = model.evaluate(test_data)
-        logging.info("Test Accuracy: {:.2f}%".format(test_results[1] * 100))
+        if isinstance(test_results, list):
+            logging.info("Test Accuracy: {:.2f}%".format(test_results[1] * 100))
+        else:
+            logging.info("Test Accuracy: {:.2f}%".format(test_results * 100))
 
     def inference_model(
         self, model: tf.keras.Model, text_vec: tf.keras.layers.TextVectorization
@@ -77,10 +80,10 @@ def inference_model(
         Returns:
             tf.keras.Model: An inference model for sentiment prediction.
         """
-        inputs = tf.keras.Input(shape=(1,), dtype=tf.string)
+        inputs: tf.Tensor = tf.keras.Input(shape=(1,), dtype=tf.string)
         process_inputs = text_vec(inputs)
         outputs = model(process_inputs)
-        inference_model = tf.keras.Model(inputs=inputs, outputs=outputs)
+        inference_model: tf.keras.Model = tf.keras.Model(inputs=inputs, outputs=outputs)
         return inference_model
 
 
@@ -121,7 +124,7 @@ def optimize(
             test_data (tf.data.Dataset): Test dataset.
         """
 
-        def _objective(trial):
+        def _objective(trial: optuna.trial.Trial) -> float:
             """
             Objective function for Optuna to optimize the model's hyperparameters.
 
@@ -141,7 +144,7 @@ def _objective(trial):
                 )
             )
             n_layers_bidirectional = trial.suggest_int("n_units_bidirectional", 1, 3)
-            for i in range(n_layers_bidirectional):
+            for i in range(n_layers_bidirectional):  # type: int
                 num_hidden_bidirectional = trial.suggest_int(
                     f"n_units_bidirectional_l{i}", 64, 128, log=True
                 )
@@ -165,8 +168,8 @@ def _objective(trial):
 
             model.add(tf.keras.layers.Dropout(self.dropout_rate))
             n_layers_nn = trial.suggest_int("n_layers_nn", 1, 2)
-            for i in range(n_layers_nn):
-                num_hidden_nn = trial.suggest_int(f"n_units_nn_l{i}", 64, 128, log=True)
+            for j in range(n_layers_nn):
+                num_hidden_nn = trial.suggest_int(f"n_units_nn_l{j}", 64, 128, log=True)
                 model.add(tf.keras.layers.Dense(num_hidden_nn, activation="gelu"))
 
             model.add(tf.keras.layers.Dropout(self.dropout_rate))
@@ -199,7 +202,9 @@ def _objective(trial):
                 )
             # Evaluate the model accuracy on the validation set.
             score = model.evaluate(test_data, verbose=1)
-            return score[1]
+            if isinstance(score, list):
+                return float(score[1])
+            return float(score)
 
         # Create an Optuna study
         study = optuna.create_study(
@@ -267,7 +272,7 @@ def get_model_api(self) -> tf.keras.Model:
         dropout_layer2 = tf.keras.layers.Dropout(self.dropout_rate)(dense_layer2)
         dense_layer3 = tf.keras.layers.Dense(32, activation="gelu")(dropout_layer2)
         outputs = tf.keras.layers.Dense(1, activation="sigmoid")(dense_layer3)
-        model = tf.keras.Model(inputs=inputs, outputs=outputs)
+        model: tf.keras.Model = tf.keras.Model(inputs=inputs, outputs=outputs)
         model.compile(
             optimizer=tf.keras.optimizers.RMSprop(),
             loss=tf.keras.losses.BinaryCrossentropy(),
@@ -282,13 +287,9 @@ def get_config(self) -> dict:
         Returns:
             dict: A dictionary containing the model's configuration.
         """
-        config = super().get_config()
-        config.update(
-            {
-                "embedding_dim": self.embedding_dim,
-                "lstm_units": self.lstm_units,
-                "dropout_rate": self.dropout_rate,
-                "max_token": self.max_token,
-            }
-        )
-        return config
+        return {
+            "embedding_dim": self.embedding_dim,
+            "lstm_units": self.lstm_units,
+            "dropout_rate": self.dropout_rate,
+            "max_token": self.max_token,
+        }

src/modules/optuna_transformer.py

@@ -61,7 +61,9 @@ def build_transformer_model(
         dropout_outputs
     )
 
-    transformer = tf.keras.Model([encoder_inputs, decoder_inputs], final_outputs)
+    transformer: tf.keras.Model = tf.keras.Model(
+        [encoder_inputs, decoder_inputs], final_outputs
+    )
 
     # Compile the model
     transformer.compile(

src/modules/sentiment_analysis_utils.py

@@ -58,7 +58,7 @@ def create_or_load_inference_model(
         return tf.keras.models.load_model(inference_model_path)
 
     logging.info("Creating and saving the inference model.")
-    trainer = ModelTrainer()
+    trainer = ModelTrainer(config_path=ModelPaths.MODEL_TRAINER_CONFIG.value)
     inference_model = trainer.inference_model(model, text_vec)
     inference_model.save(inference_model_path)
     return inference_model

src/modules/speech_to_text.py

@@ -45,7 +45,7 @@ def __init__(self, model_path: str):
         )
         self.stream.start_stream()
         self.rec = vosk.KaldiRecognizer(self.model, 16000)
-        self.recognized_text = []
+        self.recognized_text: list[str] = []
         self.recording = False
 
     def start_recording(self) -> None:

src/modules/transformer_components.py

@@ -118,7 +118,7 @@ def build(self, input_shape):
         self.layernorm_3 = tf.keras.layers.LayerNormalization()
         super().build(input_shape)
 
-    def call(self, inputs, encoder_outputs, mask=None):
+    def call(self, inputs, encoder_outputs, mask=None):  # type: ignore[override]
         causal_mask = self.get_causal_attention_mask(inputs)
         if mask is not None:
             padding_mask = tf.cast(mask[:, tf.newaxis, tf.newaxis, :], dtype="float32")
@@ -214,8 +214,8 @@ def evaluate_bleu(
             references.append([ref_sentence])
 
     # Calculate BLEU score
-    bleu_score = corpus_bleu(
-        references, candidates, smoothing_function=smoothing_function
+    bleu_score = float(
+        corpus_bleu(references, candidates, smoothing_function=smoothing_function)
     )
     logging.info(f"BLEU score evaluation completed: {bleu_score:.4f}")
     return bleu_score