add hpy opt script using optuna to yolov8

Author: khadijeh.alibabaei

README.md

@@ -36,7 +36,8 @@ apt install -y libglib2.0-0
 │   ├── README.md           <- Instructions on how to integrate your model with DEEPaaS.
 │   ├── __init__.py         <- Makes <your-model-source> a Python module
 │   ├── ...                 <- Other source code files
-│   └── config.py           <- Module to define CONSTANTS used across the AI-model python package
+│   └── config.py   <- Module to define CONSTANTS used across the AI-model python package
+│   └── hpo_yolov8 <- Hyperparameter Optimization using Optuna + Hydra + MLflow
 │
 ├── api                     <- API subpackage for the integration with DEEP API
 │   ├── __init__.py         <- Makes api a Python module, includes API interface methods
@@ -248,3 +249,7 @@ You can utilize the Swagger interface to upload your images or videos and obtain
     - A video with bounding boxes delineating objects of interest throughout.
     - A JSON string accompanying each frame, supplying bounding box coordinates, object names within the boxes, and confidence scores for the detected objects.
 
+# Hyperparameter Optimization using Optuna + Hydra + MLflow
+
+Please refer to the `README.md` inside the `yolov8_api/hpo_yolov8` directory to see how you can use these tools to automatically optimize YOLOv8 hyperparameters from the command line.
+

yolov8_api/hpo_yolov8/README.md

@@ -0,0 +1,124 @@
+# 🚀 YOLOv8 Hyperparameter Optimization with Hydra, Optuna & MLflow
+
+This repository provides a framework to **train and optimize YOLOv8 models** using **Optuna**, **Hydra**, and **MLflow**. It is designed to streamline the process of configuring, training, validating, and evaluating object detection models with automated tracking and logging.
+
+## 📁 Project Structure
+```bash
+├── configs 
+│   └── basic_train_params_pretrained.yaml  # Hydra config with model, training, and logging params 
+├── hpo.py                                  # Main script for training and hyperparameter optimization 
+├── requirements.txt                        # Python dependencies
+
+```
+## ⚙️ Features
+
+- 🔧 Hyperparameter optimization with **Optuna**
+- 🧠 Config management via **Hydra**
+- 📈 Metrics & artifact logging using **MLflow**
+- 📦 Supports training/validation with **Ultralytics YOLOv8**
+
+---
+## 🧰 Installation
+```bash
+pip install -r requirements.txt
+```
+## 🧾 Configuration
+In `config/basic_train_params_pretrained.yaml` you can
+### 🔧 Defaults & Sweeper:
+You do not need to change this part.
+```bash
+defaults:
+  - override hydra/sweeper: optuna
+```
+### 📁 MLflow Settings
+You can change the name of your MLflow experiment here.
+
+```bash
+
+mlflow_project: hpo_yolov8_kitti  # Name of the MLflow experiment
+mlflow_parent: basic_train_params_loss_30epochs_640imgsz  # The name of the parent run in MLflow
+```
+### ⚙️ Hydra + Optuna Sweep Configuration
+Here you can change the Hydra configuration.
+
+
+```bash
+
+hydra:
+  sweep:
+    dir: tmp_multirun  # Directory where multi-run outputs are stored
+  sweeper:
+    _target_: hydra_plugins.hydra_optuna_sweeper.optuna_sweeper.OptunaSweeper  # Optuna sweeper plugin
+    sampler:
+      seed: 815  # Seed for reproducibility
+    direction: maximize  # Objective direction (e.g. maximize validation fitness)
+    n_trials: 30  # Number of trials to run
+    n_jobs: 1  # Number of jobs to run in parallel
+    params:  # Hyperparameters to optimize. you can add any hyperparameters you want here
+      train_params.optimizer: choice(SGD, Adam, AdamW)
+      train_params.lr0: interval(0.001, 0.2)
+      train_params.momentum: interval(0.6, 0.999)
+      train_params.weight_decay: interval(0.00001, 0.001)
+      train_params.box: interval(0.0, 10.0)
+      train_params.cls: interval(0.0, 10.0)
+      train_params.dfl: interval(0.0, 10.0)
+```
+
+### 🏋️‍♂️ Training Parameters
+Here you can select the version of the YOLO model you want and the path to the `dataset.yaml` configuration..
+```bash
+
+train_params:
+  model: yolov8m.pt        # Pretrained model to use (YOLOv8-m)
+  epochs: 40               # Number of epochs to train
+  patience: 10             # Early stopping patience
+  box: 7.5                 # Box loss gain
+  cls: 0.5                 # Class loss gain
+  dfl: 1.5                 # Distribution Focal Loss gain
+  optimizer: 'auto'        # Optimizer (overwritten by sweep)
+  cos_lr: False            # Use cosine learning rate schedule
+  lr0: 0.01                # Initial learning rate
+  momentum: 0.937          # Momentum (used with SGD)
+  weight_decay: 0.0005     # Weight decay
+  data: &data path/to/data/kitti.yaml  # Path to dataset config
+  batch: &batch 16         # Batch size
+  imgsz: &imgsz 640        # Image size
+  save: True               # Save model checkpoints
+  cache: True              # Cache images for faster training
+  device: &device 0        # GPU device
+  workers: 8               # Number of data loading workers
+  rect: &rect True         # Use rectangular training batches
+  plots: &plots True       # Save training plots
+```
+### 🧪 Validation Parameters
+```bash
+
+val_test_params:
+  data: *data              # Use same dataset as training
+  imgsz: *imgsz            # Same image size
+  batch: *batch            # Same batch size
+  device: *device          # Same GPU
+  plots: *plots            # Generate plots
+  rect: *rect              # Use rectangular validation batches
+```
+## How to run
+To launch training and start the Optuna hyperparameter optimization:
+
+```bash
+python3 hpo.py --multirun
+```
+✅ This will:
+
+- Run training and validation across **30 trials**
+- Optimize the defined hyperparameters using **Optuna**
+- Log all **metrics**, **configs**, and **artifacts** to **MLflow**
+- Store outputs in the **`tmp_multirun/`** directory
+
+## 📚 References
+
+- [Ultralytics YOLOv8 Documentation](https://docs.ultralytics.com/)
+- [Hydra – Elegant Configuration Management](https://hydra.cc/)
+- [Hydra Optuna Sweeper Plugin](https://github.com/facebookresearch/hydra/tree/main/plugins/hydra_optuna_sweeper)
+- [Optuna – Hyperparameter Optimization Framework](https://optuna.org/)
+- [MLflow – Open Source Experiment Tracking](https://mlflow.org/)
+s

yolov8_api/hpo_yolov8/configs/basic_train_params_pretrained.yaml

@@ -0,0 +1,57 @@
+defaults:
+  - override hydra/sweeper: optuna
+
+mlflow_project: hpo_yolov8_kitti
+mlflow_parent: basic_train_params_loss_30epochs_640imgsz
+
+hydra:
+  sweep:
+    dir: tmp_multirun
+  sweeper:
+    _target_: hydra_plugins.hydra_optuna_sweeper.optuna_sweeper.OptunaSweeper
+    sampler:
+      seed: 815
+    direction: maximize
+    n_trials: 30
+    n_jobs: 1
+    params:
+      train_params.optimizer: choice(SGD, Adam, AdamW)
+      train_params.lr0: interval(0.001, 0.2)
+      train_params.momentum: interval(0.6, 0.999)
+      train_params.weight_decay: interval(0.00001, 0.001)
+      train_params.box: interval(0.0, 10.0)
+      train_params.cls: interval(0.0, 10.0)
+      train_params.dfl: interval(0.0, 10.0)
+
+train_params:
+    model: yolov8m.pt
+    epochs: 40
+    patience: 10
+    # loss params
+    box: 7.5
+    cls: 0.5
+    dfl: 1.5
+    # train params
+    optimizer: 'auto'
+    cos_lr: False
+    lr0: 0.01
+    momentum: 0.937
+    weight_decay: 0.0005
+    # default params
+    data: &data configs/kitti.yaml
+    batch: &batch 16
+    imgsz: &imgsz 640
+    save: True
+    cache: True
+    device: &device 0
+    workers: 8
+    rect: &rect True
+    plots: &plots True
+val_test_params:
+    # default params
+    data: *data
+    imgsz: *imgsz
+    batch: *batch
+    device: *device
+    plots: *plots
+    rect: *rect