Merge pull request #1 from RihabFekii/mlflow

MLflow logging and adding test stage

Makefile

@@ -1,6 +1,18 @@
+#!make
+include .env
+
 PYTHON_INTERPRETER = python3
 
 env:
  @echo ">>> Creating a python virtual environment with venv"
  $(PYTHON_INTERPRETER) -m venv env
- @echo ">>> A new virtual env is created. Activate it with:\nsource env/bin/activate ."
+ @echo ">>> A new virtual env is created. Activate it with:\nsource env/bin/activate ."
+
+
+mlflow:
+ @echo ">>> Authenticating to MLflow remote server on DagsHub"
+ export MLFLOW_TRACKING_URI=$(MLFLOW_TRACKING_URI)
+ export MLFLOW_TRACKING_USERNAME=$(MLFLOW_TRACKING_USERNAME) 
+ export MLFLOW_TRACKING_PASSWORD=$(MLFLOW_TRACKING_PASSWORD) 
+ @echo ">>> Authenticating successful!"
+

dvc.lock

@@ -8,21 +8,34 @@ stages:
  size: 28867246
  nfiles: 1482
  - path: params.yaml
- md5: a061a8a3138da65df56595927ca325f6
- size: 216
+ md5: a0c123b909925d3386149e8b024832a0
+ size: 215
  - path: src/train.py
- md5: 8cde4fe436744929ddbaeaadb37aad71
- size: 986
+ md5: 6746a939a17bbf4e2e2bd23a226b69ec
+ size: 1774
  - path: src/utils.py
- md5: 3ecd480350d0d8edb7e4ba533bb555ad
- size: 1253
+ md5: 5440f0cb53d7c318e2233a2414c32147
+ size: 1228
  outs:
  - path: models/model.pt
- md5: 31c95b64b4a92e32e26ca67d99ffc401
- size: 22495096
+ md5: 7fe4c6173c334d2e926e71236366b8e0
+ size: 6226104
  - path: reports/train_metrics.csv
- md5: 30cf68a8a216e4dcfbde8717fa882e9f
- size: 7056
+ md5: 464bce4270bb543ea8741659367b6b9d
+ size: 1008
  - path: reports/train_params.yaml
- md5: 5c58f717d9f2b23de0e7483ba7fbf2b2
- size: 1502
+ md5: 2ea9f04ec479dd1db02f44263b2c8328
+ size: 1501
+ test:
+ cmd: python src/predict.py
+ deps:
+ - path: data/raw/wildfire-raw-yolov8
+ md5: ffe9083905cf64817ca61146b3356bb6.dir
+ size: 28867246
+ nfiles: 1482
+ - path: models/model.pt
+ md5: 7fe4c6173c334d2e926e71236366b8e0
+ size: 6226104
+ - path: src/predict.py
+ md5: fa6b9e2cfac6cbe9a40e8d182a0a2f4c
+ size: 594

dvc.yaml

@@ -13,3 +13,9 @@ stages:
  metrics:
  - reports/train_metrics.csv:
  cache: false
+ test:
+ cmd: python src/predict.py
+ deps:
+ - data/raw/wildfire-raw-yolov8
+ - models/model.pt
+ - src/predict.py

params.yaml

@@ -1,9 +1,9 @@
-model_type: yolov8s.pt
+model_type: yolov8n.pt
 pretrained: True
 seed: 0
 imgsz: 640 
 batch: 8 
-epochs: 20
+epochs: 1
 optimizer: SGD # other choices=['SGD', 'Adam', 'AdamW', 'RMSProp']
 lr0: 0.01 # learning rate 
-name: 'yolov8s_exp_v0' # experiment name
+name: 'yolov8n_exp_v0' # experiment name

reports/train_metrics.csv

@@ -1,21 +1,3 @@
  epoch, train/box_loss, train/cls_loss, train/dfl_loss, metrics/precision(B), metrics/recall(B), metrics/mAP50(B), metrics/mAP50-95(B), val/box_loss, val/cls_loss, val/dfl_loss, lr/pg0, lr/pg1, lr/pg2
- 0, 2.074, 3.6688, 1.7378, 0.29774, 0.32591, 0.23183, 0.08104, 2.2504, 2.6345, 1.8202, 0.070462, 0.0032821, 0.0032821
- 1, 1.767, 1.7208, 1.4953, 0.64651, 0.53061, 0.62556, 0.29666, 1.8773, 1.8427, 1.3876, 0.040134, 0.0062879, 0.0062879
- 2, 1.7171, 1.707, 1.3988, 0.06729, 0.09524, 0.04705, 0.02481, 3.0641, 4.3916, 2.5411, 0.0094766, 0.0089638, 0.0089638
- 3, 1.7341, 1.8335, 1.4552, 0.38806, 0.47619, 0.41508, 0.17576, 1.9845, 2.1096, 1.6549, 0.008515, 0.008515, 0.008515
- 4, 1.691, 1.3376, 1.4286, 0.49323, 0.42177, 0.41415, 0.16903, 2.213, 2.0351, 1.7182, 0.008515, 0.008515, 0.008515
- 5, 1.6586, 1.3538, 1.3919, 0.62787, 0.38095, 0.39228, 0.18293, 2.0223, 2.3359, 1.6252, 0.00802, 0.00802, 0.00802
- 6, 1.6037, 1.2676, 1.3543, 0.66557, 0.47619, 0.5516, 0.25586, 1.8962, 1.6098, 1.4971, 0.007525, 0.007525, 0.007525
- 7, 1.658, 1.2554, 1.3514, 0.58698, 0.68027, 0.6487, 0.29438, 1.7871, 1.3586, 1.4083, 0.00703, 0.00703, 0.00703
- 8, 1.609, 1.1162, 1.3305, 0.6504, 0.5102, 0.53742, 0.25941, 1.9, 1.6061, 1.4134, 0.006535, 0.006535, 0.006535
- 9, 1.5784, 1.1193, 1.3391, 0.58894, 0.70068, 0.60387, 0.26491, 1.8898, 1.466, 1.4003, 0.00604, 0.00604, 0.00604
- 10, 1.6081, 1.0872, 1.332, 0.54884, 0.53741, 0.56067, 0.26937, 1.89, 1.6558, 1.4405, 0.005545, 0.005545, 0.005545
- 11, 1.6216, 1.0673, 1.3576, 0.6775, 0.7415, 0.76254, 0.40175, 1.6952, 1.1328, 1.3126, 0.00505, 0.00505, 0.00505
- 12, 1.5426, 1.0814, 1.3183, 0.63694, 0.73995, 0.67857, 0.33739, 1.8035, 1.2909, 1.3621, 0.004555, 0.004555, 0.004555
- 13, 1.527, 0.9778, 1.2827, 0.76852, 0.76871, 0.80068, 0.40906, 1.6945, 1.0812, 1.3067, 0.00406, 0.00406, 0.00406
- 14, 1.4765, 0.93611, 1.2862, 0.74653, 0.84152, 0.85997, 0.47623, 1.6398, 1.0427, 1.3348, 0.003565, 0.003565, 0.003565
- 15, 1.4623, 0.90297, 1.2588, 0.83993, 0.80952, 0.88026, 0.47516, 1.6723, 0.98866, 1.2993, 0.00307, 0.00307, 0.00307
- 16, 1.4283, 0.86243, 1.2558, 0.81461, 0.82313, 0.90594, 0.46365, 1.6496, 0.93395, 1.3134, 0.002575, 0.002575, 0.002575
- 17, 1.4264, 0.83566, 1.2381, 0.85204, 0.82993, 0.90175, 0.49576, 1.5633, 0.94734, 1.2569, 0.00208, 0.00208, 0.00208
- 18, 1.3657, 0.84115, 1.214, 0.87301, 0.87755, 0.922, 0.4962, 1.5714, 0.88971, 1.2903, 0.001585, 0.001585, 0.001585
- 19, 1.3289, 0.8171, 1.1956, 0.85437, 0.86395, 0.92797, 0.51129, 1.5971, 0.85184, 1.2837, 0.00109, 0.00109, 0.00109
+ 0, 1.9822, 4.1505, 1.6088, 0.00279, 0.83673, 0.04395, 0.0187, 2.0868, 4.2883, 1.5972, 0.070462, 0.0032821, 0.0032821
+ 1, 1.7723, 2.9226, 1.3586, 0.25019, 0.33333, 0.19749, 0.07728, 2.2085, 3.4761, 1.5367, 0.037187, 0.0033408, 0.0033408

reports/train_params.yaml

@@ -1,8 +1,8 @@
 task: detect
 mode: train
-model: yolov8s.pt
+model: yolov8n.pt
 data: /Users/rihabfeki/Desktop/wildfire-smoke-detector/data/raw/wildfire-raw-yolov8/data.yaml
-epochs: 20
+epochs: 2
 patience: 50
 batch: 8
 imgsz: 640
@@ -12,7 +12,7 @@ cache: false
 device: null
 workers: 8
 project: null
-name: yolov8s_exp_v0
+name: yolov8n_exp_v0
 exist_ok: false
 pretrained: true
 optimizer: SGD
@@ -92,4 +92,4 @@ copy_paste: 0.0
 cfg: null
 v5loader: false
 tracker: botsort.yaml
-save_dir: /Users/rihabfeki/Desktop/wildfire-smoke-detector/runs/detect/yolov8s_exp_v0
+save_dir: /Users/rihabfeki/Desktop/wildfire-smoke-detector/runs/detect/yolov8n_exp_v0

requirements.txt

@@ -1,3 +1,6 @@
 dvc==2.51.0
 ultralytics==8.0.58
+python-dotenv==1.0.0
+mlflow==2.2.2
+
 

src/predict.py

@@ -0,0 +1,23 @@
+import os
+import glob
+from pathlib import Path
+
+from ultralytics import YOLO
+
+
+root_dir = Path(__file__).resolve().parents[1] # root directory
+path_model_weights = os.path.join(root_dir, 'models/model.pt')
+data_dir = os.path.join(root_dir, "data/raw/wildfire-raw-yolov8")
+test_data_path = os.path.join(data_dir, 'test/images')
+
+
+if __name__ == '__main__':
+
+ #load trained model 
+ model = YOLO(path_model_weights)
+
+ for images in glob.glob(f"{test_data_path}/*.jpg"):
+ predictions = model.predict(images, save=True)
+ print(predictions)
+ print(type(predictions)) 
+

src/train.py

@@ -1,43 +1,67 @@
 import os
 from pathlib import Path
 
+import mlflow
 import yaml
+from dotenv import load_dotenv
 from ultralytics import YOLO
 
-from utils import save_metrics, save_model
+from utils import save_metrics_and_params, save_model
 
 
-ROOT_DIR = Path(__file__).resolve().parents[1] # root directory absolute path
-DATA_DIR = os.path.join(ROOT_DIR, "data/raw/wildfire-raw-yolov8")
-DATA_YAML = os.path.join(DATA_DIR, "data.yaml")
+load_dotenv()
+
+MLFLOW_TRACKING_URI=os.getenv('MLFLOW_TRACKING_URI')
+
+root_dir = Path(__file__).resolve().parents[1] # root directory absolute path
+data_dir = os.path.join(root_dir, "data/raw/wildfire-raw-yolov8")
+data_yaml_path = os.path.join(data_dir, "data.yaml")
+metrics_path = os.path.join(root_dir, 'reports/train_metrics.json')
 
 
 if __name__ == '__main__':
 
+ # load the configuration file 
  with open(r"params.yaml") as f:
  params = yaml.safe_load(f)
 
- # load a pre-trained model 
- pre_trained_model = YOLO(params['model_type'])
-
- # train 
- model = pre_trained_model.train(
- data=DATA_YAML,
- imgsz=params['imgsz'],
- batch=params['batch'],
- epochs=params['epochs'],
- optimizer=params['optimizer'],
- lr0=params['lr0'],
- seed=params['seed'],
- pretrained=params['pretrained'],
- name=params['name']
- )
-
- # save model
- save_model(experiment_name=params['name']) 
-
- # save metrics 
- save_metrics(experiment_name=params['name'])
+ # set the tracking uri 
+ mlflow.set_tracking_uri(MLFLOW_TRACKING_URI)
+
+ # start mlflow experiment 
+ with mlflow.start_run(run_name=params['name']):
+ # load a pre-trained model 
+ pre_trained_model = YOLO(params['model_type'])
+
+ # train 
+ model = pre_trained_model.train(
+ data=data_yaml_path,
+ imgsz=params['imgsz'],
+ batch=params['batch'],
+ epochs=params['epochs'],
+ optimizer=params['optimizer'],
+ lr0=params['lr0'],
+ seed=params['seed'],
+ pretrained=params['pretrained'],
+ name=params['name']
+ )
+
+ # log params with mlflow
+ mlflow.log_param('model_type', params['model_type'])
+ mlflow.log_param('epochs',params['epochs'])
+ mlflow.log_param('optimizer', params['optimizer'])
+ mlflow.log_param('learning_rate', params['lr0'])
+
+ # save model
+ save_model(experiment_name=params['name']) 
+
+ # save metrics csv file and training params 
+ save_metrics_and_params(experiment_name=params['name'])
+
+
+
+
+
 
 
 

src/utils.py

@@ -2,10 +2,11 @@
 import shutil
 from pathlib import Path
 
+
 ROOT_DIR = Path(__file__).resolve().parents[1] # root directory absolute path
 
 
-def save_model(experiment_name: str) -> None:
+def save_model(experiment_name: str):
  """ saves the weights of trained model to the models directory """ 
  if os.path.isdir('runs'):
  model_weights = experiment_name + "/weights/best.pt"
@@ -14,10 +15,11 @@ def save_model(experiment_name: str) -> None:
  shutil.copy(src=path_model_weights, dst=f'{ROOT_DIR}/models/model.pt')
 
 
-def save_metrics(experiment_name: str) -> None:
+def save_metrics_and_params(experiment_name: str) -> None:
  """ saves training metrics, params and confusion matrix to the reports directory """ 
  if os.path.isdir('runs'):
  path_metrics = os.path.join(ROOT_DIR, "runs/detect", experiment_name)
+
  # save experiment training metrics 
  shutil.copy(src=f'{path_metrics}/results.csv', dst=f'{ROOT_DIR}/reports/train_metrics.csv')
 
@@ -27,5 +29,3 @@ def save_metrics(experiment_name: str) -> None:
  # save training params
  shutil.copy(src=f'{path_metrics}/args.yaml', dst=f'{ROOT_DIR}/reports/train_params.yaml')
 
-
-save_model('yolov8s_exp_v0')