WildTrain Scripts Reference

This page documents all batch scripts available in the WildTrain package for model training and evaluation.

Overview

All scripts are located in wildtrain/scripts/ directory.

Quick Reference

Script	Purpose	Config File
train_classifier.bat	Train classification model	configs/classification/classification_train.yaml
eval_classifier.bat	Evaluate classifier	configs/classification/classification_eval.yaml
eval_detector.bat	Evaluate detector	Various detector configs
train_yolo.bat	Train YOLO detector	configs/detection/yolo_configs/yolo.yaml
train_mmdet.bat	Train MMDetection model	configs/detection/mmdet_configs/mmdet.yaml
register_model.bat	Register model to MLflow	configs/registration/*.yaml
run_classification_pipeline.bat	Full classification pipeline	configs/classification/classification_pipeline_config.yaml
run_detection_pipeline.bat	Full detection pipeline	configs/detection/yolo_configs/yolo_pipeline_config.yaml
run_server.bat	Run inference server	configs/inference.yaml
visualize_predictions.bat	Visualize predictions	Various configs
create_dataset.bat	Create/prepare dataset	configs/datapreparation/*.yaml

---

Classification Scripts

train_classifier.bat

Purpose: Train an image classification model using PyTorch Lightning.

Location: wildtrain/scripts/train_classifier.bat

Command:

uv run wildtrain train classifier -c configs\classification\classification_train.yaml

Configuration: wildtrain/configs/classification/classification_train.yaml

Key Parameters:

model:
  architecture: "resnet50"  # resnet18, resnet50, efficientnet_b0, etc.
  num_classes: 10
  pretrained: true
  learning_rate: 0.001

data:
  root_data_directory: "D:/data/roi_dataset"
  split: "train"
  batch_size: 32
  num_workers: 4
  image_size: 224

training:
  max_epochs: 100
  accelerator: "gpu"  # gpu, cpu
  devices: 1
  precision: 16  # 16, 32
  gradient_clip_val: 1.0

callbacks:
  early_stopping:
    monitor: "val_loss"
    patience: 10
  model_checkpoint:
    monitor: "val_acc"
    mode: "max"
    save_top_k: 3

mlflow:
  experiment_name: "classification"
  tracking_uri: "http://localhost:5000"

Example Usage:

cd wildtrain

# Edit config
notepad configs\classification\classification_train.yaml

# Train
scripts\train_classifier.bat

Output: - Trained model checkpoints - MLflow run with metrics - Training logs - Best model weights

---

eval_classifier.bat

Purpose: Evaluate a trained classification model.

Location: wildtrain/scripts/eval_classifier.bat

Command:

uv run wildtrain eval classifier -c configs\classification\classification_eval.yaml

Configuration: wildtrain/configs/classification/classification_eval.yaml

Key Parameters:

model:
  checkpoint_path: "checkpoints/best.ckpt"
  # or load from MLflow
  mlflow_model_name: "wildlife_classifier"
  mlflow_model_version: "latest"

data:
  root_data_directory: "D:/data/roi_dataset"
  split: "test"
  batch_size: 64

evaluation:
  save_predictions: true
  generate_confusion_matrix: true
  class_metrics: true

Example Usage:

cd wildtrain
scripts\eval_classifier.bat

Output: - Evaluation metrics (accuracy, precision, recall, F1) - Confusion matrix - Per-class metrics - Predictions file (if enabled)

---

Detection Scripts

train_yolo.bat

Purpose: Train YOLO object detection model.

Location: wildtrain/scripts/train_yolo.bat (or similar)

Command:

uv run wildtrain train detector -c configs\detection\yolo_configs\yolo.yaml

Configuration: wildtrain/configs/detection/yolo_configs/yolo.yaml

Key Parameters:

model:
  framework: "yolo"
  size: "n"  # n, s, m, l, x
  pretrained: true

data:
  data_yaml: "D:/data/wildlife/data.yaml"
  # data.yaml contains:
  # - train: path to train images
  # - val: path to val images
  # - nc: number of classes
  # - names: class names

training:
  epochs: 100
  imgsz: 640
  batch: 16
  optimizer: "AdamW"
  lr0: 0.001
  device: 0  # GPU device

augmentation:
  hsv_h: 0.015
  hsv_s: 0.7
  hsv_v: 0.4
  degrees: 0.0
  translate: 0.1
  scale: 0.5
  shear: 0.0
  perspective: 0.0
  flipud: 0.0
  fliplr: 0.5
  mosaic: 1.0

mlflow:
  experiment_name: "yolo_detection"

Example Usage:

cd wildtrain
scripts\train_yolo.bat

YOLO Data Format:

dataset/
├── images/
│   ├── train/
│   └── val/
├── labels/
│   ├── train/
│   └── val/
└── data.yaml

data.yaml:

train: ./images/train
val: ./images/val
nc: 3
names: ['elephant', 'giraffe', 'zebra']

---

eval_detector.bat

Purpose: Evaluate object detection model.

Location: wildtrain/scripts/eval_detector.bat

Command:

uv run wildtrain eval detector -c configs\detection\yolo_configs\yolo_eval.yaml

Example Usage:

cd wildtrain
scripts\eval_detector.bat

Output: - mAP (mean Average Precision) - mAP@50, mAP@75 - Per-class AP - Precision-Recall curves - Detection visualizations

---

Pipeline Scripts

run_classification_pipeline.bat

Purpose: Run complete classification training pipeline.

Location: wildtrain/scripts/run_classification_pipeline.bat

Configuration: wildtrain/configs/classification/classification_pipeline_config.yaml

Pipeline Steps: 1. Data validation 2. Model initialization 3. Training 4. Evaluation 5. Model registration 6. Export for deployment

Example Config:

pipeline:
  name: "wildlife_classification_v1"

data_preparation:
  validate_data: true
  augmentation: true

training:
  config_file: "configs/classification/classification_train.yaml"

evaluation:
  config_file: "configs/classification/classification_eval.yaml"

registration:
  register_to_mlflow: true
  model_name: "wildlife_classifier"
  stage: "Staging"

export:
  export_onnx: true
  export_torchscript: true

---

run_detection_pipeline.bat

Purpose: Run complete detection training pipeline.

Location: wildtrain/scripts/run_detection_pipeline.bat

Configuration: wildtrain/configs/detection/yolo_configs/yolo_pipeline_config.yaml

Pipeline Includes: - Data preparation - Training - Validation - Hyperparameter tuning (optional) - Model registration - Export

---

Model Management Scripts

register_model.bat

Purpose: Register trained model to MLflow model registry.

Location: wildtrain/scripts/register_model.bat

Command:

uv run wildtrain register <model_type> <config_file>

Example Configs:

Classifier Registration

# configs/registration/classifier_registration_example.yaml
model_path: "checkpoints/best.ckpt"
model_name: "wildlife_classifier"
model_type: "classifier"

description: "ResNet50 classifier for wildlife ROI"

tags:
  architecture: "resnet50"
  dataset: "wildlife_roi_v1"
  num_classes: "10"
  training_date: "2024-01-15"
  accuracy: "0.95"

aliases:
  - "production"
  - "latest"

Detector Registration

# configs/registration/detector_registration_example.yaml
model_path: "runs/detect/train/weights/best.pt"
model_name: "wildlife_detector"
model_type: "detector"

description: "YOLO11n detector for aerial wildlife"

tags:
  framework: "yolo"
  version: "11n"
  dataset: "wildlife_aerial_v2"
  map50: "0.89"
  training_date: "2024-01-15"

Example Usage:

cd wildtrain
scripts\register_model.bat

---

Inference Scripts

run_server.bat

Purpose: Run inference server for model deployment.

Location: wildtrain/scripts/run_server.bat

Command:

uv run wildtrain serve -c configs\inference.yaml

Configuration: wildtrain/configs/inference.yaml

Key Parameters:

server:
  host: "0.0.0.0"
  port: 8000
  workers: 2

model:
  mlflow_model_name: "wildlife_detector"
  mlflow_model_alias: "production"
  device: "cuda"

inference:
  batch_size: 8
  confidence_threshold: 0.5
  nms_threshold: 0.45

Example Usage:

cd wildtrain
scripts\run_server.bat

API Access: - Endpoint: http://localhost:8000 - Docs: http://localhost:8000/docs

Example Request:

import requests

with open("image.jpg", "rb") as f:
    response = requests.post(
        "http://localhost:8000/predict",
        files={"file": f}
    )

predictions = response.json()

---

Utility Scripts

visualize_predictions.bat

Purpose: Visualize model predictions on images.

Location: wildtrain/scripts/visualize_predictions.bat

Configuration: Various visualization configs

Features: - Draw bounding boxes - Show confidence scores - Save annotated images - Generate prediction gallery

Example Usage:

cd wildtrain
scripts\visualize_predictions.bat

---

Common Workflows

Training Workflow (Classification)

cd wildtrain

# 1. Prepare data (if needed)
scripts\create_dataset.bat

# 2. Train model
scripts\train_classifier.bat

# 3. Evaluate
scripts\eval_classifier.bat

# 4. Register to MLflow
scripts\register_model.bat

Training Workflow (Detection)

cd wildtrain

# 1. Train YOLO
scripts\train_yolo.bat

# 2. Evaluate
scripts\eval_detector.bat

# 3. Visualize predictions
scripts\visualize_predictions.bat

# 4. Register model
scripts\register_model.bat

Hyperparameter Tuning

# Use Optuna for HPO
cd wildtrain
uv run wildtrain tune classifier -c configs/classification/classification_sweep.yaml

Model Deployment

# 1. Register model
scripts\register_model.bat

# 2. Start inference server
scripts\run_server.bat

# 3. Test API
curl -X POST "http://localhost:8000/predict" -F "file=@test.jpg"

---

Configuration Examples

Complete Training Config

# configs/classification/classification_train.yaml
model:
  architecture: "resnet50"
  num_classes: 10
  pretrained: true
  learning_rate: 0.001
  weight_decay: 0.0001
  dropout: 0.5

data:
  root_data_directory: "D:/data/roi_dataset"
  split: "train"
  batch_size: 32
  num_workers: 4
  image_size: 224
  normalize: true
  mean: [0.485, 0.456, 0.406]
  std: [0.229, 0.224, 0.225]

augmentation:
  random_flip: 0.5
  random_rotation: 15
  color_jitter:
    brightness: 0.2
    contrast: 0.2
    saturation: 0.2
  random_crop: true

training:
  max_epochs: 100
  accelerator: "gpu"
  devices: 1
  precision: 16
  accumulate_grad_batches: 1
  gradient_clip_val: 1.0
  log_every_n_steps: 10

optimizer:
  type: "Adam"
  lr: 0.001
  betas: [0.9, 0.999]

scheduler:
  type: "CosineAnnealingLR"
  T_max: 100
  eta_min: 0.00001

callbacks:
  early_stopping:
    monitor: "val_loss"
    patience: 10
    mode: "min"
  model_checkpoint:
    monitor: "val_acc"
    mode: "max"
    save_top_k: 3
    filename: "epoch{epoch:02d}-acc{val_acc:.4f}"

mlflow:
  experiment_name: "wildlife_classification"
  tracking_uri: "http://localhost:5000"
  log_models: true

---

Troubleshooting

Training Crashes

Issue: Training crashes with CUDA out of memory

Solutions: 1. Reduce batch size 2. Use mixed precision (precision: 16) 3. Reduce image size 4. Enable gradient accumulation

MLflow Connection Error

Issue: Can't connect to MLflow tracking server

Solutions: 1. Start MLflow server: mlflow server --port 5000 2. Check MLFLOW_TRACKING_URI environment variable 3. Verify network connection 4. Check firewall settings

Data Loading Slow

Issue: Data loading is bottleneck

Solutions: 1. Increase num_workers (max 4 on Windows) 2. Enable pin_memory: true 3. Use SSD for data storage 4. Reduce data preprocessing complexity

Model Won't Load

Issue: Can't load trained model

Solutions: 1. Check checkpoint path is correct 2. Verify model architecture matches 3. Check for version compatibility 4. Try loading with strict=False

---

Next Steps

• WildTrain Configuration Reference
• Model Training Tutorial
• WildTrain CLI Reference
• MLflow Integration Guide