{
  "name": "307_dropout02_with_doy",
  "description": "Production Model 307: LSTM-based harvest detection (Phase 3, minimal regularization)",
  "model_info": {
    "type": "LSTM",
    "architecture": "Unidirectional LSTM with dual output heads (imminent + detected)",
    "total_parameters": 105120,
    "input_features": 14,
    "hidden_units": 256,
    "output_heads": 2,
    "training_data": "Historical multi-season CI data from multiple estates",
    "validation_method": "5-fold cross-validation",
    "device": "GPU (CUDA) or CPU fallback"
  },
  "production_scripts": {
    "baseline": {
      "script": "01_harvest_baseline_prediction.py",
      "frequency": "Run ONCE during setup",
      "purpose": "Predict all harvest dates (ground truth baseline)",
      "input": "ci_data_for_python.csv (complete historical data)",
      "output": "harvest_production_export.xlsx",
      "time_estimate": "5-30 minutes depending on data volume"
    },
    "monitoring": {
      "script": "02_harvest_imminent_weekly.py",
      "frequency": "Run WEEKLY (or daily if required)",
      "purpose": "Real-time harvest status and imminent alerts",
      "input": "ci_data_for_python.csv (recent data)",
      "output": "harvest_imminent_weekly.csv",
      "time_estimate": "1-5 minutes"
    }
  },
  "features": [
    "CI_raw",
    "7d_MA",
    "14d_MA",
    "21d_MA",
    "7d_velocity",
    "14d_velocity",
    "21d_velocity",
    "7d_min",
    "14d_min",
    "21d_min",
    "7d_std",
    "14d_std",
    "21d_std",
    "DOY_normalized"
  ],
  "model": {
    "type": "LSTM",
    "hidden_size": 256,
    "num_layers": 1,
    "dropout": 0.2
  },
  "training": {
    "imminent_days_before": 28,
    "imminent_days_before_end": 1,
    "detected_days_after_start": 1,
    "detected_days_after_end": 21,
    "k_folds": 5,
    "num_epochs": 150,
    "patience": 20,
    "learning_rate": 0.001,
    "batch_size": 4
  },
  "data": {
    "csv_path": "../lstm_complete_data.csv",
    "ci_column": "FitData",
    "test_fraction": 0.15,
    "seed": 42
  },
  "workflow_instructions": {
    "overview": "Model 307 uses a two-script approach: baseline setup + weekly monitoring",
    "step_1_baseline": {
      "description": "Establish historical harvest date reference for all fields",
      "script": "01_harvest_baseline_prediction.py",
      "when": "Run once after setting up CI extraction pipeline",
      "command": "conda activate python_gpu && python 01_harvest_baseline_prediction.py",
      "input_data": "ci_data_for_python.csv (all available historical CI data)",
      "output_file": "harvest_production_export.xlsx (ground truth baseline)",
      "columns": [
        "field - Field ID",
        "sub_field - Sub-field designation",
        "season - Season number (1, 2, 3...)",
        "year - Year of harvest",
        "season_start_date - Start of growing season",
        "season_end_date - End of season (harvest date)",
        "phase1_harvest_date - Refined harvest prediction"
      ],
      "notes": "This becomes your reference - compare all weekly monitoring against this"
    },
    "step_2_monitoring": {
      "description": "Weekly real-time harvest status and imminent alerts",
      "script": "02_harvest_imminent_weekly.py",
      "when": "Run every week (e.g., Mondays) or daily if near harvest",
      "command": "conda activate python_gpu && python 02_harvest_imminent_weekly.py",
      "input_data": "ci_data_for_python.csv (latest CI data from 02b conversion)",
      "output_file": "harvest_imminent_weekly.csv",
      "columns": [
        "field - Field ID",
        "sub_field - Sub-field designation",
        "imminent_prob - Likelihood of harvest readiness in next 28 days (0.0-1.0)",
        "detected_prob - Current harvest probability (0.0-1.0)",
        "week - ISO week number",
        "year - Year",
        "as_of_date - Latest date in dataset",
        "num_days - Days of history used"
      ],
      "alert_thresholds": {
        "imminent_high": "imminent_prob > 0.7 (prepare harvest)",
        "imminent_medium": "imminent_prob 0.5-0.7 (monitor closely)",
        "detected_high": "detected_prob > 0.6 (active harvesting)"
      }
    },
    "integration_with_r_pipeline": {
      "before_model_307": [
        "Planet 8-band download: download_8band_pu_optimized.ipynb",
        "CI extraction: 02_ci_extraction.R",
        "Convert to CSV: 02b_convert_rds_to_csv.R (outputs ci_data_for_python.csv)"
      ],
      "model_307_here": [
        "BASELINE: 01_harvest_baseline_prediction.py (run once)",
        "MONITORING: 02_harvest_imminent_weekly.py (run weekly)"
      ],
      "after_model_307": [
        "Field analysis: 09b_field_analysis_weekly.R (reads harvest predictions)",
        "Reports: 10_CI_report_with_kpis.Rmd (includes harvest status)"
      ]
    },
    "environment_requirements": {
      "python_env": "python_gpu",
      "activation": "conda activate python_gpu",
      "required_packages": [
        "torch (GPU-enabled)",
        "pandas",
        "numpy",
        "scikit-learn",
        "pyyaml",
        "openpyxl"
      ],
      "gpu": "NVIDIA GPU with CUDA (optional - falls back to CPU if unavailable)"
    }
  }
}