Upload 5 files

README.md

@@ -4,58 +4,63 @@ emoji: 🌱
 colorFrom: green
 colorTo: blue
 sdk: gradio
-sdk_version: 6.13.0
+sdk_version: 5.46.1
 app_file: app.py
 pinned: false
 license: cc-by-nc-nd-4.0
+python_version: "3.11"
 ---
 
-# 🌱 Algae Yield Predictor
+# Algae Yield Predictor
 
-This Space provides an interactive interface to **predict algal biomass, lipid, protein, and carbohydrate yields** under different culture conditions.  
-It uses augmented datasets (200k synthetic rows per target) combined with ensemble ML models (CatBoost, XGBoost, LightGBM, ExtraTrees) and a meta-stacking approach.
+This Space provides an interactive interface to predict algal biomass, lipid, protein, and carbohydrate yields under different culture conditions.
 
-[Full description continues here...]
-# 🌱 Algae Yield Predictor
+The app uses real experimental data, augmented target-specific datasets, and saved ensemble models to provide model predictions, local uncertainty intervals, response plots, species-medium constraints, and optional DOI-based literature matches.
 
-This Space provides an interactive interface to **predict algal biomass, lipid, protein, and carbohydrate yields** under different culture conditions.  
-It uses augmented datasets (200k synthetic rows per target) combined with ensemble ML models (CatBoost, XGBoost, LightGBM, ExtraTrees) and a meta-stacking approach.
+## Main features
 
----
+- Targets: biomass, lipid, protein, carbohydrate
+- Species-medium aware dropdowns
+- Curated suggested conditions by species and target
+- Selectable model: STACK, XGB, LGBM, CAT, MLP
+- Local uncertainty intervals from augmented-data nearest neighbors
+- Response plots for light, days, pH, temperature, day exposure, and night exposure
+- Optional DOI matching from `doi.csv`
+- Literature-range clamping for biologically plausible outputs
 
-## ✨ Features
-- **Targets:** biomass, lipid, protein, carbohydrate  
-- **Species–Media aware:** dropdowns restrict valid species–medium combinations  
-- **Curated suggestions:** shows recommended conditions for each species/target  
-- **Uncertainty estimates:** KNN-based local intervals (10–90%) from augmented data  
-- **Response plots:** sweep one variable (light, days, pH, etc.) and visualize prediction curve + uncertainty band  
-- **DOI references:** retrieves closest experimental setups from `doi.csv` (if provided)
+## Required repository files
 
----
+The following files/folders must be present in the Space repository root:
 
-## 🚀 How to Use
-1. Select a **target** (biomass, lipid, protein, carb).  
-2. Choose **species** and valid **growth medium**.  
-3. Adjust culture conditions:
-   - Light intensity
-   - Day/Night exposure
-   - Temperature
-   - pH
-   - Days of culture  
-4. Click **Predict + Plot** to:
-   - Get yield prediction with uncertainty band  
-   - See response curve for chosen variable  
-5. Optionally click **Find Closest DOI Matches** to explore related literature.
+```text
+app.py
+README.md
+requirements.txt
+pre-requirements.txt
+runtime.txt
+ai_al.csv
+doi.csv                         # optional
+augmented_biomass_200k.csv       # or augmented_biomass_20k.csv
+augmented_lipid_200k.csv         # or augmented_lipid_20k.csv
+augmented_protein_200k.csv       # or augmented_protein_20k.csv
+augmented_carb_200k.csv          # or augmented_carb_20k.csv
+models/
+  biomass/
+  lipid/
+  protein/
+  carb/
+```
 
----
+Each target model folder should contain the saved model artifacts expected by `app.py`, including encoders, imputer, config, base learners, MLP model, and meta model.
 
-## 🧩 Models & Data
-- **Training data:** real experimental CSV (`ai_al.csv`) + augmented synthetic sets (20k/200k).  
-- **Models:** CatBoost, XGBoost, LightGBM, ExtraTrees → stacked with RidgeCV.  
-- **Uncertainty:** derived from nearest neighbors in augmented dataset.  
+## Deployment notes
 
-If `doi.csv` is provided with experimental metadata + DOI links, the app will display closest literature matches.
+This Space is pinned to Python 3.11 because the scientific stack used here is much more reliable on Python 3.11 than Python 3.13. Python 3.13 may force source builds for packages such as pandas or create compatibility issues with TensorFlow and compiled ML libraries.
 
----
+After uploading these fixed files, use **Factory reboot** in Hugging Face Spaces to clear the old build cache.
+
+## Citation
+
+If you use this predictor or dataset, please cite:
 
-## 📂 File Structure
+Tiwari, A., Dubey, S., Sumathi, Y., Patel, A. K., & Kuo, T.-R. (2025). *Augmented and Real Microalgae Datasets for Biomass and Biochemical Composition Prediction* [Data set]. Zenodo. https://doi.org/10.5281/zenodo.17177597