Workflow for daily keep alive

.github/workflows/daily-keepalive.yml

@@ -0,0 +1,44 @@
+name: Daily Keep-Alive
+
+on:
+  schedule:
+    - cron: "0 6 * * *"
+  workflow_dispatch:
+
+permissions:
+  contents: write
+
+jobs:
+  keepalive:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+
+      - name: Update keep_alive.csv
+        run: |
+          python3 -c "
+          import datetime, csv
+          now = datetime.datetime.utcnow()
+          epoch = datetime.date(2022, 5, 26)
+          days = (now.date() - epoch).days
+          with open('keep_alive.csv', 'w', newline='') as f:
+              w = csv.writer(f)
+              w.writerow(['date', 'days_since_first_run', 'last_updated_utc'])
+              w.writerow([now.strftime('%Y-%m-%d'), days, now.strftime('%Y-%m-%dT%H:%M:%SZ')])
+          "
+
+      - name: Commit and push
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        run: |
+          git config user.name "github-actions[bot]"
+          git config user.email "github-actions[bot]@users.noreply.github.com"
+          git add keep_alive.csv
+          git diff --cached --quiet && echo "No changes" && exit 0
+          git commit -m "chore: daily keep-alive $(date -u +%Y-%m-%d)"
+          git push origin main
+          git push https://fmegahed:$HF_TOKEN@huggingface.co/spaces/fmegahed/tavr_project main

.github/workflows/sync-to-hf.yml

@@ -0,0 +1,22 @@
+name: Sync to Hugging Face Spaces
+
+on:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+jobs:
+  sync:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+
+      - name: Push to Hugging Face
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        run: |
+          git push https://fmegahed:$HF_TOKEN@huggingface.co/spaces/fmegahed/tavr_project main

.gitignore

@@ -0,0 +1,2 @@
+.Claude.md
+.claude/

CLAUDE.md

@@ -0,0 +1,53 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+Medical prediction web app for TAVR (Transcatheter Aortic Valve Replacement) in-hospital mortality risk. Uses a pre-trained PyCaret logistic regression model (L2 penalty) trained on HCUP NIS data (2012-2019). Published in *Scientific Reports*: Alhwiti, T., Aldrugh, S., & Megahed, F. M. (2023).
+
+**Primary repo:** GitHub. Auto-syncs to HF Spaces on every push to `main`.
+**Deployed at:** `https://huggingface.co/spaces/fmegahed/tavr_project` (Docker-based Space)
+
+## Running the App
+
+```bash
+# Local
+python app.py
+
+# Docker
+docker build -t tavr-app .
+docker run -p 7860:7860 tavr-app
+```
+
+The app launches on port 7860 (configurable via `PORT` or `GRADIO_SERVER_PORT` env vars).
+
+## Architecture
+
+Single-file application (`app.py`) with three layers:
+
+1. **Model loading** (`_ensure_model_file()` / `_get_model()`): Lazy-loads `final_model.pkl` with global caching. Falls back to downloading from GitHub if the local file is missing.
+
+2. **Prediction** (`predict()`): Takes 44 patient parameters (demographics, hospital info, 32 comorbidities), constructs a DataFrame with proper categorical types (including ordered categoricals for `zipinc_qrtl`), runs PyCaret's `predict_model()` with `raw_score=True`. Returns `{"Death": 0-1, "Survival": 0-1}` dict for `gr.Label`.
+
+3. **Gradio UI** (`gr.Blocks` with `gr.themes.Soft()`): 44 inputs organized in 4 tabs (Patient Demographics, Hospital Information, Comorbidities, Procedure). Explicit "Predict" button triggers prediction; output displayed as `gr.Label` with confidence bars.
+
+## GitHub Actions Workflows
+
+- **`sync-to-hf.yml`**: Pushes to HF Spaces on every `main` push. Requires `HF_TOKEN` secret (write-access).
+- **`daily-keepalive.yml`**: Runs daily at 06:00 UTC. Updates `keep_alive.csv` with current date/timestamp and pushes to both GitHub and HF to prevent Space from sleeping.
+
+## Key Constraints
+
+- **Python 3.8** — pinned in the Dockerfile, all dependencies must be compatible
+- **PyCaret 2.3.6** — requires `SKLEARN_ALLOW_DEPRECATED_SKLEARN_PACKAGE_INSTALL=True` env var for installation
+- **All dependency versions are pinned** in `requirements.txt` — changing one may break others due to tight PyCaret/scikit-learn compatibility
+- **`final_model.pkl`** is tracked via Git LFS (see `.gitattributes`)
+- The model expects exact categorical column names and types; any change to input names in `predict()` will break predictions
+- Gradio version is 3.50.2 (not 4.x) — API differs significantly from current Gradio
+
+## GitHub Setup
+
+1. Create a GitHub repo and add it as the `origin` remote
+2. Add `HF_TOKEN` secret (Settings > Secrets > Actions) with a write-access token from https://huggingface.co/settings/tokens
+3. Push to `main` — the sync workflow will mirror to HF Spaces automatically

README.md

@@ -8,3 +8,14 @@ app_port: 7860
 pinned: false
 license: cc-by-4.0
 ---
+
+# Predicting In-Hospital Mortality After TAVR
+
+A web app for predicting in-hospital mortality after Transcatheter Aortic Valve Replacement (TAVR) using a penalized logistic regression model trained on HCUP NIS data (2012-2019).
+
+**Published paper:** [Alhwiti, T., Aldrugh, S., & Megahed, F. M. (2023), *Scientific Reports*](https://www.nature.com/articles/s41598-023-37358-9.pdf)
+
+| | Link |
+|---|---|
+| GitHub | https://github.com/fmegahed/tavr_project |
+| Hugging Face Space | https://huggingface.co/spaces/fmegahed/tavr_project |

app.py

@@ -9,7 +9,6 @@ from pandas.api.types import CategoricalDtype
 from pycaret.classification import load_model, predict_model
 
 # Optional: load example data (not required for predictions, but kept since it exists in your repo)
-# If the file is missing, the app still runs.
 try:
     _ex_data = pd.read_csv("example_data2.csv")
 except Exception:
@@ -159,132 +158,179 @@ def predict(
     model = _get_model()
     pred = predict_model(model, df, raw_score=True)
 
-    # These column names depend on how the model was trained and saved
-    # This matches your original code: Score_Yes for death, Score_No for survival.
+    # Return dict with 0-1 scale for gr.Label confidence bars
     return {
-        "Death %": round(100 * float(pred["Score_Yes"].iloc[0]), 2),
-        "Survival %": round(100 * float(pred["Score_No"].iloc[0]), 2),
-        "Predicting Death Outcome": str(pred["Label"].iloc[0]),
+        "Death": float(pred["Score_Yes"].iloc[0]),
+        "Survival": float(pred["Score_No"].iloc[0]),
     }
 
 
-inputs = [
-    gr.Slider(minimum=18, maximum=100, value=80, label="Age"),
-    gr.Dropdown(choices=["Female", "Male"], value="Female", label="Sex"),
-    gr.Dropdown(
-        choices=[
-            "Asian or Pacific Islander",
-            "Black",
-            "Hispanic",
-            "Native American",
-            "White",
-            "Other",
-        ],
-        value="White",
-        label="Race",
-    ),
-    gr.Radio(choices=["Elective", "NonElective"], value="Elective", label="Elective"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Weekend"),
-    gr.Radio(
-        choices=["FirstQ", "SecondQ", "ThirdQ", "FourthQ"],
-        value="SecondQ",
-        label="Zip Income Quartile",
-    ),
-    gr.Radio(
-        choices=["Midwest", "Northeast", "South", "West"],
-        value="South",
-        label="Hospital Region",
-    ),
-    gr.Radio(
-        choices=[
-            "New England",
-            "Middle Atlantic",
-            "East North Central",
-            "West North Central",
-            "South Atlantic",
-            "East South Central",
-            "West South Central",
-            "Mountain",
-            "Pacific",
-        ],
-        value="South Atlantic",
-        label="Hospital Division",
-    ),
-    gr.Radio(
-        choices=["Urban teaching", "Urban nonteaching", "Rural"],
-        value="Urban teaching",
-        label="Hospital Location/Teaching",
-    ),
-    gr.Radio(choices=["Small", "Medium", "Large"], value="Large", label="Hospital Bedsize"),
-    gr.Radio(
-        choices=["Government_nonfederal", "Private_invest_own", "Private_not_profit"],
-        value="Private_not_profit",
-        label="Hospital Control",
-    ),
-    gr.Dropdown(
-        choices=["Private insurance", "Medicare", "Medicaid", "Self-pay", "No charge", "Other"],
-        value="Medicare",
-        label="Payee",
-    ),
-    # Comorbidities
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Anemia"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Atrial Fibrillation"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Cancer"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Cardiac Arrhythmias"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Carotid Artery Disease"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Chronic Kidney Disease"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Chronic Pulmonary Disease"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Coagulopathy"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Depression"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Diabetes Mellitus"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Drug Abuse"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Dyslipidemia"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Endocarditis"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Family History"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Fluid and Electrolyte Disorder"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Heart Failure"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Hypertension"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Known CAD"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Liver Disease"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Obesity"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Peripheral Vascular Disease"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior CABG"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior ICD"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior MI"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior PCI"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior PPM"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior TIA Stroke"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Pulmonary Circulation Disorder"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Smoker"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Valvular Disease"),
-    gr.Radio(choices=["No", "Yes"], value="No", label="Weight Loss"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Endovascular TAVR"),
-    gr.Radio(choices=["No", "Yes"], value="Yes", label="Transapical TAVR"),
-]
+# ---------------------------------------------------------------------------
+# UI — gr.Blocks with tabs
+# ---------------------------------------------------------------------------
 
-description_html = """
-<p style="font-size:16px; line-height:1.6;">
-This app predicts in-hospital mortality after TAVR using a finalized logistic regression model with L2 penalty,
-based on national inpatient data from 2012–2019 (HCUP NIS).<br>
-<br>
-Published paper:
-<a href="https://www.nature.com/articles/s41598-023-37358-9.pdf" target="_blank">
-Alhwiti, T., Aldrugh, S., & Megahed, F. M. (2023), <i>Scientific Reports</i>
-</a>
-</p>
-"""
+with gr.Blocks(theme=gr.themes.Soft(), title="TAVR Mortality Prediction") as demo:
+
+    gr.Markdown(
+        """
+        # Predicting In-Hospital Mortality After TAVR
+
+        This app predicts in-hospital mortality after Transcatheter Aortic Valve
+        Replacement (TAVR) using a logistic regression model (L2 penalty) trained
+        on national inpatient data from 2012-2019 (HCUP NIS).
+
+        **Paper:**
+        [Alhwiti, T., Aldrugh, S., & Megahed, F. M. (2023), *Scientific Reports*](https://www.nature.com/articles/s41598-023-37358-9.pdf)
+        """
+    )
+
+    with gr.Row():
+        # ---- Left: inputs ----
+        with gr.Column(scale=3):
+            with gr.Tab("Patient Demographics"):
+                with gr.Row():
+                    age = gr.Slider(minimum=18, maximum=100, value=80, label="Age")
+                    female = gr.Dropdown(choices=["Female", "Male"], value="Female", label="Sex")
+                with gr.Row():
+                    race = gr.Dropdown(
+                        choices=[
+                            "Asian or Pacific Islander",
+                            "Black",
+                            "Hispanic",
+                            "Native American",
+                            "White",
+                            "Other",
+                        ],
+                        value="White",
+                        label="Race",
+                    )
+                    pay = gr.Dropdown(
+                        choices=["Private insurance", "Medicare", "Medicaid", "Self-pay", "No charge", "Other"],
+                        value="Medicare",
+                        label="Payee",
+                    )
+                with gr.Row():
+                    elective = gr.Radio(choices=["Elective", "NonElective"], value="Elective", label="Elective")
+                    aweekend = gr.Radio(choices=["No", "Yes"], value="No", label="Weekend Admission")
+                with gr.Row():
+                    zipinc_qrtl = gr.Radio(
+                        choices=["FirstQ", "SecondQ", "ThirdQ", "FourthQ"],
+                        value="SecondQ",
+                        label="Zip Income Quartile",
+                    )
+
+            with gr.Tab("Hospital Information"):
+                with gr.Row():
+                    hosp_region = gr.Radio(
+                        choices=["Midwest", "Northeast", "South", "West"],
+                        value="South",
+                        label="Hospital Region",
+                    )
+                    hosp_bedsize = gr.Radio(
+                        choices=["Small", "Medium", "Large"],
+                        value="Large",
+                        label="Hospital Bedsize",
+                    )
+                with gr.Row():
+                    hosp_division = gr.Radio(
+                        choices=[
+                            "New England",
+                            "Middle Atlantic",
+                            "East North Central",
+                            "West North Central",
+                            "South Atlantic",
+                            "East South Central",
+                            "West South Central",
+                            "Mountain",
+                            "Pacific",
+                        ],
+                        value="South Atlantic",
+                        label="Hospital Division",
+                    )
+                with gr.Row():
+                    hosp_locteach = gr.Radio(
+                        choices=["Urban teaching", "Urban nonteaching", "Rural"],
+                        value="Urban teaching",
+                        label="Hospital Location/Teaching",
+                    )
+                    h_contrl = gr.Radio(
+                        choices=["Government_nonfederal", "Private_invest_own", "Private_not_profit"],
+                        value="Private_not_profit",
+                        label="Hospital Control",
+                    )
+
+            with gr.Tab("Comorbidities"):
+                with gr.Row():
+                    anemia = gr.Radio(choices=["No", "Yes"], value="Yes", label="Anemia")
+                    atrial_fibrillation = gr.Radio(choices=["No", "Yes"], value="Yes", label="Atrial Fibrillation")
+                    cancer = gr.Radio(choices=["No", "Yes"], value="No", label="Cancer")
+                with gr.Row():
+                    cardiac_arrhythmias = gr.Radio(choices=["No", "Yes"], value="Yes", label="Cardiac Arrhythmias")
+                    carotid_artery_disease = gr.Radio(choices=["No", "Yes"], value="No", label="Carotid Artery Disease")
+                    chronic_kidney_disease = gr.Radio(choices=["No", "Yes"], value="Yes", label="Chronic Kidney Disease")
+                with gr.Row():
+                    chronic_pulmonary_disease = gr.Radio(choices=["No", "Yes"], value="Yes", label="Chronic Pulmonary Disease")
+                    coagulopathy = gr.Radio(choices=["No", "Yes"], value="No", label="Coagulopathy")
+                    depression = gr.Radio(choices=["No", "Yes"], value="No", label="Depression")
+                with gr.Row():
+                    diabetes_mellitus = gr.Radio(choices=["No", "Yes"], value="Yes", label="Diabetes Mellitus")
+                    drug_abuse = gr.Radio(choices=["No", "Yes"], value="No", label="Drug Abuse")
+                    dyslipidemia = gr.Radio(choices=["No", "Yes"], value="Yes", label="Dyslipidemia")
+                with gr.Row():
+                    endocarditis = gr.Radio(choices=["No", "Yes"], value="No", label="Endocarditis")
+                    family_history = gr.Radio(choices=["No", "Yes"], value="No", label="Family History")
+                    fluid_and_electrolyte_disorder = gr.Radio(choices=["No", "Yes"], value="Yes", label="Fluid & Electrolyte Disorder")
+                with gr.Row():
+                    heart_failure = gr.Radio(choices=["No", "Yes"], value="Yes", label="Heart Failure")
+                    hypertension = gr.Radio(choices=["No", "Yes"], value="Yes", label="Hypertension")
+                    known_cad = gr.Radio(choices=["No", "Yes"], value="Yes", label="Known CAD")
+                with gr.Row():
+                    liver_disease = gr.Radio(choices=["No", "Yes"], value="No", label="Liver Disease")
+                    obesity = gr.Radio(choices=["No", "Yes"], value="Yes", label="Obesity")
+                    peripheral_vascular_disease = gr.Radio(choices=["No", "Yes"], value="Yes", label="Peripheral Vascular Disease")
+                with gr.Row():
+                    prior_cabg = gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior CABG")
+                    prior_icd = gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior ICD")
+                    prior_mi = gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior MI")
+                with gr.Row():
+                    prior_pci = gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior PCI")
+                    prior_ppm = gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior PPM")
+                    prior_tia_stroke = gr.Radio(choices=["No", "Yes"], value="Yes", label="Prior TIA/Stroke")
+                with gr.Row():
+                    pulmonary_circulation_disorder = gr.Radio(choices=["No", "Yes"], value="No", label="Pulmonary Circulation Disorder")
+                    smoker = gr.Radio(choices=["No", "Yes"], value="No", label="Smoker")
+                    valvular_disease = gr.Radio(choices=["No", "Yes"], value="Yes", label="Valvular Disease")
+                with gr.Row():
+                    weight_loss = gr.Radio(choices=["No", "Yes"], value="No", label="Weight Loss")
+
+            with gr.Tab("Procedure"):
+                with gr.Row():
+                    endovascular_tavr = gr.Radio(choices=["No", "Yes"], value="Yes", label="Endovascular TAVR")
+                    transapical_tavr = gr.Radio(choices=["No", "Yes"], value="Yes", label="Transapical TAVR")
+
+        # ---- Right: output ----
+        with gr.Column(scale=1):
+            predict_btn = gr.Button("Predict", variant="primary")
+            output = gr.Label(label="Prediction", num_top_classes=2)
+
+    # Wire up the button
+    all_inputs = [
+        age, female, race, elective, aweekend, zipinc_qrtl,
+        hosp_region, hosp_division, hosp_locteach, hosp_bedsize, h_contrl, pay,
+        anemia, atrial_fibrillation, cancer, cardiac_arrhythmias,
+        carotid_artery_disease, chronic_kidney_disease, chronic_pulmonary_disease,
+        coagulopathy, depression, diabetes_mellitus, drug_abuse, dyslipidemia,
+        endocarditis, family_history, fluid_and_electrolyte_disorder,
+        heart_failure, hypertension, known_cad, liver_disease, obesity,
+        peripheral_vascular_disease, prior_cabg, prior_icd, prior_mi,
+        prior_pci, prior_ppm, prior_tia_stroke, pulmonary_circulation_disorder,
+        smoker, valvular_disease, weight_loss,
+        endovascular_tavr, transapical_tavr,
+    ]
+
+    predict_btn.click(fn=predict, inputs=all_inputs, outputs=output)
 
-iface = gr.Interface(
-    fn=predict,
-    inputs=inputs,
-    outputs="text",
-    live=True,
-    title="Predicting In-Hospital Mortality After TAVR Using Preoperative Variables and Penalized Logistic Regression",
-    description=description_html,
-    css="https://bootswatch.com/5/journal/bootstrap.css",
-)
 
 if __name__ == "__main__":
-    # Works locally, in Docker, and on HF Spaces (PORT is commonly set by platforms)
     port = int(os.getenv("PORT", os.getenv("GRADIO_SERVER_PORT", "7860")))
-    iface.launch(server_name="0.0.0.0", server_port=port)
+    demo.launch(server_name="0.0.0.0", server_port=port)

keep_alive.csv

@@ -0,0 +1,2 @@
+date,days_since_first_run,last_updated_utc
+2026-02-10,1356,2026-02-10T00:00:00Z