Spaces:
Sleeping
Sleeping
Train model at Docker build time; stop committing binaries
Browse filesHugging Face requires git-LFS for binary artifacts. Instead, generate the
model during the image build (deterministic: pinned deps + fixed seed) and
keep models/ out of git.
- .gitignore +5 -3
- Dockerfile +4 -0
- models/metrics.json +0 -51
.gitignore
CHANGED
|
@@ -17,6 +17,8 @@ Thumbs.db
|
|
| 17 |
.vscode/
|
| 18 |
.idea/
|
| 19 |
|
| 20 |
-
#
|
| 21 |
-
#
|
| 22 |
-
#
|
|
|
|
|
|
|
|
|
| 17 |
.vscode/
|
| 18 |
.idea/
|
| 19 |
|
| 20 |
+
# Trained artifacts are generated at Docker build time (see Dockerfile), and
|
| 21 |
+
# locally with `python -m nephroscreen.train` — not committed (Hugging Face
|
| 22 |
+
# requires git-LFS for binaries, and build-time training keeps the repo clean).
|
| 23 |
+
models/*.joblib
|
| 24 |
+
models/*.json
|
Dockerfile
CHANGED
|
@@ -10,6 +10,10 @@ RUN pip install --no-cache-dir -r requirements.txt
|
|
| 10 |
COPY . .
|
| 11 |
RUN pip install --no-cache-dir -e .
|
| 12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
EXPOSE 8000
|
| 14 |
# Render/Cloud Run set $PORT; default to 8000 locally.
|
| 15 |
CMD ["sh", "-c", "uvicorn api.main:app --host 0.0.0.0 --port ${PORT:-8000}"]
|
|
|
|
| 10 |
COPY . .
|
| 11 |
RUN pip install --no-cache-dir -e .
|
| 12 |
|
| 13 |
+
# Train the model at build time (fast + deterministic: pinned deps, fixed seed).
|
| 14 |
+
# Keeps binaries out of git and guarantees the served model matches the code.
|
| 15 |
+
RUN python -m nephroscreen.train
|
| 16 |
+
|
| 17 |
EXPOSE 8000
|
| 18 |
# Render/Cloud Run set $PORT; default to 8000 locally.
|
| 19 |
CMD ["sh", "-c", "uvicorn api.main:app --host 0.0.0.0 --port ${PORT:-8000}"]
|
models/metrics.json
DELETED
|
@@ -1,51 +0,0 @@
|
|
| 1 |
-
{
|
| 2 |
-
"model": "RandomForestClassifier(n_estimators=100)",
|
| 3 |
-
"trained_at": "2026-07-03T10:37:56.937655+00:00",
|
| 4 |
-
"random_state": 42,
|
| 5 |
-
"test_size": 0.2,
|
| 6 |
-
"n_train": 320,
|
| 7 |
-
"n_test": 80,
|
| 8 |
-
"serving_threshold": 0.62,
|
| 9 |
-
"metrics": {
|
| 10 |
-
"Logistic Regression": {
|
| 11 |
-
"accuracy": 98.75,
|
| 12 |
-
"precision": 100.0,
|
| 13 |
-
"recall": 98.0,
|
| 14 |
-
"f1": 98.99,
|
| 15 |
-
"roc_auc": 1.0,
|
| 16 |
-
"cv_mean": 99.38,
|
| 17 |
-
"cv_std": 1.25
|
| 18 |
-
},
|
| 19 |
-
"Random Forest": {
|
| 20 |
-
"accuracy": 100.0,
|
| 21 |
-
"precision": 100.0,
|
| 22 |
-
"recall": 100.0,
|
| 23 |
-
"f1": 100.0,
|
| 24 |
-
"roc_auc": 1.0,
|
| 25 |
-
"cv_mean": 98.75,
|
| 26 |
-
"cv_std": 2.86
|
| 27 |
-
},
|
| 28 |
-
"Decision Tree": {
|
| 29 |
-
"accuracy": 97.5,
|
| 30 |
-
"precision": 98.0,
|
| 31 |
-
"recall": 98.0,
|
| 32 |
-
"f1": 98.0,
|
| 33 |
-
"roc_auc": 0.9893,
|
| 34 |
-
"cv_mean": 96.56,
|
| 35 |
-
"cv_std": 2.19
|
| 36 |
-
}
|
| 37 |
-
},
|
| 38 |
-
"top_features": {
|
| 39 |
-
"hemo": 0.1771,
|
| 40 |
-
"pcv": 0.1626,
|
| 41 |
-
"sc": 0.1132,
|
| 42 |
-
"kidney_stress_index": 0.1012,
|
| 43 |
-
"rbcc": 0.0995,
|
| 44 |
-
"sg": 0.0828,
|
| 45 |
-
"htn": 0.0469,
|
| 46 |
-
"al": 0.0412,
|
| 47 |
-
"dm": 0.0309,
|
| 48 |
-
"bu": 0.0306
|
| 49 |
-
},
|
| 50 |
-
"note": "Near-perfect scores reflect the high separability of this small UCI dataset (400 patients, 80-patient test set) \u2014 not clinical deployment readiness. See README for the honest interpretation."
|
| 51 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|