Spaces:
Build error
Build error
Update model_handler.py
Browse files- model_handler.py +12 -5
model_handler.py
CHANGED
|
@@ -15,7 +15,6 @@ class ModelHandler:
|
|
| 15 |
try:
|
| 16 |
print(f"Loading {self.model_name} on {self.device}...")
|
| 17 |
|
| 18 |
-
# Pemuatan otomatis oleh pipeline (sudah terbukti berhasil di langkah sebelumnya)
|
| 19 |
self.pipeline = BaseChronosPipeline.from_pretrained(
|
| 20 |
self.model_name,
|
| 21 |
device_map=self.device,
|
|
@@ -52,12 +51,20 @@ class ModelHandler:
|
|
| 52 |
predictions_samples = self.pipeline.predict(
|
| 53 |
data['original'],
|
| 54 |
prediction_length=horizon,
|
| 55 |
-
#
|
| 56 |
-
|
| 57 |
)
|
| 58 |
|
| 59 |
-
#
|
| 60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
|
| 62 |
return mean_predictions
|
| 63 |
|
|
|
|
| 15 |
try:
|
| 16 |
print(f"Loading {self.model_name} on {self.device}...")
|
| 17 |
|
|
|
|
| 18 |
self.pipeline = BaseChronosPipeline.from_pretrained(
|
| 19 |
self.model_name,
|
| 20 |
device_map=self.device,
|
|
|
|
| 51 |
predictions_samples = self.pipeline.predict(
|
| 52 |
data['original'],
|
| 53 |
prediction_length=horizon,
|
| 54 |
+
# FIX UTAMA: Menghapus 'n_samples' untuk menghindari error.
|
| 55 |
+
# Model akan kembali ke single trajectory prediction (deterministic)
|
| 56 |
)
|
| 57 |
|
| 58 |
+
# Karena sampling dihilangkan, asumsikan output adalah single trajectory (1D atau 2D dengan dimensi pertama 1)
|
| 59 |
+
if predictions_samples.ndim > 1 and predictions_samples.shape[0] > 1:
|
| 60 |
+
# Jika model tetap mengembalikan multiple samples (probabilistic)
|
| 61 |
+
mean_predictions = np.mean(predictions_samples, axis=0)
|
| 62 |
+
elif predictions_samples.ndim > 1 and predictions_samples.shape[0] == 1:
|
| 63 |
+
# Jika hanya satu trajectory
|
| 64 |
+
mean_predictions = predictions_samples[0]
|
| 65 |
+
else:
|
| 66 |
+
# Jika sudah 1D
|
| 67 |
+
mean_predictions = predictions_samples
|
| 68 |
|
| 69 |
return mean_predictions
|
| 70 |
|