PVNet 5.3.23__tar.gz → 5.3.25__tar.gz

{pvnet-5.3.23 → pvnet-5.3.25}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: PVNet
-Version: 5.3.23
+Version: 5.3.25
 Summary: PVNet
 Author-email: Peter Dudfield <info@openclimatefix.org>
 Requires-Python: <3.14,>=3.11

{pvnet-5.3.23 → pvnet-5.3.25}/PVNet.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: PVNet
-Version: 5.3.23
+Version: 5.3.25
 Summary: PVNet
 Author-email: Peter Dudfield <info@openclimatefix.org>
 Requires-Python: <3.14,>=3.11

{pvnet-5.3.23 → pvnet-5.3.25}/pvnet/training/lightning_module.py

@@ -129,11 +129,15 @@ class PVNetLightningModule(pl.LightningModule):
         if self.model.use_quantile_regression:
             metric_name = "val_fraction_below/fraction_below_{:.2f}_quantile"
             # Add fraction below each quantile for calibration
+            val_quantiles = np.zeros(len(self.model.output_quantiles))
             for i, quantile in enumerate(self.model.output_quantiles):
                 below_quant = y <= y_hat[..., i]
                 # Mask values small values, which are dominated by night
                 mask = y >= 0.01
-                losses[metric_name.format(quantile)] = below_quant[mask].float().mean()
+                below_quant_masked_mean = below_quant[mask].float().mean()
+                losses[metric_name.format(quantile)] = below_quant_masked_mean
+                val_quantiles[i] = below_quant_masked_mean
+            self._val_quantiles.append(val_quantiles)
 
         return losses
 
@@ -185,6 +189,8 @@ class PVNetLightningModule(pl.LightningModule):
         # Set up stores which we will fill during validation
         self.all_val_results: list[xr.Dataset] = []
         self._val_horizon_maes: list[np.array] = []
+        if self.model.use_quantile_regression:
+            self._val_quantiles: list[np.array] = []
         if self.current_epoch == 0:
             self._val_persistence_horizon_maes: list[np.array] = []
 
@@ -355,6 +361,25 @@ class PVNetLightningModule(pl.LightningModule):
                 step=self.trainer.global_step,                
             )
 
+            # Create a quantile-quantile plot
+            if self.model.use_quantile_regression:
+                val_quantiles = np.mean(self._val_quantiles, axis=0)
+                self._val_quantiles = []
+                    
+                qq_plot = wandb_line_plot(
+                    x=self.model.output_quantiles,
+                    y=val_quantiles,
+                    xlabel="Target quantile",
+                    ylabel="Fraction below quantile",
+                    title="Quantile-quantile plot",
+                    add_identity_line=True
+                )
+
+                wandb.log(
+                    {"quantile_quantile": qq_plot},
+                    step=self.trainer.global_step,                
+                )
+
             # Create persistence horizon accuracy curve but only on first epoch
             if self.current_epoch == 0:
                 persist_horizon_mae_plot = wandb_line_plot(

{pvnet-5.3.23 → pvnet-5.3.25}/pvnet/training/plots.py

@@ -14,13 +14,31 @@ def wandb_line_plot(
     y: Sequence[float], 
     xlabel: str, 
     ylabel: str, 
-    title: str | None = None
+    title: str | None = None,
+    add_identity_line: bool = False,
 ) -> wandb.plot.CustomChart:
     """Make a wandb line plot"""
-    data = [[xi, yi] for (xi, yi) in zip(x, y)]
-    table = wandb.Table(data=data, columns=[xlabel, ylabel])
-    return wandb.plot.line(table, xlabel, ylabel, title=title)
-
+    # Main series data
+    data = [[xi, yi, "model"] for xi, yi in zip(x, y)]
+    
+    # Add identity line endpoints if requested
+    if add_identity_line:
+        min_val, max_val = min(x), max(x)
+        data.append([min_val, min_val, "y=x"])
+        data.append([max_val, max_val, "y=x"])
+
+    table = wandb.Table(data=data, columns=[xlabel, ylabel, "key"])
+
+    # stroke=None creates a clean single line; stroke="key" creates multi-line legend
+    stroke_col = "key" if add_identity_line else None
+
+    return wandb.plot.line(
+        table=table, 
+        x=xlabel, 
+        y=ylabel, 
+        stroke=stroke_col, 
+        title=title
+    )
 
 def plot_sample_forecasts(
     batch: TensorBatch,

{pvnet-5.3.23 → pvnet-5.3.25}/pvnet/training/train.py

@@ -85,31 +85,29 @@ def train(config: DictConfig) -> None:
 
     # Set the output directory based in the wandb-id of the run
     if use_wandb_logger:
-        for callback in callbacks:
-            if isinstance(callback, ModelCheckpoint):
-                # Calling the .experiment property instantiates a wandb run
-                wandb_id = wandb_logger.experiment.id
-
-                # Save the run results to the expected parent folder but with the folder name
-                # set by the wandb ID
-                save_dir = "/".join(callback.dirpath.split("/")[:-1] + [wandb_id])
-
-                callback.dirpath = save_dir
-                
-                # Save the model config
-                os.makedirs(save_dir, exist_ok=True)
-                OmegaConf.save(config.model, f"{save_dir}/{MODEL_CONFIG_NAME}")
-
-                # Save the data config to the output directory and to wandb
-                data_config = config.datamodule.configuration
-                shutil.copyfile(data_config, f"{save_dir}/{DATA_CONFIG_NAME}")
-                wandb_logger.experiment.save(f"{save_dir}/{DATA_CONFIG_NAME}", base_path=save_dir)
-
-                # Save the full hydra config to the output directory and to wandb
-                OmegaConf.save(config, f"{save_dir}/{FULL_CONFIG_NAME}")
-                wandb_logger.experiment.save(f"{save_dir}/{FULL_CONFIG_NAME}", base_path=save_dir)
-                
-                break
+        ckpt_callbacks = [cb for cb in callbacks if isinstance(cb, ModelCheckpoint)]
+        if ckpt_callbacks:
+            wandb_id = wandb_logger.experiment.id
+
+            # Save run results to the expected folder but with folder name set by ID
+            save_dir = "/".join(ckpt_callbacks[0].dirpath.split("/")[:-1] + [wandb_id])
+
+            # Point every ckpt callback (best + last) at same folder
+            for cb in ckpt_callbacks:
+                cb.dirpath = save_dir
+
+            # Save model config
+            os.makedirs(save_dir, exist_ok=True)
+            OmegaConf.save(config.model, f"{save_dir}/{MODEL_CONFIG_NAME}")
+
+            # Save data config to output directory and to wandb
+            data_config = config.datamodule.configuration
+            shutil.copyfile(data_config, f"{save_dir}/{DATA_CONFIG_NAME}")
+            wandb_logger.experiment.save(f"{save_dir}/{DATA_CONFIG_NAME}", base_path=save_dir)
+
+            # Save full hydra config to output directory and to wandb
+            OmegaConf.save(config, f"{save_dir}/{FULL_CONFIG_NAME}")
+            wandb_logger.experiment.save(f"{save_dir}/{FULL_CONFIG_NAME}", base_path=save_dir)
 
     trainer: Trainer = hydra.utils.instantiate(
         config.trainer,