add combine method

quartz_solar_forecast/eval/forecast.py

@@ -62,6 +62,7 @@ def run_forecast(pv_df: pd.DataFrame, nwp_df: pd.DataFrame, nwp_source="ICON") -
             ts = datetime.fromisoformat(ts)
 
         # make pv and nwp data from GFS
+        # TODO move this to model
         nwp_xr = xr.DataArray(
             data=nwp_site_df.values,
             dims=["step", "variable"],
@@ -97,6 +98,10 @@ def run_forecast(pv_df: pd.DataFrame, nwp_df: pd.DataFrame, nwp_source="ICON") -
         times = pd.date_range(start=x.ts, periods=len(pred.powers), freq="15min")
         pred_df = pd.DataFrame({"power_wh": pred.powers}, index=times)
 
+        # only select hourly predictions
+        pred_df = pred_df.resample("1H").mean()
+        pred_df["horizon_hours"] = range(0, len(pred_df))
+
         all_predictions.append(pred_df)
 
     return pd.concat(all_predictions)

quartz_solar_forecast/eval/utils.py

@@ -0,0 +1,36 @@
+import pandas as pd
+
+
+
+
+
+def combine_forecast_ground_truth(forecast_df, ground_truth_df):
+    """
+    Combine the forecast results with the ground truth (ts, id, horizon (in hours), pred, truth, diff)
+
+
+    forecast_df should have the following columns
+    - timestamp
+    - pv_id
+    - horizon_hours
+    - power
+
+    ground_truth_df should have the following columns
+    - timestamp
+    - pv_id
+    - horizon_hours
+    - power
+
+    """
+
+    # rename power to forecast_power
+    forecast_df = forecast_df.rename(columns={"power": "forecast_power"})
+
+    # rename power to ground_truth_power
+    ground_truth_df = ground_truth_df.rename(columns={"power": "ground_truth_power"})
+
+    # merge the two dataframes
+    combined_df = pd.merge(forecast_df, ground_truth_df, on=["timestamp", "pv_id", "horizon_hours"])
+
+    return combined_df
+

quartz_solar_forecast/evaluation.py

@@ -7,6 +7,7 @@
 """
 from quartz_solar_forecast.eval.nwp import get_nwp
 from quartz_solar_forecast.eval.forecast import run_forecast
+from quartz_solar_forecast.eval.utils import combine_forecast_ground_truth
 
 import pandas as pd
 
@@ -18,6 +19,7 @@ def run_eval(testset_path):
     # Extract generation data and metadata for specific sites and timestamps for the testset from Hugging Face. (Zak)
 
     # Split data into PV inputs and ground truth. (Zak)
+    ground_truth_df = None # TODO
 
     # Collect NWP data from Hugging Face, ICON. (Peter)
     nwp_df = get_nwp(testset)
@@ -27,8 +29,10 @@ def run_eval(testset_path):
     predictions_df = run_forecast(pv_df=None, nwp_df=nwp_df)
 
     # Combine the forecast results with the ground truth (ts, id, horizon (in hours), pred, truth, diff)
+    results_df = combine_forecast_ground_truth(predictions_df, ground_truth_df)
 
     # Save file
+    results_df.to_csv("results.csv")
 
     # Calculate and print metrics: MAE