Add visualization graphs to eval

quartz_solar_forecast/eval/visualize.py

@@ -0,0 +1,71 @@
+import os
+import pandas as pd
+import matplotlib.pyplot as plt
+import seaborn as sns
+import matplotlib.dates as mdates
+import re
+
+def visualize_results(results_df):
+    '''
+    Visualize Eval Results
+
+    results_df dataframe with the following columns
+    - timestamp
+    - pv_id
+    - horizon_hours
+    - forecast_power
+    - generation_power
+    '''
+
+    # Ensure results folder exists
+    if not os.path.exists('results'):
+        os.makedirs('results')
+
+    # Prediction vs Actual values for each PV ID
+    pv_ids = results_df['pv_id'].unique()
+    for pv_id in pv_ids:
+        df_subset = results_df[results_df['pv_id'] == pv_id]
+        num_plots = len(df_subset) // 48  # Find out how many full 48-hour plots we can make
+
+        for i in range(num_plots):
+            # Extract data for the ith 48-hour segment
+            df_segment = df_subset.iloc[i*48:(i+1)*48].copy()
+
+            start_timestamp = df_segment.iloc[0]['timestamp']
+            start_timestamp_str = re.sub(r"[\s\-:]", "_", start_timestamp)
+            df_segment['timestamp'] = pd.to_datetime(df_segment['timestamp'])
+
+            plt.figure(figsize=(10, 6))
+            sns.lineplot(data=df_segment, x='timestamp', y='generation_power', label='Actual', marker='o')
+            sns.lineplot(data=df_segment, x='timestamp', y='forecast_power', label='Predicted', marker='o')
+
+            # Setting the x-ticks and labels
+            ax = plt.gca()  # Get the current Axes instance
+            x_ticks = pd.date_range(start=df_segment['timestamp'].iloc[0], periods=len(df_segment), freq='h')
+            ax.set_xticks(x_ticks)  # Set x-ticks to every hour
+
+            # Format the x-tick labels to show every 5 hours
+            ax.xaxis.set_major_locator(mdates.HourLocator(interval=5))  # Show label every 5 hours
+            ax.xaxis.set_major_formatter(mdates.DateFormatter('%H:%M'))  # Format the datetime
+            ax.xaxis.set_minor_locator(mdates.HourLocator())  # Keep a tick for every hour
+
+            # Plot title, labels, legend, and tight layout
+            plt.title(f'48HRs Predicted vs. Actual Solar Power Output for {pv_id} (Starting from {start_timestamp})')
+            plt.xlabel('Timestamp')
+            plt.ylabel('Power Output (kW)')
+            plt.legend()
+            plt.tight_layout()
+
+            # Save the figure
+            plt.savefig(f'results/pred_vs_actual_{pv_id}_{start_timestamp_str}.png')
+            plt.close()
+
+    # Distribution of Errors across all data points
+    results_df['error'] = results_df['forecast_power'] - results_df['generation_power']
+    plt.figure(figsize=(10, 6))
+    sns.histplot(results_df['error'], kde=True, bins=30)
+    plt.title('Distribution of Prediction Errors')
+    plt.xlabel('Error (kW)')
+    plt.ylabel('Frequency')
+    plt.savefig('results/error_distribution.png')
+    plt.close()

quartz_solar_forecast/evaluation.py

@@ -16,7 +16,7 @@
 from quartz_solar_forecast.eval.nwp import get_nwp
 from quartz_solar_forecast.eval.pv import get_pv_metadata, get_pv_truth
 from quartz_solar_forecast.eval.utils import combine_forecast_ground_truth
-
+from quartz_solar_forecast.eval.visualize import visualize_results 
 from dotenv import load_dotenv
 
 load_dotenv()
@@ -33,7 +33,7 @@
     )
 
 
-def run_eval(testset_path: str = "dataset/testset.csv"):
+def run_eval(testset_path: str = "dataset/testset.csv", visualize: bool = False):
 
     # load testset from csv
     testset = pd.read_csv(testset_path)
@@ -61,7 +61,8 @@ def run_eval(testset_path: str = "dataset/testset.csv"):
     metrics(results_df, pv_metadata, include_night=False)
 
     # Visualizations
-    # TODO
-
+    if visualize:
+        print("Creating visualizations")
+        visualize_results(results_df)
 
 # run_eval()

requirements.txt

@@ -3,4 +3,6 @@ xarray==2022.12.0
 pv-site-prediction==0.1.17
 pydantic==2.6.2
 huggingface_hub== 0.17.3 # only for evaluation
-python-dotenv==1.0.1
+python-dotenv==1.0.1
+seaborn==0.13.2 # only for evaluation visualization
+matplotlib==3.8.3 # only for evaluation visualization