Merge branch 'aksel'

past_weather_data.csv

@@ -15,4 +15,5 @@ date,temp,humidity,precip,windspeed,sealevelpressure,visibility,solarradiation
 2023-10-30,11.2,90.4,13.0,18.4,997.5,28.8,27.0
 2023-10-31,11.0,93.7,18.6,18.0,1000.7,17.9,29.8
 2023-11-01,12.4,88.5,4.9,25.9,997.8,32.6,31.5
-2023-11-02,11,80,8.7,46.4,976.4,33.6,21.5
+2023-11-02,11.0,80.0,8.7,46.4,976.4,33.6,21.5
+2023-11-03,9.6,83.3,7.9,32.4,981.6,31,40.1

pollution_data.csv

@@ -12,4 +12,5 @@ date,NO2,O3
 2024-10-27,27.53722134983982,20.80809239842384
 2024-10-28,23.337567567567568,26.82861788617886
 2024-10-29,16.53533209586906,23.28254887605004
-2024-10-30,22.26162162162162,18.03443548387097
+2024-10-30,22.26162162162162,18.03443548387097
+2024-10-31,24.919333333333334,20.79696

predictions_history.csv

@@ -70,3 +70,9 @@ NO2,2024-10-30,2024-10-31,29.77507241979599
 O3,2024-10-30,2024-11-01,21.135906183680472
 NO2,2024-10-30,2024-11-01,28.38872595850704
 O3,2024-10-30,2024-11-02,19.67426015042635
+O3,2024-10-31,2024-11-01,16.491393851863755
+NO2,2024-10-31,2024-11-01,17.22825222459993
+O3,2024-10-31,2024-11-02,16.874728806873033
+NO2,2024-10-31,2024-11-02,14.771381333796965
+O3,2024-10-31,2024-11-03,15.244292496093546
+NO2,2024-10-31,2024-11-03,14.606430068166452

src/predict.py

@@ -3,9 +3,9 @@ from datetime import date, datetime, timedelta
 
 import joblib
 import pandas as pd
+import torch
 from dotenv import load_dotenv
 from huggingface_hub import hf_hub_download, login
-
 from src.data_api_calls import (
     get_combined_data,
     update_pollution_data,
@@ -17,26 +17,65 @@ load_dotenv()
 login(token=os.getenv("HUGGINGFACE_DOWNLOAD_TOKEN"))
 
 
+def load_nn():
+    import torch.nn as nn
+    from huggingface_hub import PyTorchModelHubMixin
+
+    class AirPollutionNet(nn.Module, PyTorchModelHubMixin):
+        def __init__(self, input_size, layers, dropout_rate):
+            super(AirPollutionNet, self).__init__()
+            self.layers_list = nn.ModuleList()
+            in_features = input_size
+
+            for units in layers:
+                self.layers_list.append(nn.Linear(in_features, units))
+                self.layers_list.append(nn.ReLU())
+                self.layers_list.append(nn.Dropout(p=dropout_rate))
+                in_features = units
+
+            self.output = nn.Linear(in_features, 3)  # Output size is 3 for next 3 days
+
+        def forward(self, x):
+            for layer in self.layers_list:
+                x = layer(x)
+            x = self.output(x)
+            return x
+
+    model = AirPollutionNet.from_pretrained(
+        "akseljoonas/Utrecht_pollution_forecasting_NO2"
+    )
+    return model
+
+
 def load_model(particle):
     repo_id = f"elisaklunder/Utrecht-{particle}-Forecasting-Model"
     if particle == "O3":
         file_name = "O3_svr_model.pkl"
-    elif particle == "NO2":
-        file_name = "NO2_svr_model.pkl"
+        model_path = hf_hub_download(repo_id=repo_id, filename=file_name)
+        model = joblib.load(model_path)
+    else:
+        model = load_nn()
 
-    model_path = hf_hub_download(repo_id=repo_id, filename=file_name)
-    model = joblib.load(model_path)
     return model
 
 
 def run_model(particle, data):
     input_data = create_features(data=data, target_particle=particle)
     model = load_model(particle)
-    prediction = model.predict(input_data)
 
-    repo_id = f"elisaklunder/Utrecht-{particle}-Forecasting-Model"
-    file_name = f"target_scaler_{particle}.joblib"
-    path = hf_hub_download(repo_id=repo_id, filename=file_name)
+    if particle == "NO2":
+        with torch.no_grad():
+            prediction = model(torch.tensor(input_data.values, dtype=torch.float32))
+        repo_id = "akseljoonas/Utrecht_pollution_forecasting_NO2"
+        file_name = "target_scaler_NO2.joblib"
+        path = hf_hub_download(repo_id=repo_id, filename=file_name)
+    else:
+        prediction = model.predict(input_data)
+
+        repo_id = f"elisaklunder/Utrecht-{particle}-Forecasting-Model"
+        file_name = f"target_scaler_{particle}.joblib"
+        path = hf_hub_download(repo_id=repo_id, filename=file_name)
+
     target_scaler = joblib.load(path)
     prediction = target_scaler.inverse_transform(prediction)
 

weather_data.csv

@@ -13,3 +13,4 @@ date,temp,humidity,precip,windspeed,sealevelpressure,visibility,solarradiation
 2024-10-28,12.4,91.8,1.1,31.7,1021.8,12.8,27.3
 2024-10-29,13.8,95.9,0.2,20.5,1023.1,8.1,16.0
 2024-10-30,12.7,92.9,0.6,9.4,1027.5,12.5,32.8
+2024-10-31,12.5,89.9,0.0,11.2,1027.1,17.1,70.6