fix bug with choose model, add more model metadata

api/helpers.py

@@ -28,9 +28,26 @@ def load_models():
 
 def choose_model(transformed_params):
     if "askingPrice" in transformed_params:
-        return models["bostadspriser-with-askingPrice"]
 
-    return models["bostadspriser-without-askingPrice"]
+        with_asking_price_models = []
+        for name, model in models.items():
+            if "with-askingPrice" in name:
+                with_asking_price_models.append(model)
+
+        # sort by trainedAt
+        with_asking_price_models.sort(key=lambda x: x["metadata"]["trainedAt"])
+
+        return with_asking_price_models[-1]    
+    else:
+        without_asking_price_models = []
+        for name, model in models.items():
+            if "without-askingPrice" in name:
+                without_asking_price_models.append(model)
+
+        # sort by trainedAt
+        without_asking_price_models.sort(key=lambda x: x["metadata"]["trainedAt"])
+
+        return without_asking_price_models[-1]
 
 
 def get_live_listings(page: int, page_size: int):

api/models/bostadspriser-with-askingPrice-2024-01-05/metadata.json

@@ -1 +1,36 @@
-{"name": "bostadspriser-with-askingPrice", "nameWithDate": "bostadspriser-with-askingPrice-2024-01-05", "features": ["fee", "livingArea", "rooms", "runningCosts", "hasElevator", "hasBalcony", "lat", "long", "cpi", "hasHousingCooperative", "isPlot", "isWinterLeisureHouse", "isApartment", "isFarmWithForest", "isHouse", "isRowHouse", "isPairHouse", "isPairTerracedRowHouse", "isTerracedHouse", "isFarmWithoutForest", "isLeisureHouse", "isOtherHousingForm", "isFarmWithAgriculture", "age", "sinceLastRenovation", "soldYear", "soldMonth", "askingPrice"], "target": "finalPrice", "trainedAt": "2024-01-05T15:01:59.314123"}
+{
+  "name": "bostadspriser-with-askingPrice",
+  "nameWithDate": "bostadspriser-with-askingPrice-2024-01-05",
+  "features": [
+    "fee",
+    "livingArea",
+    "rooms",
+    "runningCosts",
+    "hasElevator",
+    "hasBalcony",
+    "lat",
+    "long",
+    "cpi",
+    "hasHousingCooperative",
+    "isPlot",
+    "isWinterLeisureHouse",
+    "isApartment",
+    "isFarmWithForest",
+    "isHouse",
+    "isRowHouse",
+    "isPairHouse",
+    "isPairTerracedRowHouse",
+    "isTerracedHouse",
+    "isFarmWithoutForest",
+    "isLeisureHouse",
+    "isOtherHousingForm",
+    "isFarmWithAgriculture",
+    "age",
+    "sinceLastRenovation",
+    "soldYear",
+    "soldMonth",
+    "askingPrice"
+  ],
+  "target": "finalPrice",
+  "trainedAt": "2024-01-05T15:01:59.314123"
+}

model/train.py

@@ -164,14 +164,14 @@ def evaluate_model(
     r2 = r2_score(y_test, y_pred)
 
     print(f"[{name}] MSE: " + str(mse) + ", RMSE: " + str(rmse) + ", R^2: " + str(r2))
-    return gs_model, mse, rmse, r2
+    return gs_model, gs_model.best_params_, mse, rmse, r2
 
 
 # def train, takes in a test train split and returns the best model and its results
 def train(name, X_train, y_train, X_test, y_test):
     results = {}
     for regressor_name, regressor in regressors.items():
-        model, mse, rmse, r2 = evaluate_model(
+        model, best_params, mse, rmse, r2 = evaluate_model(
             name,
             regressor_name,
             regressor["model"],
@@ -181,7 +181,7 @@ def train(name, X_train, y_train, X_test, y_test):
             X_test,
             y_test,
         )
-        results[regressor_name] = {"model": model, "MSE": mse, "RMSE": rmse, "R^2": r2}
+        results[regressor_name] = {"model": model, "bestParams": best_params, "mse": mse, "rmse": rmse, "r2": r2}
 
     return results
 
@@ -224,7 +224,7 @@ def main():
         results_df = pd.DataFrame(results)
         results_df = results_df.drop("model", axis=0)
         results_df = results_df.transpose()
-        results_df = results_df.sort_values(by=["RMSE"])
+        results_df = results_df.sort_values(by=["rmse"])
         best_model = results_df.index[0]
         print(f"[{name}] Best model: " + best_model)
 
@@ -252,6 +252,13 @@ def main():
             "features": setup["features"],
             "target": setup["target"],
             "trainedAt": now.isoformat(),
+            "model": {
+                "name": best_model,
+                "params": results[best_model]["bestParams"],
+                "mse": results[best_model]["mse"],
+                "rmse": results[best_model]["rmse"],
+                "r2": results[best_model]["r2"],
+            }
         }
         with open(f"{folder}/metadata.json", "w") as f:
             json.dump(metadata, f)

models/bostadspriser-with-askingPrice-2024-01-05/metadata.json

@@ -1 +1 @@
-{"name": "bostadspriser-with-askingPrice", "nameWithDate": "bostadspriser-with-askingPrice-2024-01-05", "features": ["fee", "livingArea", "rooms", "runningCosts", "hasElevator", "hasBalcony", "lat", "long", "cpi", "hasHousingCooperative", "isPlot", "isWinterLeisureHouse", "isApartment", "isFarmWithForest", "isHouse", "isRowHouse", "isPairHouse", "isPairTerracedRowHouse", "isTerracedHouse", "isFarmWithoutForest", "isLeisureHouse", "isOtherHousingForm", "isFarmWithAgriculture", "age", "sinceLastRenovation", "soldYear", "soldMonth", "askingPrice"], "target": "finalPrice", "trainedAt": "2024-01-05T15:01:59.314123"}
+{"name": "bostadspriser-with-askingPrice", "nameWithDate": "bostadspriser-with-askingPrice-2024-01-05", "features": ["fee", "livingArea", "rooms", "runningCosts", "hasElevator", "hasBalcony", "lat", "long", "cpi", "hasHousingCooperative", "isPlot", "isWinterLeisureHouse", "isApartment", "isFarmWithForest", "isHouse", "isRowHouse", "isPairHouse", "isPairTerracedRowHouse", "isTerracedHouse", "isFarmWithoutForest", "isLeisureHouse", "isOtherHousingForm", "isFarmWithAgriculture", "age", "sinceLastRenovation", "soldYear", "soldMonth", "askingPrice"], "target": "finalPrice", "trainedAt": "2024-01-05T15:10:07.275350", "model": {"name": "Linear Regression", "params": {}, "mse": 114645181342.62703, "rmse": 338592.94343300635, "r2": 0.9751360245953378}}

models/bostadspriser-with-askingPrice-2024-01-05/results.csv

@@ -1,2 +1,2 @@
-,MSE,RMSE,R^2
-Linear Regression,114645181342.62703,338592.94343300635,0.9751360245953378
+,bestParams,mse,r2,rmse
+Linear Regression,{},114645181342.62703,0.9751360245953378,338592.94343300635

models/bostadspriser-without-askingPrice-2024-01-05/metadata.json

@@ -1 +1 @@
-{"name": "bostadspriser-without-askingPrice", "nameWithDate": "bostadspriser-without-askingPrice-2024-01-05", "features": ["fee", "livingArea", "rooms", "runningCosts", "hasElevator", "hasBalcony", "lat", "long", "cpi", "hasHousingCooperative", "isPlot", "isWinterLeisureHouse", "isApartment", "isFarmWithForest", "isHouse", "isRowHouse", "isPairHouse", "isPairTerracedRowHouse", "isTerracedHouse", "isFarmWithoutForest", "isLeisureHouse", "isOtherHousingForm", "isFarmWithAgriculture", "age", "sinceLastRenovation", "soldYear", "soldMonth"], "target": "finalPrice", "trainedAt": "2024-01-05T15:01:52.948934"}
+{"name": "bostadspriser-without-askingPrice", "nameWithDate": "bostadspriser-without-askingPrice-2024-01-05", "features": ["fee", "livingArea", "rooms", "runningCosts", "hasElevator", "hasBalcony", "lat", "long", "cpi", "hasHousingCooperative", "isPlot", "isWinterLeisureHouse", "isApartment", "isFarmWithForest", "isHouse", "isRowHouse", "isPairHouse", "isPairTerracedRowHouse", "isTerracedHouse", "isFarmWithoutForest", "isLeisureHouse", "isOtherHousingForm", "isFarmWithAgriculture", "age", "sinceLastRenovation", "soldYear", "soldMonth"], "target": "finalPrice", "trainedAt": "2024-01-05T15:10:01.278979", "model": {"name": "Linear Regression", "params": {}, "mse": 3155263694507.91, "rmse": 1776306.1939057438, "r2": 0.3156938828418222}}

models/bostadspriser-without-askingPrice-2024-01-05/results.csv

@@ -1,2 +1,2 @@
-,MSE,RMSE,R^2
-Linear Regression,3155263694507.91,1776306.1939057438,0.3156938828418222
+,bestParams,mse,r2,rmse
+Linear Regression,{},3155263694507.91,0.3156938828418222,1776306.1939057438