traget variable deviation -> deviation rate

README.md

@@ -11,7 +11,7 @@ pinned: false
 
 # Cricket-Prophet: AI-Powered Cricket Score Prediction
 
-**Predict cricket scores with accuracy beyond traditional projections!**
+**Predict cricket scores with accuracy beyond traditional encounter-based or runrate-based predictions!**
 
 ## Key Features
 
@@ -27,7 +27,7 @@ pinned: false
 ## How It Works
 
 1. **Live score scraping:** Fetches live match data from CricBuzz.
-2. **Feature engineering:** Extracts relevant features from the match data.
+2. **Feature engineering:** Extracts relevant features from the match data and tried to predict deviation of score from projected score
 3. **Model prediction:** Applies a pre-trained Random Forest model to predict the final score.
 4. **Visualization:** Displays predictions and insights clearly.
 
@@ -39,12 +39,12 @@ Refer to the below table, think of variance as MSE of projected score
 
 | Format | Measuring after n balls played | Variance (Train) | Variance (Test) | Model MSE (Train) | Model MSE (Test) |
 | ------ | ------------------------------ | ---------------- | --------------- | ----------------- | ---------------- |
-| T20    | 30                             | 498.41           | 514.90          | 8.88              | 8.91             |
-| T20    | 60                             | 218.03           | 234.74          | 5.78              | 5.89             |
-| T20    | 90                             | 80.19            | 82.53           | 3.94              | 4.03             |
-| ODI    | 120                            | 745.17           | 702.39          | 12.50             | 14.52            |
-| ODI    | 180                            | 360.02           | 358.21          | 9.01              | 10.03            |
-| ODI    | 240                            | 134.28           | 125.87          | 5.00              | 5.93             |
+| T20    | 30                             | 498.41           | 514.90          | 9.02              | 9.26             |
+| T20    | 60                             | 218.03           | 234.74          | 5.73              | 5.89             |
+| T20    | 90                             | 80.19            | 82.53           | 3.82              | 3.79             |
+| ODI    | 120                            | 745.17           | 702.39          | 13.28             | 13.97            |
+| ODI    | 180                            | 360.02           | 358.21          | 9.12              | 9.83             |
+| ODI    | 240                            | 134.28           | 125.87          | 4.33              | 5..06            |
 
 **ODI Format**
 

features.py

@@ -7,7 +7,7 @@
 import matplotlib.pyplot as plt
 from cricksheet import get_all_matches
 
-# import ydata_profiling
+import ydata_profiling
 
 
 ## Reading IPL dataset
@@ -17,6 +17,28 @@
 
 ## Feature selection/creation and ngram creation
 
+features_for_profiling = features = [
+    # "batting_team",
+    # "bowling_team",
+    # "balls",
+    # "runs",
+    # "wickets",
+    "wkt_last_5_overs",
+    # "runrate_last_5_overs",
+    "current_RR",
+    # "average",
+    "balls_left",
+    "wkts_left",
+    # "required_RR",
+    # "projected_score_more",
+    # "min_score_more",
+    # "max_score_more",
+    # "projected_avg_score_more",
+    "runrate_last_5_overs-current_RR",
+    "deviation_from_projected_rate",
+    "deviation_from_projected",
+]
+
 features = [
     "matchid",
     "format",
@@ -41,6 +63,7 @@
     "final_score",
     "final_score_more",
     "deviation_from_projected",
+    "deviation_from_projected_rate",
 ]
 
 getformat = {"ODI": 1, "T20": 2}
@@ -113,6 +136,7 @@ def extract_features(inning):
                 inning.final_score,
                 final_score_more,
                 round(deviation_from_projected),
+                (deviation_from_projected * 6) / balls_left,
             )
         )
     return data
@@ -129,7 +153,9 @@ def save_features(innings, fname):
     Xy = [xi for Xi in Xy for xi in Xi]
     print(f"{len(Xy)=}")
     featuresdf = pd.DataFrame(Xy, columns=features)
-    # ydata_profiling.ProfileReport(featuresdf, title=fname).to_file(fname + ".html")
+    ydata_profiling.ProfileReport(
+        featuresdf[features_for_profiling], title=fname
+    ).to_file(fname + ".html")
     featuresdf.to_feather(fname)
     featuresdf.to_csv(fname + ".csv")
 

model.py

@@ -42,16 +42,19 @@
     # "projected_avg_score_more",
     "runrate_last_5_overs-current_RR",
 ]
-target = "deviation_from_projected"
+target = "deviation_from_projected_rate"
 
 
 # evaluate
 def evaluate(model, featuresdf, x_test, fname):
     predictdf = featuresdf.loc[x_test.index].copy()
     # print(predictdf.columns)
-    predictdf["h_deviation_from_projected"] = model.predict(
+    predictdf["h_deviation_from_projected_rate"] = model.predict(
         featuresdf.loc[x_test.index][features]
     )
+    predictdf["h_deviation_from_projected"] = (
+        predictdf["h_deviation_from_projected_rate"] * predictdf["balls_left"] / 6
+    )
     predictdf["error"] = (
         predictdf["h_deviation_from_projected"] - predictdf["deviation_from_projected"]
     )
@@ -123,12 +126,24 @@ def plot_feature_importance(f, imp, fname):
 
 
 def compare_util(model, fname, x_train, x_test, y_train, y_test, balls_left):
-    print(f'After {(300 if "odi" in fname.lower() else 120) - balls_left} overs')
+    print(f'After {(300 if "odi" in fname.lower() else 120) - balls_left} balls')
+
+    train_index = x_train["balls_left"] <= balls_left
+    test_index = x_test["balls_left"] <= balls_left
+    y_train_data = y_train[train_index] * x_train["balls_left"][train_index] / 6
+    y_test_data = y_test[test_index] * x_test["balls_left"][test_index] / 6
+    h_train_data = (
+        model.predict(x_train[train_index]) * x_train["balls_left"][train_index] / 6
+    )
+    h_test_data = (
+        model.predict(x_test[test_index]) * x_test["balls_left"][test_index] / 6
+    )
+
     print(
-        f"Variance: {statistics.variance(y_train[x_train['balls_left']<=balls_left])=}, {statistics.variance(y_test[x_test['balls_left']<=balls_left])=}"
+        f"Data: Train Variance: {statistics.variance(y_train_data)}, Test Variance: {statistics.variance(y_test_data)}"
     )
     print(
-        f"Model: {mse(model.predict(x_train[x_train['balls_left']<=balls_left]), y_train[x_train['balls_left']<=balls_left], squared=False)=}, {mse(model.predict(x_test[x_test['balls_left']<=balls_left]), y_test[x_test['balls_left']<=balls_left], squared=False)=}"
+        f"Model: Train MSE: {mse(h_train_data, y_train_data, squared=False)}, Test MSE: {mse(h_test_data, y_test_data, squared=False)}"
     )
 
 
@@ -199,7 +214,7 @@ def train(fname, max_depth=-1):
 
     # print(f"{model.score(x_train, y_train)=}, {model.score(x_test, y_test)=}")
     print(
-        f"{mse(model.predict(x_train), y_train, squared=False)=}, {mse(model.predict(x_test), y_test, squared=False)=}"
+        f"Train MSE: {mse(model.predict(x_train)*x_train['balls_left']/6, y_train*x_train['balls_left']/6, squared=False)}, Test MSE: {mse(model.predict(x_test)*x_test['balls_left']/6, y_test*x_test['balls_left']/6, squared=False)}"
     )
 
     compare(model, fname, x_train, x_test, y_train, y_test)

serve.py

@@ -165,8 +165,9 @@ def predict(url):
         inputdf["batting_team"] = all_teams_enc
     inputdf = inputdf[features]
     model = load_model(format)
-    h = model.predict(inputdf)
-    print(f"{h=}")
+    h_rate = model.predict(inputdf)
+    print(f"{h_rate=}")
+    h = h_rate * balls_left / 6
     projected_score_more = balls_left * current_rr / 6
     projected = math.ceil(projected_score_more + runs)
     predicted_score_more = math.ceil(h.mean() + projected_score_more)
@@ -380,7 +381,7 @@ def render(url):
 
         live_matches = get_live_matches("https://cricbuzz.com")
         if live_matches:
-            option = st.selectbox(
+            option = st.sidebar.selectbox(
                 "Choose a live match here",
                 list(live_matches.keys()) + ["Custom URL", "Simulator"],
             )
@@ -407,9 +408,15 @@ def render(url):
                     - args["current_RR"]
                 )
                 balls = 300 if format == "ODI" else 120
-                st.text(
-                    str(int((balls * args["current_RR"] / 6) + simulator(args, format)))
-                )
+                if st.button("Predict"):
+                    st.text(
+                        str(
+                            int(
+                                (balls * args["current_RR"] / 6)
+                                + simulator(args, format) * args["balls_left"] / 6
+                            )
+                        )
+                    )
             else:
                 if option == "Custom URL":
                     url = st.text_input("Enter cricbuzz match link")