Update app.py

app.py

@@ -33,6 +33,9 @@ gpas = SPD.apply(grade_to_gpa, 'GradeClass')
 
 SPD = SPD.with_column('GPA', gpas)
 
+def gpa_to_label(gpa):
+    return "PASS" if gpa >= 2.0 else "FAIL"
+
 effectiveness_array = np.array(SPD.column('StudyEffectiveness'))
 gpa_array = np.array(SPD.column('GPA'))
 
@@ -136,6 +139,7 @@ shuffled = data.sample(with_replacement=False)
 size = int(data.num_rows * 0.8)
 
 train = shuffled.take(np.arange(size))
+test = shuffled.take(np.arange(size, data.num_rows))
 
 def distance(r1, r2):
 
@@ -189,151 +193,64 @@ def knn_gpa(test_row, k):
 
     return round(gpa, 2)
 
-def random_forest_gpa(
-    study_time,
-    absences,
-    parental_support,
-    tutoring,
-    extracurricular,
-    sports,
-    music,
-    volunteering,
-    effectiveness
-):
-
-    predictions = []
-
-    np.random.seed(3)
-
-    for i in range(30):
-
-        sample = train.sample(k=train.num_rows, with_replacement=True)
-
-        sample_effectiveness = np.mean(sample.column('StudyEffectiveness'))
-        sample_gpa = np.mean(sample.column('GPA'))
-
-        gpa = (
-            (effectiveness / sample_effectiveness) *
-            sample_gpa
-        )
-
-        gpa += (study_time * 0.015)
-
-        gpa -= (absences * 0.01)
-
-        gpa += (parental_support * 0.05)
-
-        if tutoring:
-            gpa += 0.15
-
-        if extracurricular:
-            gpa += 0.05
-
-        if sports:
-            gpa += 0.05
-
-        if music:
-            gpa += 0.05
-
-        if volunteering:
-            gpa += 0.05
-
-        noise = np.random.normal(0, 0.08)
-
-        gpa += noise
-
-        if gpa < 0:
-            gpa = 0
-
-        if gpa > 4:
-            gpa = 4
-
-        predictions.append(gpa)
-
-    return round(np.mean(predictions), 2)
-
 k = 5
 
-X = []
-
-for i in np.arange(SPD.num_rows):
-
-    row = [
-        SPD.column('StudyTimeWeekly').item(i) / 20,
-        SPD.column('Absences').item(i) / 30,
-        SPD.column('Tutoring').item(i),
-        SPD.column('ParentalSupport').item(i) / 4,
-        SPD.column('Extracurricular').item(i),
-        SPD.column('Sports').item(i),
-        SPD.column('Music').item(i),
-        SPD.column('Volunteering').item(i),
-        SPD.column('StudyEffectiveness').item(i) / 40
-    ]
-
-    X.append(row)
-
-X = np.array(X)
-
-y_gpa = np.array(SPD.column('GPA')).reshape(-1, 1) / 4
-
-np.random.seed(1)
-
-W1_gpa = np.random.normal(0, 1, (9, 16))
-
-W2_gpa = np.random.normal(0, 1, (16, 1))
-
-def sigmoid(x):
-    return 1 / (1 + np.exp(-x))
-
-for i in range(20000):
-
-    hidden = sigmoid(np.dot(X, W1_gpa))
-
-    output = sigmoid(np.dot(hidden, W2_gpa))
-
-    error = y_gpa - output
+correct = 0
 
-    output_change = error * output * (1 - output)
+for i in np.arange(test.num_rows):
 
-    hidden_error = np.dot(output_change, W2_gpa.T)
+    row = test.row(i)
 
-    hidden_change = hidden_error * hidden * (1 - hidden)
+    study_time = row.item('StudyTimeWeekly')
+    absences = row.item('Absences')
+    parental_support = row.item('ParentalSupport')
+    tutoring = row.item('Tutoring')
+    extracurricular = row.item('Extracurricular')
+    sports = row.item('Sports')
+    music = row.item('Music')
+    volunteering = row.item('Volunteering')
+    effectiveness = row.item('StudyEffectiveness')
 
-    W2_gpa = W2_gpa + 0.001 * np.dot(hidden.T, output_change)
-
-    W1_gpa = W1_gpa + 0.001 * np.dot(X.T, hidden_change)
-
-def neural_network_gpa(x):
-
-    hidden = sigmoid(np.dot(x, W1_gpa))
-
-    output = sigmoid(np.dot(hidden, W2_gpa))
+    tree_gpa = predict_tree_gpa(
+        study_time,
+        absences,
+        parental_support,
+        tutoring,
+        extracurricular,
+        sports,
+        music,
+        volunteering,
+        effectiveness
+    )
 
-    gpa = float(output[0]) * 4
+    linear_gpa = linear_regression_gpa(effectiveness)
 
-    if gpa < 0:
-        gpa = 0
+    knn_prediction = knn_gpa(row, k)
 
-    if gpa > 4:
-        gpa = 4
+    predictions = [
+        gpa_to_label(tree_gpa),
+        gpa_to_label(linear_gpa),
+        gpa_to_label(knn_prediction)
+    ]
 
-    return round(gpa, 2)
+    final_prediction = max(set(predictions), key=predictions.count)
 
-def create_output(gpa):
+    actual = gpa_to_label(row.item('GPA'))
 
-    result = "PASS" if gpa >= 2.0 else "FAIL"
+    if final_prediction == actual:
+        correct += 1
 
-    return f"{result}\nGPA: {gpa}"
+ensemble_accuracy = round(correct / test.num_rows, 4)
 
-def create_average_output(gpa):
+def majority_vote(predictions):
 
-    result = "PASS" if gpa >= 2.0 else "FAIL"
+    passes = predictions.count("PASS")
+    fails = predictions.count("FAIL")
 
-    return f"""
-FINAL RESULT: {result}
+    if passes > fails:
+        return "PASS"
 
-AVERAGE GPA PREDICTION: {gpa}
-"""
+    return "FAIL"
 
 def predict_models(
     outside_study_time,
@@ -373,18 +290,6 @@ def predict_models(
         study_effect
     )
 
-    rf_gpa = random_forest_gpa(
-        study_time,
-        absences,
-        parental_support,
-        tutoring,
-        extracurricular,
-        sports,
-        music,
-        volunteering,
-        study_effect
-    )
-
     linear_gpa = linear_regression_gpa(
         study_effect
     )
@@ -408,38 +313,24 @@ def predict_models(
 
     knn_prediction = knn_gpa(test_row, k)
 
-    x = np.array([
-        study_time / 20,
-        absences / 30,
-        tutoring,
-        parental_support / 4,
-        extracurricular,
-        sports,
-        music,
-        volunteering,
-        study_effect / 40
+    tree_label = gpa_to_label(tree_gpa)
+    linear_label = gpa_to_label(linear_gpa)
+    knn_label = gpa_to_label(knn_prediction)
+
+    final_prediction = majority_vote([
+        tree_label,
+        linear_label,
+        knn_label
     ])
 
-    nn_prediction = neural_network_gpa(x)
-
-    average_gpa = round(
-        (
-            tree_gpa +
-            rf_gpa +
-            linear_gpa +
-            knn_prediction +
-            nn_prediction
-        ) / 5,
-        2
-    )
+    accuracy_output = f"Combined Model Accuracy: {ensemble_accuracy}"
 
     return (
-        create_output(tree_gpa),
-        create_output(rf_gpa),
-        create_output(linear_gpa),
-        create_output(knn_prediction),
-        create_output(nn_prediction),
-        create_average_output(average_gpa)
+        tree_label,
+        linear_label,
+        knn_label,
+        final_prediction,
+        accuracy_output
     )
 
 theme = gr.themes.Soft(
@@ -451,7 +342,6 @@ theme = gr.themes.Soft(
 )
 
 with gr.Blocks(
-    theme=theme,
     fill_height=True
 ) as app:
 
@@ -459,7 +349,7 @@ with gr.Blocks(
         """
 # Student Performance Predictor
 
-### Predict GPA and pass/fail results using multiple machine learning models
+### Predict pass/fail results using multiple machine learning models
 """
     )
 
@@ -538,51 +428,36 @@ with gr.Blocks(
 
                 tree_output = gr.Textbox(
                     label="Decision Tree",
-                    lines=2,
-                    interactive=False,
-                    scale=1
+                    lines=1,
+                    interactive=False
                 )
 
-                rf_output = gr.Textbox(
-                    label="Random Forest",
-                    lines=2,
-                    interactive=False,
-                    scale=1
-                )
-
-            with gr.Row(equal_height=True):
-
                 linear_output = gr.Textbox(
                     label="Linear Regression",
-                    lines=2,
-                    interactive=False,
-                    scale=1
+                    lines=1,
+                    interactive=False
                 )
 
+            with gr.Row(equal_height=True):
+
                 knn_output = gr.Textbox(
                     label="KNN",
-                    lines=2,
-                    interactive=False,
-                    scale=1
+                    lines=1,
+                    interactive=False
                 )
 
-            with gr.Row(equal_height=True):
-
-                nn_output = gr.Textbox(
-                    label="Neural Network",
-                    lines=2,
-                    interactive=False,
-                    scale=1
+                final_output = gr.Textbox(
+                    label="Final Majority Vote",
+                    lines=1,
+                    interactive=False
                 )
 
     with gr.Row():
 
-        average_output = gr.Textbox(
-            label="Average Prediction",
-            lines=5,
-            interactive=False,
-            max_lines=5,
-            scale=1
+        accuracy_box = gr.Textbox(
+            label="Combined Model Accuracy",
+            lines=1,
+            interactive=False
         )
 
     btn.click(
@@ -601,11 +476,10 @@ with gr.Blocks(
         ],
         outputs=[
             tree_output,
-            rf_output,
             linear_output,
             knn_output,
-            nn_output,
-            average_output
+            final_output,
+            accuracy_box
         ]
     )