new file: dataset.py

dataset.py

@@ -0,0 +1,64 @@
+import numpy as np
+
+TIMESTEPS = 120
+
+def generate_normal():
+    acc = np.random.normal(9.8, 2, (TIMESTEPS, 3))
+    gyro = np.random.normal(0, 0.5, (TIMESTEPS, 3))
+    speed = np.linspace(30, 50, TIMESTEPS)
+    return np.hstack([acc, gyro, speed.reshape(-1,1)]), 0
+
+
+def generate_brake():
+    acc = np.random.normal(9.8, 2, (TIMESTEPS, 3))
+    acc[-20:] += np.random.normal(-10, 3, (20, 3))
+    gyro = np.random.normal(0, 1, (TIMESTEPS, 3))
+    speed = np.linspace(50, 5, TIMESTEPS)
+    return np.hstack([acc, gyro, speed.reshape(-1,1)]), 0
+
+
+def generate_crash():
+    acc = np.random.normal(9.8, 2, (TIMESTEPS, 3))
+    acc[80:90] += np.random.normal(40, 12, (10, 3))
+
+    gyro = np.random.normal(0, 0.5, (TIMESTEPS, 3))
+    gyro[80:90] += np.random.normal(25, 10, (10, 3))
+
+    speed = np.linspace(60, 0, TIMESTEPS)
+    speed[90:] = 0
+
+    return np.hstack([acc, gyro, speed.reshape(-1,1)]), 1
+
+
+def generate_phone_drop():
+    acc = np.random.normal(9.8, 2, (TIMESTEPS, 3))
+    acc[60:70] += np.random.normal(25, 5, (10, 3))
+
+    gyro = np.random.normal(0, 2, (TIMESTEPS, 3))
+    speed = np.zeros(TIMESTEPS)
+
+    return np.hstack([acc, gyro, speed.reshape(-1,1)]), 0
+
+
+def create_dataset(n=5000):
+    X, y = [], []
+
+    for _ in range(n):
+        choice = np.random.choice(
+            ["normal", "brake", "crash", "phone"],
+            p=[0.25, 0.25, 0.3, 0.2]
+        )
+
+        if choice == "normal":
+            d, l = generate_normal()
+        elif choice == "brake":
+            d, l = generate_brake()
+        elif choice == "crash":
+            d, l = generate_crash()
+        else:
+            d, l = generate_phone_drop()
+
+        X.append(d)
+        y.append(l)
+
+    return np.array(X), np.array(y)

features.py

@@ -0,0 +1,18 @@
+import numpy as np
+
+def extract_features(sample):
+    acc = np.sqrt(sample[:,0]**2 + sample[:,1]**2 + sample[:,2]**2)
+
+    return [
+        np.mean(acc),
+        np.max(acc),
+        np.std(acc),
+
+        np.max(np.abs(sample[:,3:6])),
+
+        np.mean(sample[:,6]),
+        np.min(sample[:,6]),
+        np.max(sample[:,6]),
+
+        np.mean(acc[-10:])
+    ]