Машинное обучение Kaggle

#Plotting gradient descent path
#Gradient descent
from sklearn.model_selection import train_test_split
import pandas as pd
davis_df = pd.read_csv("/kaggle/input/mash-obuch-1603-new/mash-obuch-1603.csv")
 
#Get rid of impossible values
davis_df = davis_df[davis_df.Age > 10].dropna()
davis_train, davis_test = train_test_split(davis_df, train_size=0.5, stratify=davis_df.Gender)
y = davis_train['Genre']
x = davis_train['Age'].to_numpy()
 
w = 1
b = 0
alpha = 0.0001
old_w = []
 
max_norm = 4
for t in range(10000):
    l = ((1/len(x)) * np.sum((y - w * x - b)**2))
    d_yh = (w * x + b - y)
    d_w = np.mean(d_yh * x)
    d_b = np.mean(d_yh)
    norm = np.sqrt(d_w2 + d_b2)
    if norm > max_norm:
        reciprocal = norm / max_norm
        d_w /= reciprocal
        d_b /= reciprocal
        w = w - alpha * d_w
        b = b - alpha * d_b
    if t % 500 == 0:
        old_w.append(np.array((w, b)))
print(w, b)
all_ws = np.array(old_w)
ww,bb = np.mgrid[-1:1.5:100j, -0.25:0.1:100j]
L = []
heights = davis_train['Age'].values.astype(np.float64)
weights = davis_train['Genre'].values.astype(np.float64)
for w, b in zip(ww.ravel(),bb.ravel()):
    L.append(1/len(heights) * np.sum((weights - w * heights - b)**2))
L = np.array(L)
plt.figure(figsize=(10, 8))
plt.title("Contour Plot of Gradient Descent")
plt.xlabel("w")
plt.ylabel("b")
plt.contourf(ww,bb,L.reshape(ww.shape),levels=50)
for i in range(len(old_w) - 1):
    plt.annotate('', xy=all_ws[i + 1, :], xytext=all_ws[i, :],
                 arrowprops={'arrowstyle': '->', 'color': 'r', 'lw': 1},
                 va='center', ha='center')