About the deviation when learning sin waves in machine learning

 from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
import numpy as np
import matplotlib.pyplot asplt
from sklearn.model_selection import train_test_split

# === 3-a. Create Data ===
x = np.linspace (-np.pi, np.pi).reshape (-1,1)
# reduce the joint to six cycles for learning
for i in range (5):
    x = np.append(x,x)
    t=np.sin(x)

# === 3-b. Create Model ===
# Set parameters
batch_size=8# batch size
n_in=1# Number of neurons in the input layer
n_mid=20# number of neurons in the intermediate layer
n_out=1# Number of neurons in the output layer

# Generate neural network models
model=Sequential()
model.add(Dense(n_mid, input_shape=(n_in,), activation="sigmoid"))
model.add(Dense(n_out, activation="linear"))
model.compile(loss="mean_squared_error", optimizer="sgd")
print(model.summary())

# ===3-c. Model Learning ===
history=model.fit(x,t,batch_size=batch_size,epochs=2000,validation_split=0.1)

# === 3-d. Visualization of learning trends ===
loss_train=history.history ['loss']
loss_val=history.history ['val_loss']

plt.plot(np.range(len(loss_train))), loss_train, label="loss_train")
plt.plot(np.range(len(loss_val))), loss_val, label="loss_val")
plt.legend()
plt.show()

print(len(x))
# prediction
for i in range (20):
    x_train, x_test=train_test_split(x, test_size=5, shuffle=False)
    r=model.predict(x_test)
    t=np.append(t,r[4])
    x = np.append(x, x[i+1])

# === 3-e.sin wave prediction ===
print(x)
#plt.plot(x,model.predict(x))
plt.plot(x,t)
plt.show()