In [ ]:

%matplotlib inline
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

In [ ]:

from tensorflow.keras.datasets import mnist
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Input, Dense
from tensorflow.keras.optimizers import SGD, Adam

In [ ]:

!pip install -U tensorflow-addons
from tensorflow_addons.losses import TripletSemiHardLoss

Requirement already satisfied: tensorflow-addons in /usr/local/lib/python3.7/dist-packages (0.16.1)
Requirement already satisfied: typeguard>=2.7 in /usr/local/lib/python3.7/dist-packages (from tensorflow-addons) (2.7.1)

In [ ]:

(x_train, y_train), (x_test, y_test) = mnist.load_data()

x_train = x_train.reshape(-1, 28*28)/255
x_test = x_test.reshape(-1, 28*28)/255

1. 建立我們的 FaceNet 神經網路模型¶

In [ ]:

model = Sequential()
model.add(Dense(128, activation='relu', input_shape=(784,)))
model.add(Dense(64, activation='relu'))
model.add(Dense(2))

In [ ]:

model.compile(optimizer=Adam(0.001), 
                   loss=TripletSemiHardLoss(margin=0.87))

In [ ]:

history = model.fit(x_train, y_train, epochs=5)

Epoch 1/5
1875/1875 [==============================] - 8s 4ms/step - loss: 0.0483
Epoch 2/5
1875/1875 [==============================] - 8s 4ms/step - loss: 0.0364
Epoch 3/5
1875/1875 [==============================] - 7s 4ms/step - loss: 0.0296
Epoch 4/5
1875/1875 [==============================] - 7s 4ms/step - loss: 0.0257
Epoch 5/5
1875/1875 [==============================] - 7s 4ms/step - loss: 0.0219

In [ ]:

def representation(x, y):
    y_pred = model.predict(x)
    idx = np.random.randint(x.shape[0], size=2000)
    colors = ['r', 'g', 'b', 'c', 'm', 'y', 'k', 'lime', 'orange', 'purple']
    for label, c in enumerate(colors):
        label_idx = idx[y[idx]==label]
        plt.scatter(y_pred[label_idx, 0], y_pred[label_idx, 1], 
                    c=c, label=str(label))
    plt.legend();

In [ ]:

plt.figure(figsize=(12, 6))
plt.subplot(1, 2, 1)
representation(x_train, y_train)
plt.title("Training Data")
plt.subplot(1, 2, 2)
representation(x_test, y_test)
plt.title("Testing Data");