from keras.layers import Input, UpSampling2D, Dense, GlobalAveragePooling2D
from keras.applications import MobileNet
from keras.models import Model
from keras.utils import to_categorical
from keras.datasets import cifar10
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt
import numpy as np
import pickle, os, zipfile, glob
#1から訓練させる
def create_cnn():
net = MobileNet(input_shape=(128,128,3), weights=None, include_top=False)
# upsampling(32->128)
input = Input((32,32,3))
x = UpSampling2D(4)(input)
x = net(x)
x = GlobalAveragePooling2D()(x)
x = Dense(10, activation="softmax")(x)
model = Model(input, x)
model.summary()
return model
def train(n_labeled_data):
model = create_cnn()
(X_train, y_train), (X_test, y_test) = cifar10.load_data()
indices = np.arange(X_train.shape[0])
np.random.shuffle(indices)
y_test_true = np.ravel(y_test)
X_train = X_train[indices[:n_labeled_data]] / 255.0
X_test = X_test / 255.0
y_train = to_categorical(y_train[indices[:n_labeled_data]], 10)
y_test = to_categorical(y_test, 10)
model.compile("adam", loss="categorical_crossentropy", metrics=["acc"])
if not os.path.exists("result_mobilenet"):
os.mkdir("result_mobilenet")
hist = model.fit(X_train, y_train, batch_size=min(n_labeled_data, 256),
validation_data=(X_test, y_test), epochs=100).history
with open(f"result_mobilenet/history_{n_labeled_data:05}.dat", "wb") as fp:
pickle.dump(hist, fp)
# tsne-plot
emb_model = Model(model.input, model.layers[-2].output)
embedding = emb_model.predict(X_test)
proj = TSNE(n_components=2).fit_transform(embedding)
cmp = plt.get_cmap("tab10")
plt.figure()
for i in range(10):
select_flag = y_test_true == i
plt_latent = proj[select_flag, :]
plt.scatter(plt_latent[:,0], plt_latent[:,1], color=cmp(i), marker=".")
plt.savefig(f"result_mobilenet/embedding_{n_labeled_data:05}.png")
if __name__ == "__main__":
n_batches = [500, 1000, 5000, 10000]
for nb in n_batches:
train(nb)
with zipfile.ZipFile("result_mobilenet.zip", "w") as zip:
for f in glob.glob("result_mobilenet/*"):
zip.write(f)
