In [1]:
from IPython.display import Image, display
使用Inception模型来做Transfer Learning。输入图片，并存储模型处理之后的Transfer Value。Transfer Value是尚未用于分类的输出，里面保存了一张图片的必要信息。当所有图片都处理之后，并存储在磁盘上。接下来使用transfer-value作为其他神经网络的输入，将transfer-value作为新数据样本，并训练新的神经网络分类结果。
In [2]:
Image('data/08_transfer_learning_flowchart.png')
Out[2]:
In [3]:
%matplotlib inline
In [33]:
import matplotlib.pyplot as plt
import tensorflow as tf
import numpy as np
import time
import os
import inception
import prettytensor as pt
import matplotlib.cm as cm
In [5]:
import cifar10
In [6]:
from cifar10 import num_classes
In [7]:
cifar10.maybe_download_and_extract()
In [8]:
class_names = cifar10.load_class_names()
class_names
Out[8]:
加载训练数据：训练图片、整型标签、one-hot编码过的标签
In [9]:
images_train, cls_train, labels_train = cifar10.load_training_data()
In [10]:
images_test, cls_test, labels_test = cifar10.load_test_data()
In [11]:
print("Size of:")
print("- Training-set:\t\t{}".format(len(images_train)))
print("- Test-set:\t\t{}".format(len(images_test)))
In [12]:
def plot_images(images, cls_true, cls_pred=None, smooth=True):
assert len(images) == len(cls_true)
fig, axes = plt.subplots(3, 3)
hspace = 0.3 if cls_pred is None else 0.6
fig.subplots_adjust(hspace=hspace, wspace=0.3)
if smooth:
interpolation = 'spline16'
else:
interpolation = 'nearest'
for i, ax in enumerate(axes.flat):
if i < len(images):
ax.imshow(images[i], interpolation=interpolation)
cls_true_name = class_names[cls_true[i]]
if cls_pred is None:
xlabel = "True: {0}".format(cls_true_name)
else:
cls_pred_name = class_names[cls_pred[i]]
xlabel = "True: {0}\nPred: {1}".format(cls_true_name, cls_pred_name)
ax.set_xlabel(xlabel)
ax.set_xticks([])
ax.set_yticks([])
plt.show()
In [13]:
# Get the first images from the test-set.
images = images_test[0:9]
# Get the true classes for those images.
cls_true = cls_test[0:9]
# Plot the images and labels using our helper-function above.
plot_images(images=images, cls_true=cls_true, smooth=False)
In [14]:
inception.maybe_download()
In [15]:
model = inception.Inception()
In [16]:
from inception import transfer_values_cache
In [17]:
file_path_cache_train = os.path.join(cifar10.data_path, 'inception_cifar10_train.pkl')
file_path_cache_test = os.path.join(cifar10.data_path, 'inception_cifar10_test.pkl')
In [18]:
print (u"使用Inception模型，开始处理输入的验证图片，输出transfer-value并存储")
images_scaled = images_train * 255.0
transfer_values_train = transfer_values_cache(cache_path=file_path_cache_train,
model=model,
images=images_train)
In [20]:
print("使用Inception模型，开始处理输入的测试图片，输出transfer-value并存储 ...")
# Scale images because Inception needs pixels to be between 0 and 255,
# while the CIFAR-10 functions return pixels between 0.0 and 1.0
images_scaled = images_test * 255.0
# If transfer-values have already been calculated then reload them,
# otherwise calculate them and save them to a cache-file.
transfer_values_test = transfer_values_cache(cache_path=file_path_cache_test,
images=images_scaled,
model=model)
In [19]:
transfer_values_train.shape
Out[19]:
In [21]:
transfer_values_test.shape
Out[21]:
In [22]:
transfer_values_test[0].shape
Out[22]:
In [23]:
def plot_transfer_values(i):
print ("input image: ")
plt.imshow(images_test[i], interpolation='nearest')
plt.show()
print("Transfer-values for the image using Inception model:")
img = transfer_values_test[i]
img = img.reshape((32, 64))
plt.imshow(img, interpolation='nearest', cmap='Reds')
plt.show()
In [24]:
plot_transfer_values(16)
In [25]:
plot_transfer_values(17)
In [26]:
from sklearn.decomposition import PCA
In [36]:
pca = PCA(n_components=2)
In [43]:
transfer_values = transfer_values_train[0:5000]
In [44]:
cls = cls_train[0:5000]
In [45]:
transfer_values_reduced = pca.fit_transform(transfer_values)
In [46]:
transfer_values_reduced.shape
Out[46]:
In [47]:
def plot_scatter(values, cls):
cmap = cm.rainbow(np.linspace(0.0, 1.0, num_classes))
colors = cmap[cls]
x = values[:, 0]
y = values[:, 1]
plt.scatter(x, y, color=colors)
plt.show()
In [48]:
plot_scatter(transfer_values_reduced, cls)
In [49]:
from sklearn.manifold import TSNE
In [50]:
pca = PCA(n_components=50)
transfer_values_50d = pca.fit_transform(transfer_values)
In [51]:
tsne = TSNE(n_components=2)
In [ ]:
transfer_values_reduced = tsne.fit_transform(transfer_values_50d)