我们试想一个实际问题,春天到了,我们要买衣服了,同时,作为服装厂商,也要开始发布新的衣服了,如果你作为一个服装厂商的技术顾问,请你分析出什么样的衣服属于今年的流行趋势,你会怎么做?
首先,作为技术宅男的你,我不认为你会对流行元素有那么多的关注,不会去看什么巴黎时装周,你能做的就是根据各种各样的数据进行分析预测。你可能会在店铺中进行一些扫码填写调查表发放一些优惠券,还可能去各大时尚网站去扒一些评论分析文章,还可能去微博这种公开的社交平台扒一些时尚博主的自拍分享等,你可以想尽办法获取各种各样的数据,你要做的是用好这些数据分析出流行趋势。
对于这种分析预测深度学习无疑是适合的模型工具,问题是既有这种调查表得来的数据,适合普通的层堆叠进行训练的模型,还有这种评论分析文章这种文本数据,适合循环神经网络的文章,还有各种图片需要分析,适合卷积神经网络的模型,这可怎么办?根据我们现有的知识,我们可以考虑分别训练不同的网络模型,对他们进行加权分析,这固然是一种方法,但是这个权也太随机了,各个数据之间也都割裂开了,分别处理算不上好的方法。那怎么办?
将他们联合起来,进行联合学习,大致的样子就是这样:
当然,联合模型不只能同时接收多种输入,还可以同时给出输出,比如流行的元素和价格,多个输入多个输出,这就不单纯是一个前面我们提到的各种线性的网络,更像是一个图的感觉,对,就是那种类似于数据结构中的图。就像数据结构中的图要比线性链表复杂得多一样,深度学习中的这种联合模型也比单一的一种模型要复杂得多。
函数式 API
这是预备知识,所谓函数式方式,很幸运 Keras 支持这样的方式,就是把层看成函数,这个函数的参数和返回值都是张量,这样的函数组合起来就能构成网络模型,看下面的代码例子:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 def run (): seq_model = Sequential() seq_model.add(layers.Dense(32 , activation='relu' , input_shape=(64 ,))) seq_model.add(layers.Dense(32 , activation='relu' )) seq_model.add(layers.Dense(10 , activation='softmax' )) seq_model.summary() input_tensor = Input(shape=(64 ,)) x = layers.Dense(32 , activation='relu' )(input_tensor) x = layers.Dense(32 , activation='relu' )(x) output_tensor = layers.Dense(10 , activation='softmax' )(x) model = Model(input_tensor, output_tensor) model.summary()
这两种定义网络模型的方式是等价的,Input 是用来定义张量的,原始数据可以转换成张量用于网络训练,Model 是用来找到从 input_tensor 到 output_tensor 的路径,将其加入到网络中。
多输入模型
假设一个模型在进行一个简化版的图灵测试,这个模型要回答一个问题,这个问题的答案放在一篇文章中也由这个模型去处理,那么这个模型就是拥有多个输入的一个网络,如下:
其中有许多的知识我们之前讨论过了,如 LSTM 是循环神经网络,用于处理文本。这里的 Concatenate 需要解释一下:Concatenate 用于连接,可以将输入的问题和答案文本相关联起来。用代码构建的网络进行 summary 查看是这个样子的:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 import time import kerasimport numpy as np from keras import Inputfrom keras import layersfrom keras.models import Model def run (): text_vocabulary_size = 10000 question_vocabulary_size = 10000 answer_vocabulary_size = 500 text_input = Input(shape=(None ,), dtype='int32' , name='text' ) embedded_text = layers.Embedding(text_vocabulary_size, 64 )(text_input) encoded_text = layers.LSTM(32 )(embedded_text) question_input = Input(shape=(None ,), dtype='int32' , name='question' ) embedded_question = layers.Embedding(question_vocabulary_size, 32 )(question_input) encoded_question = layers.LSTM(16 )(embedded_question) concatenated = layers.concatenate([encoded_text, encoded_question], axis=-1 ) answer = layers.Dense(answer_vocabulary_size, activation='softmax' )(concatenated) model = Model([text_input, question_input], answer) model.summary() model.compile (optimizer='rmsprop' , loss='categorical_crossentropy' , metrics=['acc' ]) num_samples = 1000 max_length = 100 text = np.random.randint(1 , text_vocabulary_size, size=(num_samples, max_length)) question = np.random.randint(1 , question_vocabulary_size, size=(num_samples, max_length)) answers = np.random.randint(answer_vocabulary_size, size=(num_samples)) answers = keras.utils.to_categorical(answers, answer_vocabulary_size) model.fit([text, question], answers, epochs=10 , batch_size=128 ) model.fit({'text' : text, 'question' : question}, answers, epochs=10 , batch_size=128 ) if __name__ == "__main__" : time_start = time.time() run() time_end = time.time() print('Time Used: ' , time_end - time_start)
多输出模型
针对这个问题,实际用到的非常频繁,比如用户画像,我爸爸的百度首页就和我妈妈的不一样,虽然他们在设置里都没填写个人信息,但是还是知道他们分别喜欢什么,这是怎么做到的呢?
通过浏览的新闻,大致描述出用户的收入年龄等,再由后台对新闻的 Tag 标签,推送可能更感兴趣的新闻。注意这里的性别分类和收入分类就是不同的,一个是二分类,一个是多分类,这也需要定义不同的损失函数。想要完成这种任务,在对模型进行编译的时候,进行定义就是很好的办法。这里要注意,不平衡的损失贡献会影响模型的训练,如果某一项损失值特别大时,会导致模型会先对其进行优化而较少考虑或几乎不考虑其他任务的优化,这不是我们想看到的,因此我们还可以定义不同的损失对最终损失贡献的大小,进行损失加权:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 def run (): vocabulary_size = 50000 num_income_groups = 10 posts_input = Input(shape=(None ,), dtype='int32' , name='posts' ) embedded_posts = layers.Embedding(256 , vocabulary_size)(posts_input) x = layers.Conv1D(128 , 5 , activation='relu' )(embedded_posts) x = layers.MaxPooling1D(5 )(x) x = layers.Conv1D(256 , 5 , activation='relu' )(x) x = layers.Conv1D(256 , 5 , activation='relu' )(x) x = layers.MaxPooling1D(5 )(x) x = layers.Conv1D(256 , 5 , activation='relu' )(x) x = layers.Conv1D(256 , 5 , activation='relu' )(x) x = layers.GlobalMaxPooling1D()(x) x = layers.Dense(128 , activation='relu' )(x) age_prediction = layers.Dense(1 , name='age' )(x) income_prediction = layers.Dense(num_income_groups, activation='softmax' , name='income' )(x) gender_prediction = layers.Dense(1 , activation='sigmoid' , name='gender' )(x) model = Model(posts_input, [age_prediction, income_prediction, gender_prediction]) model.summary() model.compile (optimizer='rmsprop' , loss={'age' : 'mse' , 'income' : 'categorical_crossentropy' , 'gender' : 'binary_crossentropy' }, loss_weights={'age' : 0.25 , 'income' : 1. , 'gender' : 10. })
图结构
当然,这里的图结构指的是有向无环图,层与层之间无环,层内部可以有环(循环神经网络),前文提到了,层就像函数,张量是输入也是输出,那么就可以由这些向量像搭积木一样组合复杂的网络图结构由上面多输入和多输出的例子,你应该也大致知道应该怎么组合了吧,这里给出两个非常著名的例子,如下:
Inception 模块
本模块有助于学习空间特征和每个通道的特征,这比联合学习这两种特征更加有效。本模型内置在 keras.applications.inception_v3.InceptionV3 中,对于处理 ImageNet 数据集有很好的效率和准确度。
1 2 3 4 5 6 7 8 9 branch_a = layers.Conv2D(128 , 1 , activation='relu' , strides=2 )(x) branch_b = layers.Conv2D(128 , 1 , activation='relu' )(x) branch_b = layers.Conv2D(128 , 3 , activation='relu' , strides=2 )(branch_b) branch_c = layers.AveragePooling2D(3 , strides=2 )(x) branch_c = layers.Conv2D(128 , 3 , activation='relu' )(branch_c) branch_d = layers.Conv2D(128 , 1 , activation='relu' )(x) branch_d = layers.Conv2D(128 , 3 , activation='relu' )(branch_d) branch_d = layers.Conv2D(128 , 3 , activation='relu' , strides=2 )(branch_d) output = layers.concatenate([branch_a, branch_b, branch_c, branch_d], axis=-1 )
残差连接
本模块有助于解决梯度消失(反馈信号传递到更底部的层时,信号非常微弱)和表示瓶颈(某些层的参数过少,会导致某些特征永久消失,后面的层无法获取特征信息)的问题,具体的做法就是让前面某层的输出作为后面某层的输入:
1 2 3 4 5 x = ... y = layers.Conv2D(128 , 3 , activation='relu' , padding='same' )(x) y = layers.Conv2D(128 , 3 , activation='relu' , padding='same' )(y) y = layers.Conv2D(128 , 3 , activation='relu' , padding='same' )(y) y = layers.add([y, x])
共享层权重
函数的特点是复用,可以多次调用,这里的函数式编程,也是一样,一个层定义了一次,可以多次复用,这个很好理解:
1 2 3 4 5 6 7 8 9 10 11 lstm = layers.LSTM(32 ) left_input = Input(shape=(None , 128 )) left_output = lstm(left_input) right_input = Input(shape=(None , 128 )) right_output = lstm(right_input) merged = layers.concatenate([left_output, right_output], axis=-1 ) predictions = layers.Dense(1 , activation='sigmoid' )(merged) model = Model([left_input, right_input], predictions)
这是层的复用,扩大范围,模型也可以是输出向量,因此模型也是可以当做一个层的,这与前面文章的预定义网络有异曲同工之妙:
1 2 3 4 5 6 7 xception_base = applications.Xception(weights=None , include_top=False ) left_input = Input(shape=(250 , 250 , 3 )) right_input = Input(shape=(250 , 250 , 3 )) left_features = xception_base(left_input) right_input = xception_base(right_input) merged_features = layers.concatenate([left_features, right_input], axis=-1 )
总结
本来我的每篇文章都控制在 1000 字左右,但是本篇文章实在是有点收不住了,他们之间的联系太紧密了,没办法,那就算一篇长文。其中的内容是一脉相承的,主要记住一点:因为这种函数式的 API,所以就可以构建复杂的网络结构,而不仅仅是单纯的网络层的堆叠,基于此,可以构建出多输入、多输出以及各种情况复杂的网络,并且因为这种函数式的特点,因此有各种复用的情况。