2017-07-為什么選

為什么選高移植性在一個(gè)/CPUs或GPUs的筆記本電腦，服靈活性適用AndroidWindowsiOSLinux20,000commits6000TF-reposin1 七

為什么選使用Tensorflow的公●七

importtensorflowas

如果有同學(xué)想要更簡(jiǎn)單的用TFLearn(tf.contrib.learn):習(xí)慣scikit-learnfit函數(shù) TFSlimtf.contrib.slim這是TensorFlow里的七

概念Graph和DataFlowGraphby七

概念Graph和Graphby七

Tensor+Graphby

Tensor是什在tensorflow里，大家可以理解成一個(gè)n0-dtensor標(biāo)量/數(shù)scalarnumber)1-dtensor:向量vector2-dtensor…

Numpyvs

在Tensorflow的計(jì)算圖里importtensorflowasx,y是什么Tensorflow在你沒(méi)有指定名稱(chēng)的時(shí)候會(huì)自動(dòng)命3a5x=3a5y=在“這幅圖”節(jié)點(diǎn):operators,variables,and邊

3a5在Tensorflow的計(jì)算圖3a5importtensorflowasa=tf.add(3,print>>Tensor("Add:0",shape=(),結(jié)果不是如何取到結(jié)果

初始化session，完成操importtensorflowasa=tf.add(3,5)sess=tf.Session()printsess.run(a)Session會(huì)在計(jì)算圖里找到a的依賴(lài)，把依賴(lài)的節(jié)點(diǎn)都進(jìn)行計(jì)

建議的session寫(xiě)法如importtensorflowasa=tf.add(3,sess=withtf.Session()asprintSession會(huì)在計(jì)算圖里找到a的依賴(lài)，把依賴(lài)的節(jié)點(diǎn)都進(jìn)行計(jì)

我們來(lái)看一個(gè)復(fù)雜一點(diǎn)點(diǎn)的例x2y3=tf.add(x,=tf.mul(x,=tf.pow(op2,withtf.Session()asop3=

我們不需要的依賴(lài)，其實(shí)不會(huì)進(jìn)行計(jì)算，比如下面的例x=y=add_op=tf.add(x,y)mul_op=tf.mul(x,y)useless=tf.mul(x,add_op)pow_op=tf.pow(add_op,withtf.Session()asz=

如果我需要運(yùn)行幾個(gè)運(yùn)算節(jié)點(diǎn)，比如上例中的x=y=add_op=tf.add(x,y)mul_op=tf.mul(x,y)useless=tf.mul(x,add_op)pow_op=tf.pow(add_op,withtf.Session()asz,not_useless=sess.run([pow_op,

更全的格式是下面這樣，我們把所有需要的變量編成list放到fetchesx=y=…tf.Session.run(fetches,options=None,

指定CPUorGPU去完成session里的運(yùn)算構(gòu)建witha=tf.constant([1.0,2.0,3.0,4.0,5.0,6.0],b=tf.constant([1.0,2.0,3.0,4.0,5.0,6.0],name='b')=tf.matmul(a,#構(gòu)建session，設(shè)置log_device_placement為sess=運(yùn)行op定義的運(yùn)print

回想useless例

了解圖結(jié)構(gòu)/可視化的利importtensorflowasa=b=x=tf.add(a,withtf.Session()aswriter=tf.summary.FileWriter('./graphs,printsess.run(x)writer.close(關(guān)閉writer

了解圖結(jié)構(gòu)/$python 瀏覽 七

看這24了解圖結(jié)構(gòu)/可視化的利importtensorflowasa=b=x=tf.add(a,withtf.Session()aswriter=tf.summary.FileWriter('./graphs,printsess.run(x)writer.close(關(guān)閉writer

自定義常量名稱(chēng)可以自己指importtensorflowasa=tf.constant(2,b=tf.constant(3,name=“b")x=tf.add(a,b,name=“add”)withtf.Session()assess:writer=tf.summary.FileWriter('./graphs,sess.graph)printsess.run(x)writer.close(關(guān)閉

和numpy其實(shí)tf.constant(value,dtype=None,name='Const',importtensorflowasa=tf.constant([2,2],b=tf.constant([[0,1],[2,3]],x=tf.add(a,b,name="add")y=tf.mul(a,b,name="mul")withtf.Session()assess:x,y=sess.run([x,printx,#>>[58][6

七

tf.random_normal(shape,mean=0.0,dtype=tf.float32,seed=None,tf.truncated_normal(shape,mean=0.0,dtype=tf.float32,seed=None,tf.random_uniform(shape,minval=0,dtype=tf.float32,seed=None,name=None)tf.random_shuffle(value,seed=None,name=None)tf.random_crop(value,size,seed=None,name=None)tf.multinomial(logits,num_samples,seed=None,name=None)tf.random_gamma(shape,alpha,beta=None,dtype=tf.float32,seed=None,七

a=tf.constant([3,b=tf.constant([2,tf.add(a,b)#>>[5tf.add_n([a,b,b])#>>[7tf.mul(a,b)#>>[6tf.matmul(a,b)#>>tf.matmul(tf.reshape(a,[1,2]),tf.reshape(b,[2,1]))#>>tf.div(a,b)#>>[1tf.mod(a,b)#>>[1七

七

tf.constant是op，而tf.Variable是一個(gè)類(lèi)，初始化的對(duì)象有多個(gè)#createvariableawithscalarvaluea=tf.Variable(2,name="scalar")#createvariablebasab=#createvariable

3],asa2x2c=tf.Variable([[0,1],[2,3]],#createvariableWas784x10tensor,filledwithW=七

tf.constant是op，而tf.Variable是一個(gè)類(lèi)，初始化的對(duì)象有多個(gè)x=tf.Variable(...)x.initializer初始化x.value 的x.assign(..寫(xiě)入的x.assign_add(...)七

init=withtf.Session()asinit_ab=tf.variables_initializer([a,b],withtf.Session()aswithtf.Session()assess:

W是一個(gè)700x100隨W=tf.Variable(tf.truncated_normal([700,withtf.Session()assess:printWprint>>Tensor("Variable/read:0",shape=(700, 七

Tensorflowtf.placeholder(dtype,shape=None,#createaplaceholderoftypefloat32-‐bit,shapeisavectorof3a=tf.placeholder(tf.float32,#createaconstantoftypefloat32-‐bit,shapeisavectorof3b=tf.constant([5,5,5],#usetheplaceholderasyouwouldaconstantorac=a+b#Shortfortf.add(a,withtf.Session()assess:printsess.run(c)#Errorbecauseadoesn’thaveany

Tensorflowtf.placeholder(dtype,shape=None,#createaplaceholderoftypefloat32-‐bit,shapeisavectorof3a=tf.placeholder(tf.float32,#createaconstantoftypefloat32-‐bit,shapeisavectorof3b=tf.constant([5,5,5],#usetheplaceholderasyouwouldaconstantorac=a+b#Shortfortf.add(a,withtf.Session()as#feed[1,2,3]toplaceholderaviathedict{a:[1,2,printsess.run(c,{a:[1,2,#thetensoraisthekey,notthestring#>>[6,7,

Tensorflowtf.placeholder(dtype,shape=None,#createoperations,tensors,etc(usingthed