‘îŁñ›È−wﬁÁŁÊ”À„±I --Tensorflow‡ð”g‡Á‡½ﬂ»ŁÊ›ð’Í--

Size: px

Start display at page:

Download "‘îŁñ›È−wﬁÁŁÊ”À„±I --Tensorflow‡ð”g‡Á‡½ﬂ»ŁÊ›ð’Í--"

もえりやたけ
5 years ago
Views:

1 I Tensorflow ( ) ( ) Tensorflow ( ) 1 / 39

2 Tensorflow I Tensorflow Python Python(+Python Anaconda) Tensorflow Tensorflow 1 Anaconda Prompt 2 Anaconda Prompt (base) C:\Users\komori>conda create -n tensorflow python=3.5 3 tensorflow (base) C:\Users\komori>activate tensorflow 4 pip(pip installs Packages ) tensorflow (base) C:\Users\komori>pip install --ignore-installed --upgrade tensorflow cp35-cp35m-win_amd64.whl 5 idle (base) C:\Users\komori>idle ( ) Tensorflow ( ) 2 / 39

3 Tensorflow I Tensorflow Google C++ Python (flow) Google Tensorflow ( ) Tensorflow ( ) 3 / 39

4 Tensorflow II ( ) Tensorflow ( ) 4 / 39

5 Tensor I Tensor Tensorflow (tensor) (flow) tensor >>> import numpy as np # python >>> M1=np.array([1,2,3]) # >>> M1 array([1, 2, 3]) >>> M1.ndim # M1 1 >>> M1.shape # M1 (3,) >>> M2=np.array([[1,2,3],[4,5,6]]) # >>> M2 array([[1, 2, 3], [4, 5, 6]]) >>> M2.ndim # M2 2 >>> M2.shape # M2 (2, 3) ( ) Tensorflow ( ) 5 / 39

6 Tensor II tensor( ) >>> M3=np.array([[[1,2,3],[4,5,6]],[[7,8,9],[10,11,12]],[[1,1,1],[1,1,1]],[[2,2,2],[ >>> M3 # array([[[ 1, 2, 3], [ 4, 5, 6]], [[ 7, 8, 9], [10, 11, 12]], [[ 1, 1, 1], [ 1, 1, 1]], [[ 2, 2, 2], [ 2, 2, 2]]]) >>> M3.ndim # M3 3 >>> M3.shape # M3 (4, 2, 3) ( ) Tensorflow ( ) 6 / 39

7 Tensor III ( ) Tensorflow ( ) 7 / 39

8 Tensorflow I Tensorflow 2 Tensorflow 1 (dataflow graph) 2 Tensorflow (session) node( ): (operation, op), tensor( ) edge( ): tensor( ) ( ) Tensorflow ( ) 8 / 39

9 Tensorflow II tensor ( tensor!) tensor constant: tensor( ) Variable: tensor( ) trainable( ) non-trainable( ) w sess.run(w.initializer), sees.run(tf.global variables initializer()) placeholder: tensor. session feed dict={ } tf.initialize all variables 2017 ( ) Tensorflow ( ) 9 / 39

10 Tensorflow I Tensorflow activate tensorflow activate C:\\Users\\komori\\Desktop>activate tensorflow (tensorflow) tensorflow activate (tensorflow)c:\\users\\komori\\desktop> python (idle: integrated development environment) python (tensorflow)c:\\users\\komori\\desktop>idle python a.py python a.py ( ) Tensorflow ( ) 10 / 39

11 Session I Session (operation) (tensor) >>> import tensorflow as tf # tensorflow import tf # Build a graph. >>> a = tf.constant(5.0) >>> b = tf.constant(6.0) >>> c = tf.constant(7.0) >>> d = a * b + c # Launch the graph in a session. >>> sess = tf.session() # Evaluate the tensor d. >>> print(sess.run(d)) # 37.0 >>> sess.close() # session >>> print(d)#d node Tensor("mul:0", shape=(), dtype=float32) ( ) Tensorflow ( ) 11 / 39

12 I Tensorboard >>> import tensorflow as tf # tensorflow import tf # Build a graph. >>> a=tf.constant(5.0, name="a") # >>> b=tf.constant(6.0, name="b") # >>> c=tf.constant(7.0, name="c") >>> d=a * b + c >>> tf.summary.scalar( d, d) # d # Launch the graph in a session. >>> sess = tf.session() >>> writer=tf.summary.filewriter(./log, sess.graph) # log graph >>> sess.close() # session >>> writer.close() # writer tensorboard ( ) Tensorflow ( ) 12 / 39

13 II tensorboard (tensorflow) C:\\Users\\komori\\Desktop>tensorboard --logdir=./log C:\\Users\\komori\\Desktop>tasklist # PID, 2528 C:\\Users\\komori\\Desktop>taskkill /F /PID 2528 C:\\Users\\komori\\Desktop>help taskkill # help ( ) Tensorflow ( ) 13 / 39

14 Iris I pandas pandas C:\\Users\\komori\\Desktop>pip install pandas pip : Pip installs Packages, Pip installs Python pandas : panel data (x it, i = 1,..., n, t = 1,..., T), iris >>> import pandas as pd >>> iris=pd.read_csv("iris100.csv") >>> iris.iloc[0:3,] #iloc, integer location Sepal.Length Sepal.Width Petal.Length Petal.Width Species setosa setosa setosa ( ) Tensorflow ( ) 14 / 39

15 Iris II iris ( ) >>> iris.index Int64Index([ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 92, 93, 94, 95, 96, 97, 98, 99, 100], dtype= int64 ) >>> iris.columns Species ], dtype= object ) >>> iris.shape (100, 5) ( ) Tensorflow ( ) 15 / 39

16 Iris III matplotlib pandas C:\\Users\\komori\\Desktop>pip install matplotlib iris >>> import matplotlib.pyplot as plt #matplotlib pyplot plt >>> plt.scatter(iris.iloc[0:50,0],iris.iloc[0:50,1]) >>> plt.scatter(iris.iloc[49:100,0],iris.iloc[49:100,1]) >>> plt.xlabel(iris.columns[0]) >>> plt.ylabel(iris.columns[1]) >>> plt.show() ( ) Tensorflow ( ) 16 / 39

17 Iris I ( ) Tensorflow ( ) 17 / 39

18 tf.matmul I matmul ( ) >>> import tensorflow as tf >>> impoart numpy as np >>> X=np.array([[1,2,3,4],[5,6,7,8]]) >>> W=np.array([1,1,-1,-1]) # W: 1 4 >>> W.shape=(4,1) # 4 1 >>> A=tf.matmul(X,W) >>> with tf.session() as sess: #: print(sess.run(a)) [[-4] [-4]] with sess.close() ( ) ( ) Tensorflow ( ) 18 / 39

19 a I tf.assign update.py import tensorflow as tf a = tf.variable(0, name="a") b= tf.assign(a, a+1) with tf.session() as sess: sess.run(tf.global_variables_initializer()) #tf.variable print(sess.run(b)) for _ in range(3): # for _ print(sess.run(b)) tf.assign(ref,value): Update ref by assigning value to it. ( ) Tensorflow ( ) 19 / 39

20 (fetching) I sess.run() fetch.py import tensorflow as tf a=tf.constant(1) # default graph #a=1 # default graph ( ) b=2 c=3 d=a+b e=a*b+c with tf.session() as sess: res1, res2 = sess.run ([d,e]) # d,e fetch print(res1,res2) fetch: [ ] ( ) Tensorflow ( ) 20 / 39

21 I graph.py import tensorflow as tf g = tf.graph() #graph g with g.as_default():# Define operations and tensors in g. a=tf.constant(1) # graph g (tensor)a b=2 c=3 d=a+b e=a*b+c A=tf.constant(30) # default graph (tensor)a with tf.session(graph=g) as sess: print(sess.run([d,e])) with tf.session(graph=tf.get_default_graph()) as sess: print(sess.run(a)) default graph tf.get default graph() ( ) Tensorflow ( ) 21 / 39

22 placeholder I placeholder.py import tensorflow as tf import numpy as np x = tf.placeholder(tf.int32, shape=(3, 3)) # y = tf.matmul(x, x)+1 with tf.session() as sess: A = np.random.randint(10,size=(3,3)) # print(a) print(sess.run(y, feed_dict={x: A})) # (feed) np.random.randint() np.random.rand() np.random.normal() np.random.binomial() ( ) Tensorflow ( ) 22 / 39

23 tf.reduce sum I reduce.py import tensorflow as tf x = tf.constant([[1, 1, 1], [1, 1, 1]]) a1=tf.reduce_sum(x) a2=tf.reduce_sum(x, 0) a3=tf.reduce_sum(x, 1) a4=tf.reduce_sum(x, [0, 1]) with tf.session() as sess: x, a1, a2, a3, a4=sess.run([x,a1,a2,a3,a4]) print("x= %s,a1=%s, a2=%s, a3=%s, a4=%s" % (x,a1,a2,a3,a4)) x= [[1 1 1] [1 1 1]],a1=6, a2=[2 2 2], a3=[3 3], a4=6 reduce sum axis. %s: string cf %d cf. reduce prod ( ) Tensorflow ( ) 23 / 39

24 Iris I logistic.py import numpy as np import pandas as pd import tensorflow as tf iris=pd.read_csv("iris100.csv") # data=iris.iloc[:,0:4] #4 X=tf.placeholder(tf.float32,[None,4]) W=tf.Variable(tf.zeros([4,1])) w0=tf.variable(tf.zeros(1)) Fx=tf.matmul(X,W)+w0 #Fx Fx=tf.reshape(Fx,shape=[100]) #Fx prob=tf.sigmoid(fx) #Fx prob y=tf.concat([tf.zeros(50),tf.ones(50)],0) #iris tf.concat([t1,t2], axis) tf.sigmoid: y = 1 / (1 + exp(-x)) y: setosa 0, sersicolor 1 ( ) Tensorflow ( ) 24 / 39

25 Iris II logistic.py( ) likelihood=tf.reduce_sum(y*tf.log(prob)+(1-y)*tf.log(1-prob)) loss=-likelihood learning_rate=0.001 # train= tf.train.gradientdescentoptimizer(learning_rate).minimize(loss) likelihood: loss 2 (tf.reduce mean(tf.square(y-prob))) (learning rate=0.1 ) η: W = W η loss W Optimizer GradientDescentOptimizer: AdagradOptimize: AdaGrad MomentumOptimizer: AdamOptimizer: Adam ( ) Tensorflow ( ) 25 / 39

26 Iris III Session logistic.py( ) with tf.session() as sess: sess.run(tf.global_variables_initializer()) #Variables for i in range(100): #100 sess.run(train,feed_dict={x:data}) if i % 10 == 0: #10 print("%5d:(loss,w,w0)=(%f, %s,%.4f)" % (i, sess.run(loss,feed_dict={x:data}), sess.run(tf.reshape(w,shape=[4])), sess.run(w0))) print("%d:prob=%s" % (i, sess.run([prob[0:50],prob[49:100]], feed_dict={x:data}))) sess.run(train,feed dict={x:data}) ( ) Tensorflow ( ) 26 / 39

27 Iris IV 99:prob=[array([ , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ], dtype=float32), array([ , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ], dtype=float32)] ( ) Tensorflow ( ) 27 / 39

28 (name scope, name space) I Tensorboard tensorflow tf.name scope tf.name scope with with tf.variable_scope("probability"): with tf.name_scope("x"): X=tf.placeholder(tf.float32,[None,4]) with tf.name_scope("w"): W=tf.Variable(tf.zeros([4,1])) with tf.name_scope("w0"): w0=tf.variable(tf.zeros(1)) with tf.name_scope("fx"): Fx=tf.matmul(X,W)+w0 Fx=tf.reshape(Fx,shape=[100]) with tf.name_scope("prob"): prob=tf.sigmoid(fx) W W.name ( ) Tensorflow ( ) 28 / 39

29 (name scope, name space) II >>> print(w.name,w0.name) probability/w/variable:0 probability/w0/variable:0 0 index. 2 0 index >>> a,b=tf.nn.top_k([1,4,2], 1) # k index >>> a.name TopKV2_2:0 >>> b.name TopKV2_2:1 ( ) Tensorflow ( ) 29 / 39

30 (summary) I tf.summary tensorboard with tf.name_scope( summary ): tf.summary.scalar("loss",loss) tf.summary.scalar("mean.w",tf.reduce_mean(w)) tf.summary.histogram("hist.w",w) (sess.run ) merged = tf.summary.merge_all() ( ) Tensorflow ( ) 30 / 39

31 (summary) II log (logistic.writer.py) log_dir = C:\Users\komori\Desktop\log #log path with tf.session() as sess: sess.run(tf.global_variables_initializer()) #Variable writer = tf.summary.filewriter(log_dir, sess.graph) # log print(sess.run([loss,y],feed_dict={x:data})) for i in range(100): #100 summary,_=sess.run([merged,train],feed_dict={x:data}) writer.add_summary(summary, i) #merged summary log writer.close() #writer tensorboard (tensorflow) C:\Users\komori\Desktop>tensorboard --logdir="./log" ( ) Tensorflow ( ) 31 / 39

32 Tensorboard (scalars) I ( ) Tensorflow ( ) 32 / 39

33 Tensorboard (distributions) I ( ) Tensorflow ( ) 33 / 39

34 Tensorboard (histgrams) I ( ) Tensorflow ( ) 34 / 39

35 Tensorboard (graph) I ( ) Tensorflow ( ) 35 / 39

36 Tensorboard (graph ) I ( ) Tensorflow ( ) 36 / 39

37 I Tensorflow Tensorflow R pandas ( ) Tensorflow ( ) 37 / 39

38 I logistic.writer.py import numpy as np import pandas as pd import tensorflow as tf iris=pd.read_csv("iris100.csv") data=iris.iloc[:,0:4] with tf.variable_scope("probability"): with tf.name_scope("x"): X=tf.placeholder(tf.float32,[None,4]) with tf.name_scope("w"): W=tf.Variable(tf.zeros([4,1])) WW=tf.Variable(tf.zeros([4,1])) with tf.name_scope("w0"): w0=tf.variable(tf.zeros(1)) with tf.name_scope("fx"): Fx=tf.matmul(X,W)+w0 Fx=tf.reshape(Fx,shape=[100]) with tf.name_scope("prob"): prob=tf.sigmoid(fx) with tf.name_scope("label"): y=tf.concat([tf.zeros(50),tf.ones(50)],0) ( ) Tensorflow ( ) 38 / 39

39 II logistic.writer.py( ) with tf.name_scope("loss"): likelihood=tf.reduce_sum(y*tf.log(prob)+(1-y)*tf.log(1-prob)) loss=-likelihood learning_rate=0.001 train= tf.train.gradientdescentoptimizer(learning_rate).minimize(loss) with tf.name_scope( summary ): tf.summary.scalar("loss",loss) tf.summary.scalar("mean.w",tf.reduce_mean(w)) tf.summary.histogram("hist.w",w) merged = tf.summary.merge_all() log_dir = C:\Users\komori\Desktop\log with tf.session() as sess: sess.run(tf.global_variables_initializer()) writer = tf.summary.filewriter(log_dir, sess.graph) print(sess.run([loss,y],feed_dict={x:data})) for i in range(100): summary,_=sess.run([merged,train],feed_dict={x:data}) writer.add_summary(summary, i) writer.close() ( ) Tensorflow ( ) 39 / 39

Visual Python, Numpy, Matplotlib

Visual Python, Numpy, Matplotlib 1 / 57 Contents 1 2 Visual Python 3 Numpy Scipy 4 Scipy 5 Matplotlib 2 / 57 Contents 1 2 Visual Python 3 Numpy Scipy 4 Scipy 5 Matplotlib 3 / 57 3 Visual Python: 3D Numpy,