notebook.community

Edit and run



In [11]:

    
import numpy as np
#source ~/tensorflow/bin/activate
import tensorflow as tf
import seaborn









    



:0: FutureWarning: IPython widgets are experimental and may change in the future.



In [4]:

    
a = np.array([1,2,3])



In [4]:

    
a









    Out[4]:





array([1, 2, 3])



In [5]:

    
a.shape









    Out[5]:





(3,)



In [7]:

    
b = np.array([[1,2,3],[4,5,6]])



In [8]:

    
b









    Out[8]:





array([[1, 2, 3],
       [4, 5, 6]])



In [9]:

    
b.shape









    Out[9]:





(2, 3)



In [10]:

    
a = np.zeros((2,2))



In [11]:

    
a









    Out[11]:





array([[ 0.,  0.],
       [ 0.,  0.]])



In [12]:

    
b = np.ones((1,2))



In [13]:

    
b









    Out[13]:





array([[ 1.,  1.]])



In [14]:

    
b.shape









    Out[14]:





(1, 2)



In [15]:

    
b = np.array([1,1])



In [16]:

    
b









    Out[16]:





array([1, 1])



In [17]:

    
b.shape









    Out[17]:





(2,)



In [18]:

    
b = np.array([[1,1]])



In [19]:

    
b









    Out[19]:





array([[1, 1]])



In [20]:

    
b.shape









    Out[20]:





(1, 2)



In [21]:

    
a = np.array([[1,2,3,4], [5,6,7,8], [9,10,11,12]])



In [22]:

    
a









    Out[22]:





array([[ 1,  2,  3,  4],
       [ 5,  6,  7,  8],
       [ 9, 10, 11, 12]])



In [23]:

    
b = a[:2, 1:3] #righe 0 e 1, colonne 1 e 2



In [24]:

    
b









    Out[24]:





array([[2, 3],
       [6, 7]])



In [25]:

    
a = np.array([[1,2,3,4], [5,6,7,8], [9,10,11,12]])



In [26]:

    
row_r1 = a[1, :] #indexing reduces the rank



In [27]:

    
row_r1









    Out[27]:





array([5, 6, 7, 8])



In [28]:

    
row_r2 = a[1:2, :] #slicing keep the same rank



In [29]:

    
row_r2









    Out[29]:





array([[5, 6, 7, 8]])



In [34]:

    
col_r1 = a[:, 1] #second col



In [35]:

    
col_r1









    Out[35]:





array([ 2,  6, 10])



In [36]:

    
col_r1.shape









    Out[36]:





(3,)



In [38]:

    
col_r2 = a[:, 1:2]



In [39]:

    
col_r2









    Out[39]:





array([[ 2],
       [ 6],
       [10]])



In [40]:

    
col_r2.shape









    Out[40]:





(3, 1)



In [41]:

    
a = np.array([[1,2], [3, 4], [5, 6]])



In [42]:

    
a









    Out[42]:





array([[1, 2],
       [3, 4],
       [5, 6]])



In [43]:

    
a.shape









    Out[43]:





(3, 2)



In [47]:

    
print a[[0, 1, 2], [0, 1, 0]] #il primo set da' l'indice delle righe, il secondo delle colonne



In [50]:

    
x = np.array([[1,2],[3,4]], dtype=np.float64)



In [52]:

    
x.ndim









    Out[52]:





2



In [53]:

    
a = np.array([0, 1, 2])



In [62]:

    
a









    Out[62]:





array([0, 1, 2])



In [63]:

    
a.shape









    Out[63]:





(3,)



In [64]:

    
a1 = np.tile(a, 2)



In [65]:

    
a1.shape









    Out[65]:





(6,)



In [66]:

    
a2 = np.tile(a, (2, 2))



In [67]:

    
a2









    Out[67]:





array([[0, 1, 2, 0, 1, 2],
       [0, 1, 2, 0, 1, 2]])



In [68]:

    
a2.shape









    Out[68]:





(2, 6)



In [70]:

    
a.ndim









    Out[70]:





1



In [71]:

    
a3 = np.tile(a, (2, 1, 2))



In [72]:

    
a3









    Out[72]:





array([[[0, 1, 2, 0, 1, 2]],

       [[0, 1, 2, 0, 1, 2]]])



In [73]:

    
a3.ndim









    Out[73]:





3



In [74]:

    
a3.shape









    Out[74]:





(2, 1, 6)



In [78]:

    
b = np.array([[1, 2], [3, 4]])
b









    Out[78]:





array([[1, 2],
       [3, 4]])



In [76]:

    
b.shape









    Out[76]:





(2, 2)



In [79]:

    
b1 = np.tile(b, 2)



In [84]:

    
b1.shape









    Out[84]:





(2, 4)



In [81]:

    
b2 = np.tile(b, (2, 1))



In [82]:

    
b2









    Out[82]:





array([[1, 2],
       [3, 4],
       [1, 2],
       [3, 4]])



In [83]:

    
b2.shape









    Out[83]:





(4, 2)



In [85]:

    
x = np.array([[[1],[2],[3]], [[4],[5],[6]]])



In [86]:

    
x









    Out[86]:





array([[[1],
        [2],
        [3]],

       [[4],
        [5],
        [6]]])



In [87]:

    
x.shape









    Out[87]:





(2, 3, 1)



In [5]:

    
x1 = np.array([[[1,2,3], [4,5,6]], [[1,2,3], [4,5,6]]])



In [6]:

    
x1









    Out[6]:





array([[[1, 2, 3],
        [4, 5, 6]],

       [[1, 2, 3],
        [4, 5, 6]]])



In [7]:

    
x1.shape









    Out[7]:





(2, 2, 3)



In [96]:

    
x1.ndim









    Out[96]:





3



In [9]:

    
x1.shape[2]









    Out[9]:





3



In [98]:

    
x1 = np.array([1, 2, 3, 4, 5])
x2 = np.array([5, 4, 3])



In [99]:

    
x1









    Out[99]:





array([1, 2, 3, 4, 5])



In [100]:

    
x2









    Out[100]:





array([5, 4, 3])



In [101]:

    
x1_new = x1[:, np.newaxis]



In [102]:

    
x1_new









    Out[102]:





array([[1],
       [2],
       [3],
       [4],
       [5]])



In [104]:

    
x3 = x1_new + x2



In [106]:

    
x3









    Out[106]:





array([[ 6,  5,  4],
       [ 7,  6,  5],
       [ 8,  7,  6],
       [ 9,  8,  7],
       [10,  9,  8]])



In [4]:

    
tf1 = tf.zeros([2, 3])



In [6]:

    
tf1.shape









    Out[6]:





TensorShape([Dimension(2), Dimension(3)])



In [7]:

    
tf2 = tf.ones([2, 3,4])



In [8]:

    
tf2.shape









    Out[8]:





TensorShape([Dimension(2), Dimension(3), Dimension(4)])



In [9]:

    
tf2









    Out[9]:





<tf.Tensor 'ones:0' shape=(2, 3, 4) dtype=float32>



In [12]:

    
tf3 = tf.placeholder(tf.float32, shape=[None,3], name="train_inputs")



In [13]:

    
tf3









    Out[13]:





<tf.Tensor 'train_inputs:0' shape=(?, 3) dtype=float32>



In [9]:

    
np.zeros([2, 3] +
         [3 * (3 + 1) / 2], dtype=np.float32)









    Out[9]:





array([[[ 0.,  0.,  0.,  0.,  0.,  0.],
        [ 0.,  0.,  0.,  0.,  0.,  0.],
        [ 0.,  0.,  0.,  0.,  0.,  0.]],

       [[ 0.,  0.,  0.,  0.,  0.,  0.],
        [ 0.,  0.,  0.,  0.,  0.,  0.],
        [ 0.,  0.,  0.,  0.,  0.,  0.]]], dtype=float32)



In [16]:

    
a = np.zeros((2,2))



In [17]:

    
a









    Out[17]:





array([[ 0.,  0.],
       [ 0.,  0.]])



In [18]:

    
a.shape









    Out[18]:





(2, 2)



In [ ]:



In [5]:

    
b = np.ones((2,2))



In [6]:

    
b









    Out[6]:





array([[ 1.,  1.],
       [ 1.,  1.]])



In [13]:

    
b1 = np.sum(b, axis=0)



In [14]:

    
b1









    Out[14]:





array([ 2.,  2.])



In [15]:

    
b1.shape









    Out[15]:





(2,)



In [2]:

    
tf.InteractiveSession()









    Out[2]:





<tensorflow.python.client.session.InteractiveSession at 0x7f207c076850>



In [3]:

    
a = tf.zeros((2,2))



In [4]:

    
a









    Out[4]:





<tf.Tensor 'zeros:0' shape=(2, 2) dtype=float32>



In [5]:

    
b = tf.ones((2,2))



In [6]:

    
b









    Out[6]:





<tf.Tensor 'ones:0' shape=(2, 2) dtype=float32>



In [7]:

    
tf.reduce_sum(b, reduction_indices=1).eval()









    Out[7]:





array([ 2.,  2.], dtype=float32)



In [9]:

    
a.get_shape()









    Out[9]:





TensorShape([Dimension(2), Dimension(2)])



In [12]:

    
ta = tf.zeros((2,2))



In [15]:

    
print(ta)









    Out[15]:





<tf.Tensor 'zeros_1:0' shape=(2, 2) dtype=float32>



In [16]:

    
print(ta.eval())









    



[[ 0.  0.]
 [ 0.  0.]]



In [29]:

    
b2 = tf.reduce_sum(b, reduction_indices=1).eval()[0:1]



In [30]:

    
b2









    Out[30]:





array([ 2.], dtype=float32)



In [38]:

    
a = tf.constant(5.0)
b = tf.constant(6.0)



In [39]:

    
a









    Out[39]:





<tf.Tensor 'Const_2:0' shape=() dtype=float32>



In [32]:

    
c = a*b



In [33]:

    
c  #ho creato l'oggetto ma non l'ho ancora valutato









    Out[33]:





<tf.Tensor 'mul:0' shape=() dtype=float32>



In [37]:

    
with tf.Session() as sess:
    print(sess.run(c))
    print(c.eval())



In [42]:

    
W1 = tf.ones((2,2))  #this is a constant that still need to be executed
W2 = tf.Variable(tf.zeros((2,2)), name="weights") 
W1
W2









    Out[42]:





<tensorflow.python.ops.variables.Variable at 0x7f205c58b910>



In [43]:

    
with tf.Session() as sess:
        print(sess.run(W1)) #running the constant
        sess.run(tf.initialize_all_variables())
        print(sess.run(W2)) #running the variable









    



[[ 1.  1.]
 [ 1.  1.]]
WARNING:tensorflow:From <ipython-input-43-e74ec20ca67b>:3: initialize_all_variables (from tensorflow.python.ops.variables) is deprecated and will be removed after 2017-03-02.
Instructions for updating:
Use `tf.global_variables_initializer` instead.






    



WARNING:tensorflow:From <ipython-input-43-e74ec20ca67b>:3: initialize_all_variables (from tensorflow.python.ops.variables) is deprecated and will be removed after 2017-03-02.
Instructions for updating:
Use `tf.global_variables_initializer` instead.






    



[[ 0.  0.]
 [ 0.  0.]]



In [2]:

    
# Regression in tensorflow
X_data = np.arange(100, step=.1)
y_data = X_data + 20 * np.sin(X_data/10)



In [3]:

    
X_data.shape









    Out[3]:





(1000,)



In [4]:

    
y_data.shape









    Out[4]:





(1000,)



In [5]:

    
n_samples = 1000 #ho 1000 osservazioni per una variabile unidimensionale
batch_size = 1000



In [6]:

    
# Tensorflow is finicky about shapes, so resize
X_data = np.reshape(X_data, (n_samples,1))
X_data.shape









    Out[6]:





(1000, 1)



In [7]:

    
y_data = np.reshape(y_data, (n_samples,1))



In [8]:

    
y_data.shape









    Out[8]:





(1000, 1)



In [9]:

    
# Define placeholders for input - creo degli element che 
# conterranno i dati
# considero solamente 100 dati per volta (batch_size)
X = tf.placeholder(tf.float32, shape=(batch_size, 1))
y = tf.placeholder(tf.float32, shape=(batch_size, 1))



In [10]:

    
X









    Out[10]:





<tf.Tensor 'Placeholder:0' shape=(1000, 1) dtype=float32>



In [11]:

    
y









    Out[11]:





<tf.Tensor 'Placeholder_1:0' shape=(1000, 1) dtype=float32>



In [12]:

    
# Define variables to be learned
with tf.variable_scope("linear-regression"):
    # il coefficiente
    W = tf.get_variable("weights", (1, 1),
                        initializer=tf.random_normal_initializer())
    # l'intercetta
    b = tf.get_variable("bias", (1,),
                        initializer=tf.constant_initializer(0.0))
    y_pred = tf.matmul(X, W) + b
    # funzione da minimizzare
    loss = tf.reduce_sum((y - y_pred)**2/n_samples)



In [13]:

    
# Ottimizzazione rispetto alle variabile con i dati assegnati 
# ai placeholder
# specifico il metodo di ottimizzazione: fai train con AdamOptimizer
opt = tf.train.AdamOptimizer()



In [14]:

    
opt_operation = opt.minimize(loss) #l'operazione di ottimiz.



In [31]:

    
#devo runnare all'interno di una sessione
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    #print(sess.run(y_pred, feed_dict={X: X_data}))
    #print(sees.run(loss, feed_dict={X: X_data, y: y_data}))
    for i in range(500):
        sess.run([opt_operation], feed_dict={X: X_data, y: y_data})
    # Il comando deva stare cosi indentato perchè è parte della
    # sessione che apro con with
    curr_W, curr_b, curr_loss = sess.run([W,b,loss], 
                                         feed_dict={X: X_data, y: y_data})



In [32]:

    
print("W: %s b: %s loss: %s"%(curr_W, curr_b, curr_loss))









    



W: [[-0.03080422]] b: [ 0.46191812] loss: 4001.26



In [23]:

    
#print(sess.run([W,b, loss])
opt_operation









    Out[23]:





<tf.Operation 'Adam' type=NoOp>



In [8]:

    
print(opt_operation)









    



---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-8-02322271820a> in <module>()
----> 1 print(opt_operation)

NameError: name 'opt_operation' is not defined



In [9]:

    
x1 = tf.sqrt(2.0)



In [10]:

    
x1









    Out[10]:





<tf.Tensor 'Sqrt_4:0' shape=() dtype=float32>



In [15]:

    
x1 = tf.ones([2, 3] +
         [3 * (3 + 1) / 2], dtype=np.float32)



In [16]:

    
x1.shape









    Out[16]:





TensorShape([Dimension(2), Dimension(3), Dimension(6)])



In [18]:

    
x1[1, :, :]









    Out[18]:





<tf.Tensor 'strided_slice_1:0' shape=(3, 6) dtype=float32>



In [ ]: