TensorFlow Cheat Sheet¶

In [1]:
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' 
# os.environ['PYTHONWARNINGS'] = 'ignore'
In [2]:
import tensorflow as tf
import numpy as np
In [3]:
%precision 2
ipython_plain = get_ipython().display_formatter.formatters['text/plain']
ipython_plain.for_type(np.float64, ipython_plain.lookup_by_type(float));

tf.Tensor¶

In [4]:
tf.constant(2, 'float32', [3])

Metal device set to: Apple M1 Max
Out[4]:
<tf.Tensor: shape=(3,), dtype=float32, numpy=array([2., 2., 2.], dtype=float32)>
In [5]:
tf.constant(np.random.uniform(size=[2,2]).astype('float32'))
Out[5]:
<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
array([[0.06, 0.09],
       [0.61, 0.82]], dtype=float32)>
In [6]:
t3 = tf.range(3)
t3
Out[6]:
<tf.Tensor: shape=(3,), dtype=int32, numpy=array([0, 1, 2], dtype=int32)>
In [7]:
t23 = tf.constant(np.array([[1,2,3], [4,5,6]]))
t23
Out[7]:
<tf.Tensor: shape=(2, 3), dtype=int64, numpy=
array([[1, 2, 3],
       [4, 5, 6]])>

tf.Variable¶

In [8]:
with tf.device('/CPU:0'):  # Apple TF-GPU bug
    var1 = tf.Variable(tf.constant(2, 'float32', [3]), name='var1')
var1
Out[8]:
<tf.Variable 'var1:0' shape=(3,) dtype=float32, numpy=array([2., 2., 2.], dtype=float32)>
In [9]:
var1.assign(var1 + 1)
var1
Out[9]:
<tf.Variable 'var1:0' shape=(3,) dtype=float32, numpy=array([3., 3., 3.], dtype=float32)>
In [10]:
var1[1]
Out[10]:
<tf.Tensor: shape=(), dtype=float32, numpy=3.0>
In [11]:
var1[1].assign(9)
var1
Out[11]:
<tf.Variable 'var1:0' shape=(3,) dtype=float32, numpy=array([3., 9., 3.], dtype=float32)>

tf.Operation¶

In [12]:
t23
Out[12]:
<tf.Tensor: shape=(2, 3), dtype=int64, numpy=
array([[1, 2, 3],
       [4, 5, 6]])>
In [13]:
t23 + 1
Out[13]:
<tf.Tensor: shape=(2, 3), dtype=int64, numpy=
array([[2, 3, 4],
       [5, 6, 7]])>
In [14]:
t23 > 2
Out[14]:
<tf.Tensor: shape=(2, 3), dtype=bool, numpy=
array([[False, False,  True],
       [ True,  True,  True]])>
In [15]:
tf.reduce_sum(t23)
Out[15]:
<tf.Tensor: shape=(), dtype=int64, numpy=21>
In [16]:
tf.reduce_sum(t23, axis=0)
Out[16]:
<tf.Tensor: shape=(3,), dtype=int64, numpy=array([5, 7, 9])>
In [17]:
tf.reduce_sum(t23, axis=0, keepdims=True)
Out[17]:
<tf.Tensor: shape=(1, 3), dtype=int64, numpy=array([[5, 7, 9]])>
In [18]:
tf.reduce_max(t23)
Out[18]:
<tf.Tensor: shape=(), dtype=int64, numpy=6>
In [19]:
tf.reduce_mean(t23)
Out[19]:
<tf.Tensor: shape=(), dtype=int64, numpy=3>
In [20]:
tf.argmax(t23, axis=1)
Out[20]:
<tf.Tensor: shape=(2,), dtype=int64, numpy=array([2, 2])>
In [21]:
tf.cumsum(v3)
Out[21]:
<tf.Tensor: shape=(3,), dtype=int64, numpy=array([1, 3, 6])>
In [22]:
t32 = tf.transpose(t23)
t32
Out[22]:
<tf.Tensor: shape=(3, 2), dtype=int64, numpy=
array([[1, 4],
       [2, 5],
       [3, 6]])>
In [23]:
tf.matmul(t23, t32)
Out[23]:
<tf.Tensor: shape=(2, 2), dtype=int64, numpy=
array([[14, 32],
       [32, 77]])>

Reshaping¶

In [24]:
tf.squeeze(tf.constant(0, 'int32', [3,2,1]))
Out[24]:
<tf.Tensor: shape=(3, 2), dtype=int32, numpy=
array([[0, 0],
       [0, 0],
       [0, 0]], dtype=int32)>
In [25]:
tf.expand_dims(t23, -1).shape
Out[25]:
TensorShape([2, 3, 1])
In [26]:
tf.expand_dims(t23, 0).shape
Out[26]:
TensorShape([1, 2, 3])
In [ ]: