Note: 以下绝大部分内容来自 Kyubyong 的GitHub
**-
Part 1：Constants, Sequences, and Random Values
import numpy as np
from scipy import misc
import tensorflow as tf
import matplotlib.pyplot as plt
sess = tf.InteractiveSession()
"""
Ex1. Create a tensor of the shape [2, 3] with all elements set to zero.
"""
X = tf.zeros([2,3])
print (X.eval())
"""
Ex2. Let x be a tensor of [[1,2,3], [4,5,6]].
Create a tensor of the same shape and dtype as x with all elements set to zero.
"""
x = tf.convert_to_tensor(np.array([[1,2,3],[4,5,6]]))
# x = tf.constant([[1,2,3],[4,5,6]])
X = tf.zeros_like(x)
print (X.eval())
"""
Ex3. Create a tensor of the shape [3, 2], with all elements of 5.
"""
out1 = tf.ones([3,2])*5
out2 = tf.fill([3,2],5)
print (out1.eval())
print (out2.eval())
"""
Ex4. Create a constant tensor of [[1, 3, 5], [4, 6, 8]].
"""
out = tf.constant([[1,3,5],[4,6,8]])
print (out.eval())
"""
Ex5. Create a constant tensor of the shape [2, 3], with all elements set to 4.
"""
out = tf.constant(4, shape=[2,3])
print (out.eval())
"""
Ex6. Create a 1-D tensor of 50 evenly spaced elements between 5 and 10 inclusive.
"""
out = tf.linspace(5.0,10.0,50)
print (out.eval())
"""
Ex7. Create a tensor which looks like [10, 12, 14, 16, ..., 100].
"""
out = tf.range(10, limit=102, delta=2) # [10, 102)
print (out.eval())
"""
Ex8. Create a random tensor of the shape [3, 2], with elements from a normal distribution of mean=0, standard deviation=2.
"""
out = tf.random_normal([3,2], mean=0, stddev=2)
print (out.eval())
"""
Ex9. Create a random tensor of the shape [3, 2], with all elements from a uniform distribution that ranges from 0 to 2 (exclusive).
"""
out = tf.random_uniform([3,2], minval=0, maxval=2) # [minval, maxval)
print (out.eval())
"""
Ex10. Let x be a tensor of [[1, 2], [3, 4], [5, 6]]. Shuffle x along its first dimension.
"""
x = tf.constant([[1,2],[3,4],[5,6]])
out = tf.random_shuffle(x)
print (x.eval())
print (out.eval())
"""
Ex11. Let x be a random tensor of the shape [10, 10, 3], with elements from a unit normal distribution.
Crop x with the size of [5, 5, 3].
"""
x = tf.random_normal([10,10,3])
out = tf.random_crop(x, [5,5,3])
print (out.eval())
# An image example
img = misc.imread('cat.png')
im = tf.Variable(img)
im_ = tf.random_crop(im, [100,100,3])
init = tf.global_variables_initializer()
sess.run(init)
ims = im.eval()
ims_ = im_.eval()
plt.subplot(1,2,1)
plt.imshow(ims)
plt.subplot(1,2,2)
plt.imshow(ims_)
plt.show()
Randomly crop
Part 2：Control Flow
import numpy as np
import tensorflow as tf
"""
Ex1. Let x and y be random 0-D tensors. Return x + y if x < y and x - y otherwise.
"""
x = tf.random_uniform([])
y = tf.random_uniform([])
out = tf.cond(tf.less(x,y), lambda: tf.add(x,y), lambda: tf.sub(x,y))
print (out.eval())
"""
Ex2. Let x and y be 0-D int tensors randomly selected from 0 to 5. Return x + y if x < y, x - y elif x > y, 0 otherwise.
"""
x = tf.random_uniform([], minval=0, maxval=5, dtype=tf.int32)
y = tf.random_uniform([], minval=0, maxval=5, dtype=tf.int32)
out = tf.case({tf.less(x,y): lambda: tf.add(x,y), tf.greater(x,y): lambda: tf.sub(x,y)},
default=lambda: tf.constant(0), exclusive=True)
print (out.eval())
# Alternative
x = np.random.randint(0, 5)
y = np.random.randint(0, 5)
out = np.select([x > y, x < y, x == y], [x + y, x - y, 0])
print (out)
"""
Ex3. Let x be a tensor [[-1, -2, -3], [0, 1, 2]] and y be a tensor of zeros with the same shape as x.
Return a boolean tensor that yields Trues if x equals y element-wise.
"""
x = tf.constant([[-1,-2,-3], [0,1,2]])
y = tf.zeros_like(x)
out = tf.equal(x, y)
print (out.eval())
"""
Ex4. Let x be a tensor [[-1, -2, -3], [0, 1, 2]] and y be a tensor of zeros with the same shape as x.
Return a boolean tensor that yields Trues if x does not equal y element-wise.
"""
x = tf.constant([[-1,-2,-3], [0,1,2]])
y = tf.zeros_like(x)
out = tf.not_equal(x, y)
print (out.eval())
"""
Ex5. Let x be a tensor [[-1, -2, -3], [0, 1, 2]] and y be a tensor of zeros with the same shape as x.
Return a boolean tensor that yields Trues if x is greater than or equal to y element-wise.
"""
x = tf.constant([[-1,-2,-3], [0,1,2]])
y = tf.zeros_like(x)
out = tf.greater_equal(x, y)
print (out.eval())
"""
Ex6. Let x be a tensor [[-1, -2, -3], [0, 1, 2]] and y be a tensor of zeros with the same shape as x.
Return a boolean tensor that yields Trues if x is less than or equal to y elementwise.
"""
x = tf.constant([[-1,-2,-3], [0,1,2]])
y = tf.zeros_like(x)
out = tf.less_equal(x, y)
print (out.eval())
"""
Ex7. Let x be a 0-D tensor randomly selected from -5 to 5. Return a boolean tensor that yields Trues if x
is less than 3 and x is greater than 0.
"""
x = tf.random_uniform([], minval=-5, maxval=5)
out = tf.logical_and(x<3, x>0)
print (out.eval())
"""
Ex8. Let x be a tensor [[1, 2], [3, 4]], y be a tensor [[5, 6], [7, 8]], and z be a boolean tensor
[[True, False], [False, True]]. Create a 2*2 tensor such that each element corresponds to x if True, otherise y.
"""
x = tf.constant([[1,2],[3,4]])
y = tf.constant([[5,6],[7,8]])
z = tf.constant([[True,False], [False,True]])
out = tf.select(z, x, y)
print (out.eval())
"""
Ex9. Let x be a tensor [1, 2, 3, ..., 100]. Extract elements of x greater than 30.
"""
x = tf.range(1, 101)
out = tf.gather_nd(x, tf.where(x>30))
print (out.eval())
Part 3：Math
import numpy as np
import tensorflow as tf
print (tf.__version__) # 1.0.0
# ------ Arithmetic Operators ------ #
"""
Ex1. Add x and y element-wise.
"""
x = tf.constant([1,2,3])
y = tf.constant([4,5,6])
out = tf.add(x,y)
# out = x + y
print (out.eval())
"""
Ex2. Subtract y from x element-wise.
"""
x = tf.constant([1,2,3])
y = tf.constant([4,5,6])
out = tf.subtract(x,y)
# out = x - y
print (out.eval())
"""
Ex3. Multiply x by y element-wise.
"""
x = tf.constant([1,2,3])
y = tf.constant([4,5,6])
out = tf.multiply(x,y)
# out = x * y
print (out.eval())
"""
Ex4. Multiply x by 5 element-wise.
"""
x = tf.constant([1,2,3])
out = tf.scalar_mul(5, x)
# out = 5 * x
print (out.eval())
"""
Ex5. Predict the result of this.
"""
x = tf.constant([9,11,13])
y = tf.constant([4,5,6])
out1 = tf.div(x,y) # out1 = x//y
out2 = tf.truediv(x,y) # out2 = x/y
print (out1.eval())
print (out2.eval())
"""
Ex6. Get the remainder of x / y element-wise.
"""
x = tf.constant([9,11,13])
y = tf.constant([4,5,6])
out = tf.mod(x,y)
# out = x % y
print (out.eval())
"""
Ex7. Compute the pairwise cross product of x and y.
"""
x = tf.constant([1,2,3])
y = tf.constant([4,5,6])
out = tf.cross(x, y) # tf.cross == np.cross
print (out.eval())
# ------ Basic Math Functions ------ #
"""
Ex8. Add x, y, and z element-wise.
"""
x = tf.constant([1,2,3])
y = tf.constant([4,5,6])
z = tf.constant([7,8,9])
out = tf.add_n([x,y,z])
# out = x + y + z
print (out.eval())
"""
Ex9. Compute the absolute value of x.
"""
x = tf.constant([[-1,-2,-3],[4,-5,6]])
out = tf.abs(x)
print (out.eval())
"""
Ex10. Compute numerical negative value of x, elemet-wise.
"""
x = tf.constant([[-1,-2,-3],[4,-5,6]])
out = tf.negative(x)
# out = -x
print (out.eval())
"""
Ex11. Compute an element-wise indication of the sign of x, element-wise.
"""
x = tf.constant([-5,-3,0,3,5])
out = tf.sign(x)
print (out.eval())
"""
Ex12. Compute the reciprocal of x, element-wise.
"""
x = tf.constant([1,2,3], dtype=tf.float32)
out = tf.reciprocal(x)
# out = 1/x
print (out.eval())
"""
Ex13. Compute the square of x, element-wise.
"""
x = tf.constant([1,2,-3])
out = tf.square(x)
#out = x**2
print (out.eval())
"""
Ex14. Predict the results of this, paying attention to the difference among the family functions.
"""
x = tf.constant([2.1, 1.5, -2.5, -2.9])
out1 = tf.round(x) # Rounds the values of a tensor to the nearest integer, element-wise.
out2 = tf.floor(x) # Returns element-wise largest integer not greater than x.
out3 = tf.ceil(x) # Returns element-wise smallest integer in not less than x.
print (out1.eval())
print (out2.eval())
print (out3.eval())
"""
Ex15. Compute square root of x element-wise.
"""
x = tf.constant([1., 2., 3., 4.], dtype=tf.float64)
# x = tf.constant([1., 2., 3., 4.]) # default: float32
out = tf.sqrt(x) # Note: in TensorFlow, the input tensor must be a float, whereas in Numpy, an integer is okay.
print (out.eval())
"""
Ex16. Compute the reciprocal of square root of x element-wise.
"""
x = tf.constant([1.0,4,9])
out = tf.rsqrt(x)
print (out.eval())
"""
Ex17. Compute x^y, element-wise.
"""
x = tf.constant([[1,2,3],[4,5,6]])
y = tf.constant([[1,2,1],[2,1,2]])
out = tf.pow(x,y) # tf.pow == np.power
print (out.eval())
"""
Ex18. Compute e^x, element-wise.
"""
x = tf.constant([[1,2,3],[4,5,6]], dtype=tf.float32)
out = tf.exp(x) # tf.exp == np.exp
# out = tf.pow(np.e, x)
print (out.eval())
"""
Ex19. Compute natural logarithm of x element-wise.
"""
x = tf.constant([1, np.e, np.e**2])
out = tf.log(x) # tf.log == np.log
print (out.eval())
"""
Ex20. Compute the max of x and y element-wise.
"""
x = tf.constant([1,3,5])
y = tf.constant([3,1,7])
out = tf.maximum(x, y)
# out = tf.where(x>y, x, y)
print (out.eval())
"""
Ex21. Compute the min of x and y element-wise.
"""
x = tf.constant([1,3,5])
y = tf.constant([3,1,7])
out = tf.minimum(x, y)
# out = tf.where(x", out.eval())
"""
Ex14. Predict the results of these.
"""
x = tf.constant([[True,True],[False,False]])
outs = [tf.reduce_all(x), # logical and
tf.reduce_all(x, axis=0),
tf.reduce_all(x, axis=1, keep_dims=True),
"",
tf.reduce_any(x), # logical or
tf.reduce_any(x, axis=0),
tf.reduce_any(x, axis=1, keep_dims=True),
]
for out in outs: # If you remove the common suffix "reduce_", you will get the same result in numpy.
if out == "": print ()
else: print ("->", out.eval())
"""
Ex15. Predict the results of these.
"""
x = tf.constant([[0,1,0],[1,1,0]])
outs = [tf.count_nonzero(x),
tf.count_nonzero(x, axis=0),
tf.count_nonzero(x, axis=1, keep_dims=True)]
for out in outs:
print ("->", out.eval())
"""
Ex16. Complete the einsum function that would yield the same result as the given function.
"""
_x = np.arange(1, 7).reshape((2, 3))
_y = np.arange(1, 7).reshape((3, 2))
x = tf.convert_to_tensor(_x, dtype=tf.float32)
y = tf.convert_to_tensor(_y, dtype=tf.float32)
# Matrix multiplication
out1 = tf.einsum('ij,jk->ik', x, y)
out1_ = tf.matmul(x, y)
assert np.allclose(out1.eval(), out1_.eval())
# Dot product
flattened = tf.reshape(x, [-1]) # shape: (6,)
out2 = tf.einsum('i,i->', flattened, flattened)
out2_ = tf.reduce_sum(tf.multiply(flattened, flattened))
assert np.allclose(out2.eval(), out2_.eval())
# Outer product
"""
The operation tf.expand_dims() is useful if you want to add a batch dimension to
a single element. For example, if you have a single image of shape
[height, width, channels], you can make it a batch of 1 image with
expand_dims(image, 0), which will make the shape [1, height, width, channels].
"""
expanded_a = tf.expand_dims(flattened, 1) # shape: (6, 1)
expanded_b = tf.expand_dims(flattened, 0) # shape: (1, 6)
out3 = tf.einsum('i,j->ij', flattened, flattened)
out3_ = tf.matmul(expanded_a, expanded_b)
assert np.allclose(out3.eval(), out3_.eval())
# Transpose
out4 = tf.einsum('ij->ji', x) # shape: (3, 2)
out4_ = tf.transpose(x)
assert np.allclose(out4.eval(), out4_.eval())
# ------ Scan ------ #
"""
Ex1. Compute the cumulative sum of x along axis 1.
"""
x = tf.constant([[1,2,3], [4,5,6]])
out = tf.cumsum(x, axis=1) # tf.cumsum == np.cumsum
print (out.eval())
"""
Ex2. Compute the cumulative product of x along axis 1.
"""
x = tf.constant([[1,2,3], [4,5,6]])
out = tf.cumprod(x, axis=1) # tf.cumprod == np.cumprod
print (out.eval())
# ------ Segmentation ------ #
"""
Ex3. Compute the sum along the first two elements and the last two elements of x separately.
"""
x = tf.constant([[1,2,3,4],
[-1,-2,-3,-4],
[-10,-20,-30,-40],
[10,20,30,40]])
out = tf.segment_sum(x, [0, 0, 1, 1])
print (out.eval())
"""
Ex4. Compute the product along the first two elements and the last two elements of x separately.
"""
x = tf.constant([[1,2,3,4],
[1,1/2,1/3,1/4],
[1,2,3,4],
[-1,-1,-1,-1]])
out = tf.segment_prod(x, [0, 0, 1, 1])
print (out.eval())
"""
Ex5. Compute the minimum along the first two elements and the last two elements of x separately.
"""
x = tf.constant([[1,4,5,7],
[2,3,6,8],
[1,2,3,4],
[-1,-2,-3,-4]])
out = tf.segment_min(x, [0, 0, 1, 1])
print (out.eval())
"""
Ex6. Compute the mean along the first two elements and the last two elements of x separately.
"""
x = tf.constant([[1,4,5,7],
[2,3,6,8],
[1,2,3,4],
[-1,-2,-3,-4]], dtype=tf.float32)
out = tf.segment_mean(x, [0, 0, 1, 1])
print (out.eval())
"""
Ex7. Compute the sum along the second and fourth and the first and third elements of x separately in the order.
"""
x = tf.constant([[1,4,5,7],
[2,3,6,8],
[1,2,3,4],
[-1,-2,-3,-4]])
out = tf.unsorted_segment_sum(x, [1, 0, 1, 0], num_segments=2)
print (out.eval())
# ------ Sequence Comparison and Indexing ------ #
"""
Ex8. Get the indices of maximum and minimum values of x along the second(row) axis.
"""
_x = np.random.permutation(10).reshape(2, 5)
print (_x)
x = tf.convert_to_tensor(_x)
out1 = tf.argmax(x, axis=1) # Returns the index with the largest value across axes of a tensor.
out2 = tf.argmin(x, axis=1)
print (out1.eval())
print (out2.eval())
"""
Ex9. Find the unique elements of x that are not present in y.
"""
x = tf.constant([1,2,3,4,5])
y = tf.constant([0,1,2,3])
out, idx = tf.setdiff1d(x, y)
print (out.eval())
print (idx.eval())
"""
Ex10. Return the elements of x, if x < 4, otherwise x*10.
"""
_x = np.arange(1, 10).reshape(3, 3)
x = tf.convert_to_tensor(_x)
out = tf.where(x < 4, x, x*10) # # tf.where == np.where
print (out.eval())
"""
Ex11. Get unique elements and their indices from x.
"""
_x = np.array([1, 2, 6, 4, 2, 3, 2])
x = tf.convert_to_tensor(_x)
out, indices = tf.unique(x)
print (out.eval())
print (indices.eval())
# Note that tf.unique keeps the original order, whereas np.unique sorts the unique members.
_out, _indices = np.unique(_x, return_inverse=True)
print ("sorted unique elements =", _out)
print ("indices =", _indices)
"""
Ex12. Compute the edit distance between hypothesis and truth.
"""
hypothesis = tf.SparseTensor([[0, 0],[0, 1],[0, 2],[0, 4]],
["a", "b", "c", "a"], (1, 5))
# Note that this is equivalent to the dense tensor: [["a", "b", "c", 0, "a"]]
truth = tf.SparseTensor([[0, 0],[0, 2],[0, 4]],
["a", "c", "b"], (1, 6))
# This is equivalent to the dense tensor: [["a", 0, "c", 0, "b", 0]]
out1 = tf.edit_distance(hypothesis, truth, normalize=False)
out2 = tf.edit_distance(hypothesis, truth, normalize=True)
print (out1.eval()) # 2 <- one deletion ("b") and one substitution ("a" to "b")
print (out2.eval()) # 0.6666 <- 2 / 6
Part 4：Tensor Transformations
import numpy as np
import tensorflow as tf
print (tf.__version__) # 1.0.0
sess = tf.InteractiveSession()
"""
Ex1. Let x be a tensor of [["1.1", "2.2"], ["3.3", "4.4"]]. Convert the datatype of x to float32.
"""
x = tf.constant([["1.1", "2.2"], ["3.3", "4.4"]]) # Defaults to tf.float32
out = tf.string_to_number(x)
print out.eval()
"""
Ex2. Let x be a tensor [[1, 2], [3, 4]] of int32. Convert the data type of x to float64.
"""
x = tf.constant([[1, 2], [3, 4]], dtype=tf.int32)
out1 = tf.to_double(x) # Casts a tensor to type float64.
out2 = tf.cast(x, tf.float64)
assert np.allclose(out1.eval(), out2.eval())
print out1.eval()
"""
Ex3. Let x be a tensor [[1, 2], [3, 4]] of int32. Convert the data type of x to float32.
"""
x = tf.constant([[1, 2], [3, 4]], dtype=tf.int32)
out1 = tf.to_float(x) # Casts a tensor to type float32.
out2 = tf.cast(x, tf.float32)
assert np.allclose(out1.eval(), out2.eval())
print out1.eval()
"""
Ex4. Let x be a tensor [[1, 2], [3, 4]] of float32. Convert the data type of x to int32.
"""
x = tf.constant([[1, 2], [3, 4]], dtype=tf.float32)
ou
t1 = tf.to_int32(x) # Casts a tensor to type int32.
out2 = tf.cast(x, tf.int32)
assert np.allclose(out1.eval(), out2.eval())
print out1.eval()
"""
Ex5. Let x be a tensor [[1, 2], [3, 4]] of float32. Convert the data type of x to int64.
"""
x = tf.constant([[1, 2], [3, 4]], dtype=tf.float32)
out1 = tf.to_int64(x) # Casts a tensor to type int64.
out2 = tf.cast(x, tf.int64)
assert np.allclose(out1.eval(), out2.eval())
print out1.eval()
assert np.allclose(out1.eval(), arr.astype(np.int64))
"""
Ex6. Let x be a tensor of [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]].
Create the tensor representing the shape of x.
"""
x = tf.constant([[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]])
out = tf.shape(x)
print out.eval()
"""
Ex7. Let x be a tensor of [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]]) and y be a tensor [10, 20].
Create a list of tensors representing the shape of X and Y.
"""
x = tf.constant([[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]])
y = tf.constant([10, 20])
out_x, out_y = tf.shape_n([x, y])
print out_x.eval(), out_y.eval()
"""
Ex8. Let x be a tensor of [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]].
Create a tensor representing the size (total number of elements) of x.
"""
arr = np.array([[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]])
x = tf.constant(arr)
out = tf.size(x)
print out.eval()
assert out.eval() == arr.size
"""
Ex9. Let x be a tensor of [[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]].
Create a tensor representing the rank (number of dimensions) of x.
"""
arr = np.array([[[1, 2], [3, 4]], [[5, 6], [7, 8]], [[9, 10], [11, 12]]])
x = tf.constant(arr)
out = tf.rank(x)
print out.eval()
assert out.eval() == arr.ndim
"""
Ex10. Let x be tf.ones([2, 2, 3]). Reshape x so that the size of the second dimension equals 6 (means the size is [2,6]).
"""
x = tf.ones([2, 2, 3])
out = tf.reshape(x, [-1, 6])
print out.eval()
"""
Ex11. Let x be tf.ones([10, 10, 1, 1]). Remove all the dimensions of size 1 in x.
"""
x = tf.ones([10, 10, 1, 1])
out = tf.squeeze(x) # Removes dimensions of size 1 from the shape of a tensor.
print out.eval().shape
assert np.allclose(out.eval(), np.squeeze(np.ones([10, 10, 1, 1])))
"""
Ex12. Let X be tf.ones([10, 10, 1, 1]). Remove the third and fourth dimensions in x.
"""
x = tf.ones([10, 10, 1, 1])
out = tf.squeeze(x, [2,3])
print out.eval().shape
"""
Ex13. Let x be a tensor:
[[[1, 1, 1], [2, 2, 2]],
[[3, 3, 3], [4, 4, 4]],
[[5, 5, 5], [6, 6, 6]]].
Extract the [[3, 3, 3], [4, 4, 4]] from x.
"""
arr = np.array([[[1, 1, 1],
[2, 2, 2]],
[[3, 3, 3],
[4, 4, 4]],
[[5, 5, 5],
[6, 6, 6]]])
x = tf.constant(arr)
out = tf.slice(x, [1, 0, 0], [1, 2, 3]) # tf.slice(input_, begin, size, name=None)
print out.eval()
assert np.allclose(out.eval(), arr[1, :, :])
"""
Ex14. Let x be a tensor of
[[1 2]
[3 4]
[5 6]
[7 8]
[9 10]].
Extract the [[1, 2], [5, 6], [9, 10]]] from x.
"""
x = tf.reshape(tf.range(1, 11), [5, 2])
out = tf.strided_slice(x, [0], [5], [2]) # tf.strided_slice(input_, begin, end, strides=None, ...)
print out.eval()
arr = np.reshape(np.arange(1, 11), [5, 2])
assert np.allclose(out.eval(), arr[[0, 2, 4]])
"""
Ex15. Let x be a tensor of
[[1 2 3 4 5]
[6 7 8 9 10]].
Split x into 5 tensors along the second dimension.
"""
x = tf.reshape(tf.range(1, 11), [2, 5])
out = tf.split(x, num_or_size_splits=5, axis=1)
print [each.eval() for each in out]
"""
Ex16. Lex X be a tensor
[[1 2 3]
[4 5 6].
Create a tensor looking like
[[1 2 3 1 2 3 1 2 3 ]
[4 5 6 4 5 6 4 5 6 ]].
"""
x = tf.reshape(tf.range(1, 7), [2, 3])
out = tf.tile(x, [1, 3])
print out.eval()
"""
Ex17. Lex x be a tensor
[[1 2 3]
[4 5 6].
Pad 2 0's before the first dimension, 3 0's after the second dimension.
"""
x = tf.reshape(tf.range(1, 7), [2, 3])
out = tf.pad(x, [[2, 0], [0, 3]]) # paddings=[[2, 0], [0, 3]] indicates [[上,下],[左,右]]
print out.eval()
"""
Ex18. Lex x be a tensor
[[1 2 3]
[4 5 6]]
and y be a tensor
[[7 8 9]
[10 11 12]]
Concatenate x and y so that a new tensor looks like
[[1, 2, 3, 7, 8, 9], [4, 5, 6, 10, 11, 12]].
"""
x = tf.constant([[1, 2, 3], [4, 5, 6]])
y = tf.constant([[7, 8, 9], [10, 11, 12]])
out = tf.concat([x, y], axis=1) # equal np.concatenate((x, y), axis=1))
print out.eval()
"""
Ex19. Let x be a tensor [1, 4], y be a tensor [2, 5], and z be a tensor [3, 6].
Create a single tensor from these such that it looks [[1, 2, 3], [4, 5, 6]].
"""
x = tf.constant([1, 4])
y = tf.constant([2, 5])
z = tf.constant([3, 6])
out = tf.stack([x, y, z], axis=1) # tf.stack([x, y, z], axis=0) ==> [[1, 4], [2, 5], [3, 6]]
print out.eval()
"""
Ex20. Let x be a tensor [[1, 2, 3], [4, 5, 6]]. Convert X into 3 tensors:
[1, 4], [2, 5], [3, 6].
"""
x = tf.constant([[1, 2, 3], [4, 5, 6]])
out = tf.unstack(x, axis=1)
print [each.eval() for each in out]
"""
Ex21. Let X = tf.resahpe(tf.range(1, 1*2*3*4+1), [1, 2, 3, 4]). Reverse the last dimension.
[[[[ 1 2 3 4] [[[[ 4 3 2 1]
[ 5 6 7 8] [ 8 7 6 5]
[ 9 10 11 12]] [12 11 10 9]]
===>
[[13 14 15 16] [[16 15 14 13]
[17 18 19 20] [20 19 18 17]
[21 22 23 24]]]] [24 23 22 21]]]]
"""
X = tf.reshape(tf.range(1, 1*2*3*4+1), [1, 2, 3, 4])
out = tf.reverse(X, axis=[3])
print out.eval()
"""
Ex22. Let x = tf.constant([[1, 2, 3], [4, 5, 6]]). Transpose x.
"""
x = tf.constant([[1, 2, 3], [4, 5, 6]])
out = tf.transpose(x)
print out.eval()
"""
Ex23. Let x be a tensor [[1, 2, 3], [4, 5, 6], [7, 8, 9]]. Get the first, and third rows.
"""
x = tf.constant([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
out1 = tf.gather(x, [0, 2])
out2 = tf.gather_nd(x, [[0], [2]])
assert np.allclose(out1.eval(), out2.eval())
print out1.eval()
"""
Ex24. Let x be a tensor [[1, 2, 3], [4, 5, 6], [7, 8, 9]]. Get the 5 and 7.
"""
x = tf.constant([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
out = tf.gather_nd(x, [[1, 1], [2, 0]])
print out.eval()
"""
Ex25. Let x be a tensor [2, 2, 1, 5, 4, 5, 1, 2, 3]. Get the unique elements and their counts.
e.g. # tensor 'x' is [1, 1, 2, 4, 4, 4, 7, 8, 8]
y, idx, count = unique_with_counts(x)
y ==> [1, 2, 4, 7, 8]
idx ==> [0, 0, 1, 2, 2, 2, 3, 4, 4]
count ==> [2, 1, 3, 1, 2]
"""
x = tf.constant([2, 2, 1, 5, 4, 5, 1, 2, 3])
out1, _, out2 = tf.unique_with_counts(x)
print out1.eval(), out2.eval()