Errata

print ans
should be replaced with
print(ans)

As of Python 3.0, "print" was replaced by the function "print()"

The figure is supposed to represent the model described in the text.
However the graph includes a bias variable node "b" while the text doesn't mention any bias.

Note from the Author or Editor:
add to the image caption:
"Here, b is a bias term that could be added to the model"

The definition of the subtraction shortcut as
tf.subtract() a-b Subtracts a from b, element-wise.

should be

tf.subtract() a-b Subtracts b from a, element-wise.

I guessing that the LaTex formula used was this:
$$ H(p, q) = - \Sigma_x p(x) log q(x) $$

But this is correct, which puts the x UNDER the Summation sign instead of making x a SUBSCRIPT of Sigma:
$$ H(p, q) = - \sum_x p(x) log q(x) $$

Note from the Author or Editor:
Thanks!

On page 46 , Logistic regression.

The sigmoid function is written as Pr(y_i=1 | x_i) = 1/(1+exp(w*x_i+b) but it should instead be

written as

Pr(y_i=1 | x_i) = 1/(1+exp(-w*x_i-b)

Thanks,

Theophilus Siameh

In
"we will use a tf.Variable and pass one value for train (.5) and another for test (1.0)."

tf.Variable should be tf.placeholder instead.

"Next we have two consecutive layers of convolution and pooling, each with 5x5 convolutions and 64 feature maps, followed by a single fully connected layer with 1,024 units."

The first convolution layer has only 32 feature maps, not 64 as written.

cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(y_conv,y_))

should be

cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=y_conv, labels=y_))

Note from the Author or Editor:
Thanks! This will be fixed in future versions.

code reads:
conv3_drop = tf.nn.dropout(conv3_flat, keep_prob=keep_prob)

full1 = tf.nn.relu(full_layer(conv3_flat, F1))

this doesn't make sense, it should read:

conv3_drop = tf.nn.dropout(conv3_flat, keep_prob=keep_prob)

full1 = tf.nn.relu(full_layer(conv3_drop, F1))

Note from the Author or Editor:
Thank you for catching this! We have already fixed it in the git repo, and will make the change in future versions of the boo.

The code does not work: ValueError: Dimensions must be equal ...

Instead of using the lstm_cell instance multiple times:

cell = tf.contrib.rnn.MultiRNNCell(cells=[lstm_cell]*num_LSTM_layers,state_is_tuple=True)

one instance per LSTM layer should be created - for example in this way:

cell = tf.contrib.rnn.MultiRNNCell(cells=[tf.contrib.rnn.BasicLSTMCell(hidden_layer_size, forget_bias=1.0) for i in range(num_LSTM_layers)], state_is_tuple=True)

Note from the Author or Editor:
On page 93, under Stacking multiple LSTMs, replace these two lines:

lstm_cell = tf.contrib.rnn.BasicLSTMCell(hidden_layer_size, forget_bias=1.0)
cell = tf.contrib.rnn.MultiRNNCell(cells=[lstm_cell]*num_LSTM_layers, state_is_tuple=True)

With:

lstm_cell_list = [tf.contrib.rnn.BasicLSTMCell(hidden_layer_size,forget_bias=1.0) for ii in range(num_LSTM_layers)]
cell = tf.contrib.rnn.MultiRNNCell(cells=lstm_cell_list, state_is_tuple=True)