nvdli / ldl Goto Github PK

when executing the code c16e2 i get the following error

TypeError: int() argument must be a string, a bytes-like object or a number, not 'tuple'

For certain platforms, we have observed hangs in the c11e1_autocomplete and c12e1_autocomplete_embedding examples.

The first code cell has a variable pointing to a file:

INPUT_FILE_NAME = '../data/frankenstein.txt'

Then, in the second cell, this line fails

file = open(INPUT_FILE_NAME, 'r', encoding='utf-8-sig')

hello, I found this line w.shape is (784,) but w shape is (785,). I dont know why input layer concatenate 1.0, It's bias?

Thanks.

We have observed a memory leak problem in the TensorFlow versions of the NAS examples:
c17e4_nas_random_hill
c17e5_nas_evolution

In c5e2_mnist_learning_conf5 notebook for PyTorch, it's missing softmax layer. There is ReLU layer but missing softmax layer.

Thanks!

I can't find or use pickle.Gzip and don't know this, quite a beginner

However, the code snippet below contains three calls to MNIST. The reason for that is that we want to first access the training dataset to compute the mean and standard deviation.

trainset = MNIST(root='./pt_data', train=True, download=True, transform=transform)

But why the second call defining trainset? You already loaded trainset once to define mean and stdev. Why call it a second time?

Hi, the examples that use text files with Byte Order Marker (BOM), like frankenstein.txt, should use encoding='utf-8-sig' instead of utf-8 to skip the BOM, otherwise you get this:

In [4]: with open('frankenstein.txt', 'r', encoding='utf-8') as f:
   ...:     text = text_to_word_sequence(f.read())
   ...: print(text[0:5])

['\ufeffthe', 'project', 'gutenberg', 'ebook', 'of']

I realize it doesn't matter much, but it was confusing to see that word in there. One more note, np.int is deprecated.

Great book, thanks!