I noticed that there are a few typos and a couple programming errors in the RNN jupyter notebook (pulled Aug. 6th, 2017).
Many (4) times: LTSM vs LSTM - including near the top where it’s spelled out “Long Term Short Memory”
Notice here that each input is a sequence (or vector) of length 4
should be
… of length 5
X.append(odd_nums[5:7])
shouldn’t exist, because it should only use up to p-1 samples. In this case X is 6 sequences long, but there are only 5 values for y.
In section 1.5, the plotting is offset by 1 space. This is because the dataset is plotted without an x-axis counterpart and defaults to 0-index, whereas the other elements are plotted along with a 1-index-based x-axis value.
In section 2.5
"where each vector has 32 entries"
should be
"… 33 entries", because there are 26 letters, 6 punctuation marks and 1 space character
It’s not clear in section 2.6 why the softmax activation function is suggested as an entirely separate layer vs being an activation function for layer 2.
There’s an extra TODO that’s already been done in the provided notebook.
TODO: fit to our larger dataset
created
20h
last reply
7h
2
replies
7
views
3
users
1
like
1
link
cgearhart
commented
6 years ago
I noticed that there are a few typos and a couple programming errors in the RNN jupyter notebook (pulled Aug. 6th, 2017).