Week_1 Feedback Typos & Change in a fucntion(assignment2_dinosaurs)

Course Q&A Deep Learning Specialization Sequence Models
1

Hello,
I already finished the specialization in March 2021. Now,I’m redoing the assignments for the new version of the course5. Although for the week 1, there is no Tensorflow in the assignments (just numpy…), but since we lost the old assignments (old version), I wanted to redo them too.

Part1

I detected a little typo ,in assignment1: “Building_a_Recurrent_Neural_Network_Step_by_Step”:

under paragraph “2 - Long Short-Term Memory (LSTM) Network”, at “Update” level gate 𝚪𝑖 ", at the 3rd sentence:

  • “When a unit in the update gate is close to 1, it allows the value of the candidate 𝐜̃ ⟨𝑡⟩ to be passed onto the hidden state 𝐜⟨𝑡⟩”, ==> should become: "When a unit in the update gate is close to 1, it allows the corresponding value in the candidate 𝐜̃ ⟨𝑡⟩ to be passed onto the cell state 𝐜⟨𝑡⟩ "

and:

  • “When a unit in the update gate is close to 0, it prevents the corresponding value in the candidate from being passed onto the hidden state.” ===> should become:
    “When a unit in the update gate is close to 0, it prevents the corresponding value in the candidate from being passed onto the cell state.”

Part2

it is just a test I did (by changing the function model (Exercice 4), of the assignment 2 (week1): "Dinosaurus_Island_Character_level_language_model "

In fact, I didn’t like the generated names of the dinosaurs (as we were asked to do, that leads to All test passesd) :grin: in particular the names : “Meusspanchodtashuarhiaspantaxia” or “Toraperlethosdarenitochusthiamamumamaon” :upside_down_face:
In fact in fact by comparing them with the names of the dataset, there are sometimes names too long and even too far from the logic of the names in the dataset (and especially not beautiful, no clear rhyme …)

I tried to make these changes and it gave me the most beautiful names :blush:

-Change 1
in the function sample (Exercice 2)
I just changed one line:
while (idx != newline_character and counter != 10) # 10 instead of 50

-Change 2
I added this function before the function model (Exercice 4)

def create_X_Y(seq):
    X_input_txt = seq[:-1] 
    Y_target_txt = seq[1:] 
    return X_input_txt, Y_target_txt

In fact, instead of adding [None] to get X
and instead of adding [ix_newline] to X (shifted by one char)…

-Change 3
I deleted X and Y (of the assignment),in the function model, and I replaced them by

X,Y = create_X_Y(single_example_ix)

And I got beautiful names (but also lower loss ) :blush:
like:
Nedalosara
Jytronlero
Infaberpee
Lytorapisa
Yusoghator
…

2

Thanks for your suggestions.

1 Like
3

If I understand your ‘‘Part 2’’ correctly: by limiting the characters to 10 instead of 50 you get better names & loss?

Because i also do not like the 3-4 letter names and neigther the ones produced like “Mystolojmiaterltasaurus”.