Spaces:
Sleeping
Sleeping
Ans
commited on
Commit
•
ef63b9b
1
Parent(s):
85ca6f3
Second prototype
Browse files- {protoypes → prototypes}/L4_Project_first.ipynb +0 -0
- prototypes/prototypes.md +34 -0
- timelog.md +12 -0
{protoypes → prototypes}/L4_Project_first.ipynb
RENAMED
File without changes
|
prototypes/prototypes.md
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Prototypes
|
2 |
+
|
3 |
+
* Text-generation with rhyme and rhythm
|
4 |
+
* Ans Farooq
|
5 |
+
* 2390370f
|
6 |
+
* Jake Lever
|
7 |
+
|
8 |
+
## Intro
|
9 |
+
|
10 |
+
* This file contains information and reflections about each prototype of the project.
|
11 |
+
|
12 |
+
## 29/10/2021 - First Prototype - _L4_Project_first.ipynb_
|
13 |
+
|
14 |
+
### Current state/Improvements made
|
15 |
+
* The first prototype combines a causal language model and a masked language model, GPT-2 and RoBERTa, to generate the starts of sentences and fill in the rest until the end rhyming word.
|
16 |
+
|
17 |
+
* For this prototype, the starting word and end rhyming word of each line was pre-determined and hard-coded. This was temporary as I was focused on getting GPT-2 and RoBERTa working and generating some coherent lines of text.
|
18 |
+
|
19 |
+
### Future improvements
|
20 |
+
|
21 |
+
* Use a Python library to generate rhyming words
|
22 |
+
* Use user input for the topic of the limerick
|
23 |
+
* Feed summary of topic from wikipedia to GPT-2 before it generates the start of each line
|
24 |
+
|
25 |
+
## 16/11/2021 - Second Prototype - _L4_Project_second.ipynb_
|
26 |
+
|
27 |
+
### Current state
|
28 |
+
* The second prototype uses GPT-2 and RoBERTA, but instead of hard-coded starting and end rhyming words, it uses the *pronouncing* Python library to find rhyming words and the *wikipedia* library to feed a summary of the topic to GPT-2 before generation.
|
29 |
+
|
30 |
+
### Future improvements
|
31 |
+
* Word frequencies/counts for filtering rhyming library words
|
32 |
+
* Generate the first line a few times until a decent end word, e.g noun, hand-built filters, or word list
|
33 |
+
* Have a look into improving the rhyme finding, better libraries? Phonetics? Could just filter out words with multiple pronunciations.
|
34 |
+
|
timelog.md
CHANGED
@@ -70,3 +70,15 @@
|
|
70 |
|
71 |
* *0.5 hours* Meeting with supervisor
|
72 |
* *1 hour* Created GitHub repository and uploaded dissertation and project template from moodle
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
70 |
|
71 |
* *0.5 hours* Meeting with supervisor
|
72 |
* *1 hour* Created GitHub repository and uploaded dissertation and project template from moodle
|
73 |
+
|
74 |
+
### 29 Oct 2021
|
75 |
+
* *0.5 hours* Created prototypes directory on GitHub and saved first prototype directly from Colab to GitHub
|
76 |
+
|
77 |
+
## Week 7
|
78 |
+
|
79 |
+
### 2 Nov 2021
|
80 |
+
* *5 hours* Used Python library to find rhyming words and wikipedia library to feed topic summary into GPT-2 before text generation
|
81 |
+
|
82 |
+
### 3 Nov 2021
|
83 |
+
|
84 |
+
* *0.5 hours* Meeting with supervisor
|