Hello, if you have any need, please feel free to consult us, this is my wechat: wx91due
CBU5201 mini-project submission
What is the problem?
This year's mini-project considers the problem of predicting whether a narrated story is true or not. Specifically, you will build a machine learning model that takes as an input an audio recording of 3-5 minutes of duration and predicts whether the story being narrated is true or not.
Which dataset will I use?
A total of 100 samples consisting of a complete audio recording, a Language attribute and a Story Type attribute have been made available for you to build your machine learning model. The audio recordings can be downloaded from:
https://github.com/CBU5201Datasets/Deception
A CSV file recording the Language attribute and Story Type of each audio file can be downloaded from:
https://github.com/CBU5201Datasets/Deception/blob/main/CBU0521DD_stories_attributes.csv
What will I submit?
Your submission will consist of one single Jupyter notebook that should include:
- Text cells, describing in your own words, rigorously and concisely your approach, each implemented step and the results that you obtain,
- Code cells, implementing each step,
- Output cells, i.e. the output from each code cell,
Your notebook should have the structure outlined below. Please make sure that you run all the cells and that the output cells are saved before submission.
Please save your notebook as:
- CBU5201_miniproject.ipynb
How will my submission be evaluated?
This submission is worth 16 marks. We will value:
- Conciseness in your writing.
- Correctness in your methodology.
- Correctness in your analysis and conclusions.
- Completeness.
- Originality and efforts to try something new.
(4 marks are given based on your audio submission from stage 1.)
The final performance of your solutions will not influence your grade. We will grade your understanding. If you have an good understanding, you will be using the right methodology, selecting the right approaches, assessing correctly the quality of your solutions, sometimes acknowledging that despite your attempts your solutions are not good enough, and critically reflecting on your work to suggest what you could have done differently.
Note that the problem that we are intending to solve is very difficult. Do not despair if you do not get good results, difficulty is precisely what makes it interesting and worth trying.
Show the world what you can do
Why don't you use GitHub to manage your project? GitHub can be used as a presentation card that showcases what you have done and gives evidence of your data science skills, knowledge and experience. Potential employers are always looking for this kind of evidence.
-------------------------------------- PLEASE USE THE STRUCTURE BELOW THIS LINE --------------------------------------------
4 Implemented ML prediction pipelines
Describe the ML prediction pipelines that you will explore. Clearly identify their input and output, stages and format of the intermediate data structures moving from one stage to the next. It's up to you to decide which stages to include in your pipeline. After providing an overview, describe in more detail each one of the stages that you have included in their corresponding subsections (i.e. 4.1 Transformation stage, 4.2 Model stage, 4.3 Ensemble stage).
5 Dataset
Describe the datasets that you will create to build and evaluate your models. Your datasets need to be based on our MLEnd Deception Dataset. After describing the datasets, build them here. You can explore and visualise the datasets here as well.
If you are building separate training and validatio datasets, do it here. Explain clearly how you are building such datasets, how you are ensuring that they serve their purpose (i.e. they are independent and consist of IID samples) and any limitations you might think of. It is always important to identify any limitations as early as possible. The scope and validity of your conclusions will depend on your ability to understand the limitations of your approach.
If you are exploring different datasets, create different subsections for each dataset and give them a name (e.g. 5.1 Dataset A, 5.2 Dataset B, 5.3 Dataset 5.3) .