Colin-Codes / IntentClassifier-ML-Project

Pyhton, Keras, SciKit-Learn, Matplotlib: Machine learning research project around classification of intent behind tech support emails in order to enable automatic follow up.
0 stars 0 forks source link

Complete IDV and IPR #21

Closed Colin-Codes closed 4 years ago

Colin-Codes commented 5 years ago

https://herts.instructure.com/courses/44545/assignments/24540

Colin-Codes commented 5 years ago

750 - 1200 words, PDF format

The Interim Project Report must contain the following sections:
1) Completed work (up to 600 words): work completed since the submission of the Extended Project Proposal; you MUST discuss if you involve human subjects in your project. You also need to provide the protocol number of your ethics approval when applicable. You do not need to include a full literature review or things that have been already covered in your EPP. 2) Work to be done (up to 600 words): work to be completed before the submission of the Final Project Report, including what will be involved in evaluating the project and its product. This part should allow your assessors to judge if your final solution is appropriate and achievable.

Marks will be awarded for: • Quality and amount of practical work done on the project itself, as judged from the report (3 marks). • Discussion of issues relating to work done and planned to be done on the project (2 marks). • Description of problems you have faced and overcome, and other challenges that remain to be resolved (2 marks). • Presenting the issues in a logical, well ordered way (2 marks). • Readability, grammar and spelling (1 mark).

IPR is worth 5% of the overall module marks so the above 10 marks is scaled to 5 in total. It is marked and given feedback by the second marker of your project.

Colin-Codes commented 5 years ago

5% of total Record in MP4, not high quality.

An Interim Demonstration Video (IDV) is a video demonstration of your project work focusing on the practical side. You will have had about 60% of the total time on your project when you submit this IDV.

The IDV should record both your interaction with what you have produced for your project known as screen activity, and your speech of presenting your work, plus your image captured by a web camera.

The basic hardware facilities you need for VPL are a web camera and a headset. The web camera is ideally with 640 x 480-pixel resolution. The software you will use is called Open Broadcast Software (OBS) Studio. It is an open source, screen recording software, available for both Windows and Mac users. You can download the latest stable version from:

https://obsproject.com/

Use of OBS is fairly straightforward. There is an Instruction Menu to help you to create your FDV on Canvas module site. It is important that you read this menu before making the video.

The screen activities include: opening and closing an interface/document/application; operating on an application; mouse indication of what you are explaining.

The web camera should be placed on top of your desktop or laptop, with your face clearly visible. The web camera capture of your image should be placed on top right hand corner of the screen, one eighth of your screen size. Your image shouldn’t obscure your screen activities, e.g., when you present your ERD, the whole ERD should be visible. Missing of you image in IDV will be treated as incomplete submission.

While you present your work, your speech is recorded. It is recommended that you write a simple speech script to help you to organize your talk. In your speech, you only describe what each task is set to do and what your findings are: be brief and in points, to avoid reading the large part of or full of your document.

Your IDV should last no more than 10 minutes. A 50% over the limit will cap your mark to the half of the total, i.e., 2.5.

Please do not use any slides to aid your demonstration in the IDV. Please also do not repeat what have been written in your IPR as your assessors will read your report in details.

Marks will be awarded for: • The amount of practical work done on the project itself, as judged from relevant evidence provided in the video (2 marks). • The quality of the work done on the project, as judged from relevant evidence provided in the video (2 marks). • The description of problems the student has faced and overcome, and other challenges that remain to be resolved (2 marks). • Whether or not the video presents the issues in a logical, well ordered way (2 marks). • Whether or not the supporting evidences are presented visibly and clearly (2 marks). IPR is worth 5% of the overall module marks so the above 10 marks is scaled to 5 in total. It is marked and given feedback by the second marker of your project.

Colin-Codes commented 4 years ago
Colin-Codes commented 4 years ago

Mention:

Scoring for validation Architecture of Neural nets

Colin-Codes commented 4 years ago

Script:

Data processing

Demo - sample import Anonymisation / Generalisation - useful for smaller data sets Automation - easier in future Biggest tasks - Classification and Assurance

KNN

Demo - optimisation of k Demo - prediction of test set Demo - Prediction with enhanced training set Most promising - little work, best results Flawed by smaller classes when k is required to be higher Also no idea of certainty - perhaps factorial or word drop out approach

Neural networks

Tour - Support.py for consistency Demo - LSTM.py: consistently predicts Account all the time - overtraining?

High batch size reduces noise