We are a garment sizing recommendation software company based in Estonia and London. Our software is used by fashion retailers to give their customers a better decision on what size to purchase. Right now the app works purely using past shopper data and sizing feedback, no size charts required.
We do not use any machine learning or advanced algorithms to make the sizing predictions, and therefore the current build has various weaknesses and can cause inaccuracies. We also do not utilise any external datasets.
To improve the software and utilize machine learning to make more precise sizing recommendations, we are seeking a data scientist with python experience and general experience in solving similar situations to our use-case.
To summarise what we do have and what we want:
- Gender, height, weight, body shape, fit preference, age - from previous customers
- Fit feedback on if the product they purchased fitted them or not.
We DON'T have:
- Use of public datasets, such as:
[login to view URL]
[login to view URL]
At the moment we are relying only on our own data from the specific retailers existing customer base. This means that we cannot give predictions on size until the product has a specific number of purchases from customers who provided their body data. We would like to combine datasets such as this one above in order to provide recommendations when we don't yet have purchase data. The question is, how can we utilise this public dataset in our use-case and combine it with our brand specific purchase data?
- Currently fit feedback has no impact on recommendations. The way we are generating recommendations is based solely on PHP now (finding close data within a specific height/weight range), not an ML model.
- We do not have a model that can take garment measurements and predict sizing based solely on the garment measurements. See this article
[login to view URL]
I would like to know if it is a viable option to create recommendations based on garment measurements provided by a specific retailer. OR should we stick to utilising clothing fit datasets like mentioned further up? What is the best way of cold starting a product size predictions? If a product has no purchase data, how should we generate recommendations - using the public purchase dataset, or by requesting garment measurements from the brand and utilising them in a model?
We already had an ML model made by a data scientist using random forests and the problem we encounted is that if there was one row in the database where a customer provided fit feedback stating a product fitted them, it would generate a recommendation of 100% certainty that the new visitor should choose this size. It did not consider other close values. If there was only one entry for someone of the same height, weight with fit feedback of 'fitted' then ignoring other data. I understand this is how most algorithms work, but how to we solve this?
You will be responsible for reviewing our current process, reviewing and planning implementation of improved algorithm, utilising public datasets combined with our own data and building a new solution. Your knowledge and work will be fundamental to the improvement of our software so it is vital that we hire something who is highly talented with great mathematical skill.
Your application should be integrated with our PHP application, most likely through an API. You should be able to support our developer on integrating your system.
This is a part time role that will typically be paid on a fixed fee or hourly fee for completing certain tasks.
If you think this is the perfect job for you, please get in touch with us!
19 freelanceria on tarjonnut keskimäärin 543$ tähän työhön
I have been working in python for a year and being from engineering background I'm familiar to ML algo. Can't say exactly how many days however 20 - 30 days will be sufficient for good implementation.