Community mailing list archives
Re: Forecasting using Machine Learning methodsby
Demand forecasting is just one possible application. Over time companies tend to put more and more data into Odoo. They start out with a few modules and eventually can put large chunks of their data in. Over time this can grow into a lot of records. There are a lot of applications where far less than 100k records are needed to get benefit from ML. But let the user select which algorithm works best after testing.
Perhaps the module needs to have a base framework (like odoo-connector.com) where different statistically methods could be incorporated into it. The main goal being that any object in Odoo can be referenced. Select 2 or more data (in) objects, specify size of a test data set (if using ML) that could be randomly selected, select an algorithm and select an output (like total sales for the month, demand for Product A, number of employees to hire, etc.).
On Mon, Jan 4, 2016 at 10:08 AM, Roman Gurinovich <firstname.lastname@example.org> wrote:
Hi Chris,there is no single and the best tool to use for data analysis at this stage of software development. Package / library should be selected for the business need specifically.You've asked about forecasting for supply chain. As long as most Odoo customers are SME there is no benefit of Machine Learning approaches over more simple statistical analysis. Consider 100 000 records at least for effective learning. I would advise one of the next algorithms for your task:- moving average / auto regression- extrapolation- LPCTry your approach in R standalone then move to Python (as long as Odoo is Python based too) on production. Library to use:On Sun, Jan 3, 2016 at 10:35 AM, Gunnar Wagner <email@example.com> wrote:On 1/3/2016 8:54 AM, Chris Jones wrote:<blockquote cite="mid:CAK1zTiahVwsSOkF7-AP39WXKX0f8OrOK5Q847TX2XEOh+oMLxA@mail.gmail.com" type="cite">
All the modules we've built are for v7. I would love to upgrade to v8/9 but really need a paying customer to cover the cost.integrating your modules into OCA projects would probably draw some additional developer attention/power to them (also in regards to porting)
<blockquote cite="mid:CAK1zTiahVwsSOkF7-AP39WXKX0f8OrOK5Q847TX2XEOh+oMLxA@mail.gmail.com" type="cite">
On Thu, Dec 31, 2015 at 10:57 AM, Kristian Koci <firstname.lastname@example.org> wrote:
Are these modules all for OpenERP v7?
Also, this possibly Sickit based new one would be for v7?
Would be nice to collaborate on it.
On Thu, Dec 31, 2015 at 2:13 PM, Chris Jones <email@example.com> wrote:
Lokad looks interesting but is a bit expensive and wouldn't be integrated directly into Odoo. I'd really prefer to use open source Python libraries. Scikit Learn looks like a really good place to start.
I created a new folder in my GIT repository for it called forecast_ml. https://github.com/saasier1/saasier_public_addons
I have a friend who works in the Watson group and I was at the RoboUniverse Show (showing off a boxing robot I built, overthrowrobotics.com for those interested) in San Diego a couple of weeks ago and saw a presentation on Watson. Super cool, seriously almost gave me the chills what it can do. But I didn't see anything on regression or forecasting that would help for this project. They have great natural language and classifications tools and a solid API.
On Wed, Dec 30, 2015 at 7:47 PM, Marcelo Bello <firstname.lastname@example.org> wrote:
Chris, I myself look forward to the day I will have a good integration of Odoo with Lokad, the best demand planning solution that I know of (their quantile algorithm worked best for my business). The beauty of Lokad in my opinion is the output which works for the typical demand planning case but also lets one easily answer a question like "I can buy no more than X dollars worth of goods, what should I buy now?".
My point is, if you just want to play with machine learning/AI/neural network stuff then sure go ahead. Now if you instead have real demand from your customers to come up with a solid demand planning solution than I would say go with what is already out there, demand planning is serious stuff, can have a huge impact on the bottom line of your clients and usually people/businesses care about the credentials of who built/maintain their demand planning solution.
If you happen to check and like Lokad and decide to give it a try, let me know (I am a paying customer of Lokad, using it for more than 2 years now).
On Wed, Dec 30, 2015 at 10:27 PM, Chris Jones <email@example.com> wrote:
I'd like to build a module for demand forecasting that uses neural networks, SVM (Support Vector Machines) or some similar machine learning algorithm.
The idea would be that you could select in the web interface which objects you wanted to be data points (input) and what you want the output to be (Sales for January 2016, Demand for Product A, etc).
The thing that makes it powerful is that any data set in ERP could be used, it could be 15 different objects and the algorithm would find associations and correlations that we couldn't intuitively on our own. In my head I can handle 3 dimensions of data that might impact Sales, but my brain will start to turn into mush by 5, let alone 15.
The module could also let the user specify which module they want to try and possibly upload or specify some data to test against to see which specific algorithm works best.
I've been doing some work with other machine learning algorithms in C# for a robotics project but I'm still a bit new to it. I know there are some great Python machine learning libraries out there.
Is anyone interested in this? Thoughts?
This email has been sent from a virus-free computer protected by Avast.
Gunnar Wagner | Iris Germanica Co., Ltd. | Jin Qian Gong Lu 385, 8-201, Feng Xian District, 201404 Shanghai, P.R. CHINA
+86.159.0094.1702 | skype: professorgunrad | wechat: 15900941702
Director, XPANSA Group | ERP, BI, E-commerce, Data Mining and DMS consultingXpansa CLOUD | Full Business IT Infrastructure | https://xpansa.com/products/xpansa-odoo-alfresco-bi-saas/mERP | Odoo Android App | merpapp.com/// site : xpansa.com
/// mail : firstname.lastname@example.org