MLJAR is an automated machine learning platform. With mljar, you can train great machine learning models without coding. It works with binary classification and regression tasks. It can do hyperparameters tuning and model selection. The preprocessing to deal with missing values and categorical columns is available. All models trained in the service are by default deployed in the cloud and can be accessed be REST API or python or R APIs. User can also download a model and use it locally.
In the long term, I would like to connect machine learning with databases. User will be able to train machine learning models by writing a SQL query to database. After the best model is trained (with AutoML of course!), all new rows that will appear in the database will be used for computing predictions. But first I would like to create AutoML platform :)
I think that AutoML will be a standard in ML. You don't want to run fit on a single algorithm, just call AutoML and specify how long you would like to wait for final model (you can also add constraints on model size and model prediction speed). Waht is more, I would like to have ML model always with explainer. I don't see disadvantages of using AutoML vs single model.
Thank you! I believe AutoML has huge value and can be a new standard in ML. I would like to see the adoption of AutoML - that's why I'm open sourcing it. Otherwise, I would need a lot of money for marketing and sales to make it popular.
Before open sourcing I was looking at Metabase and Redash solutions, and I was very impressed with their business model - I would like to achieve something similar. The goal is to be ramen profitable.
I will offer hosted paid version (SaaS offer similar to redash) and I will offer paid white labeled embedded machine learning platform (similar to metabase). Plus paid support, paid feature requests, ML consulting. Does it make sense?
I wish someone built something like this for generic classification. You give the system a bunch of folders. Each folder is a label containing corresponding samples. The system then creates a replica of the folder structure with no content. Each time you give it a new piece of data it places it in one of those newly created folders.
This is an interface truly anyone could use. Just call it "intelligent folders" or something. The user doesn't even need to know which algorithm it uses. Just split sample into training and test data at random and choose the algo that give best results.
I was working on this for text data, but then switched jobs and don't have energy to make this in my spare time right now.
LingPipe http://alias-i.com/lingpipe/ had an example that did this IIRC it was their newsgroup example http://alias-i.com/lingpipe/demos/tutorial/classify/read-me...., I adapted it in to running a trial classification of Danish Laws. I remember I had greater than 80% precision without fiddling with settings and for less than a days worth of work setting up, don't remember the recall but it was also pretty reasonable.
I built something similar for images. It's command line based, so one could simply create a bash script to copy data to a folder based on a prediction.
I'm thinking about such a thing but for databases. "Intelligent table" where one column is a prediction from ML. Each newly inserted row is automatically used as input for ML model and prediction is inserted into database. Maybe it will be easily extendable to work with folders.
Aren't there many similar tools out there? like IBM cloud, Google's AutoML etc.,
Why did you decide to create something like this? I'm interested in hearing your opinions on what additional/new values does your service provide over the existing similar SaaS tools.
I've built mljar for myself. When I was starting working on it (2016) there were no such solutions or they were very expensive (over 50k/year).
I want to have a service where I can train many models and be able to check every model (for example check learning curves). I want a solution that can train many models in parallel in the cloud, so I don't need to heat my laptop and dont need to wait a lot. I think I achieved this. Is it better than other solutions? Hmmm, it is very similar to others (at the end, they all train some ML models), but ...
After creating the AutoML solution I come to the conclusion that AutoML is broken: https://pplonski.github.io/automatic-machine-learning-is-bro... - even if you can easily train ML model (good or bad - doesn't matter), there is still hard to use/apply machine learning in real life.
Right now, I think that AutoML is just a brick in the solution that should offer automatization. There should be a service similar to Zappier but with machine learning - you can join your data and services with ML models which live in your data ecosystem and use ML for automatization.
I did not realize you started working on this from 2016. I guess you could have easily commercialized this then since were way ahead of the competition. But I do understand that it's hard to offer this as stand-alone service. Anyways, good luck with your project!
19 comments
[ 3.1 ms ] story [ 52.6 ms ] threadMLJAR is an automated machine learning platform. With mljar, you can train great machine learning models without coding. It works with binary classification and regression tasks. It can do hyperparameters tuning and model selection. The preprocessing to deal with missing values and categorical columns is available. All models trained in the service are by default deployed in the cloud and can be accessed be REST API or python or R APIs. User can also download a model and use it locally.
Right now it is offered as a SaaS. I'm working on the open source version. The AutoML engine is already open source https://github.com/mljar/mljar-supervised
I've compared mljar performance on binary classification tasks with auto-sklearn and H2O and it works very well https://github.com/mljar/automl_comparison
In the long term, I would like to connect machine learning with databases. User will be able to train machine learning models by writing a SQL query to database. After the best model is trained (with AutoML of course!), all new rows that will appear in the database will be used for computing predictions. But first I would like to create AutoML platform :)
If Auto ML has commercial value, why are you open-sourcing it?
I think it would be a bad idea to open-source your solution, unless you plan on competing on services, rather than an Auto ML product.
Before open sourcing I was looking at Metabase and Redash solutions, and I was very impressed with their business model - I would like to achieve something similar. The goal is to be ramen profitable.
Open sourcing the core solution hugely dilutes your value proposition - I hope that you will reconsider your decision.
This is an interface truly anyone could use. Just call it "intelligent folders" or something. The user doesn't even need to know which algorithm it uses. Just split sample into training and test data at random and choose the algo that give best results.
I was working on this for text data, but then switched jobs and don't have energy to make this in my spare time right now.
https://github.com/vergeml/vergeml
I want to have a service where I can train many models and be able to check every model (for example check learning curves). I want a solution that can train many models in parallel in the cloud, so I don't need to heat my laptop and dont need to wait a lot. I think I achieved this. Is it better than other solutions? Hmmm, it is very similar to others (at the end, they all train some ML models), but ...
After creating the AutoML solution I come to the conclusion that AutoML is broken: https://pplonski.github.io/automatic-machine-learning-is-bro... - even if you can easily train ML model (good or bad - doesn't matter), there is still hard to use/apply machine learning in real life.
Right now, I think that AutoML is just a brick in the solution that should offer automatization. There should be a service similar to Zappier but with machine learning - you can join your data and services with ML models which live in your data ecosystem and use ML for automatization.