0

Are there tools that would run many machine learning fits without much setup?

Such tool could be useful for discovering what kind of model could be the best fit, without trying each of them manually.

mavavilj
  • 416
  • 1
  • 3
  • 12
  • There may, or may not be, lots of tools. Depends on the definition of "without much setup". Can you be clearer? What do you consider "too much"? There are tools as part of AWS and Azure and Google Cloud. You can use h2o automl, data bricks, scikit-learn, R tidymodels, R caret, SAS, SPSS, Java Weka, etc. We can go on. – Craig Aug 09 '22 at 12:35
  • @Craig Ideally I was envisioning something between 1-5 LOC: read data to table and do a single function call with e.g. the number of models to produce. Or at least so that 1 LOC would produce all parts of a particular algorithm so that one doesn't have to figure out what accuracy metrics to use etc. – mavavilj Aug 09 '22 at 13:40
  • A lot of tools (AWS, Google Cloud, Azure, Data Robot) can do this in a few lines of code. But you still need to setup AWS/Google/Azure/DataRobot. Same with h2o automl, R caret, etc. Need to setup R or H2o (then R or Python or Java or use their webui). A google search for automl can get you a lot of info. But you still need to setup the tool. For the metrics, choosing the appropriate metric(s) to the problem you are solving is important. Otherwise a model not optimal for your problem might be chosen. – Craig Aug 09 '22 at 14:00
  • @Craig But many methods do come with their default, intended performance metric. Or possibly all the methods would use the same metric to allow comparison. – mavavilj Aug 09 '22 at 14:01
  • If building classification, for example, which of the metrics best fit the [problem](https://scikit-learn.org/stable/modules/model_evaluation.html#model-evaluation)? Is there only 1 cutoff value used or multiple? Is there a weighting? Is everyone OK if the model is wrong where the loss is low (take fraud for example) but the model must be accurate on high loss. I guess some of the automl tools can use the metric you choose - I have not used all of these tools. – Craig Aug 09 '22 at 14:10

0 Answers0