[HN Gopher] Show HN: Machine learning automation from creating t...
       ___________________________________________________________________
        
       Show HN: Machine learning automation from creating to using models
       in production
        
       Author : nidhaloff
       Score  : 54 points
       Date   : 2021-06-23 11:27 UTC (2 days ago)
        
 (HTM) web link (github.com)
 (TXT) w3m dump (github.com)
        
       | barbazoo wrote:
       | Awesome tool!
       | 
       | A few things I found so far:
       | 
       | - When running "fit", I get an error when I use the "--yaml_file"
       | parameter mentioned in the readme: "Unrecognized argument ->
       | yaml_file". Using "-yml" works
       | 
       | - When using the model for prediction, what format should
       | "path_to_your_test_dataset.csv" be in? Is that a csv in the same
       | format as the training set except for the column we want to
       | predict? Edit: I just found
       | https://github.com/nidhaloff/igel/blob/v0.4.0/examples/data/...
        
         | nidhaloff wrote:
         | Hi, thanks for your feedback! you are right, the
         | argument/parameter is actually yaml_path and not yaml_file.
         | This is a typo in the docs/readme, thanks for this finding ;)
        
       | nidhaloff wrote:
       | Hi all,
       | 
       | Igel is a delightful tool to help you create, validate and use
       | machine learning model (also in production) without writing code.
       | You can use the integrated command line or the graphical
       | interface.
       | 
       | Igel uses FastAPI and uvicorn to serve your trained model, due to
       | their high performance.
        
         | mkl wrote:
         | Hi, this looks interesting to me, but am I correct that it
         | doesn't support image data?
         | 
         | I spotted a typo in the README: "63 different machine learning
         | model in igel" should say "models".
         | 
         | The feature "Supports all state of the art machine learning
         | models" seems absurd. How could it possibly be true? Surely
         | there are many SOTA models not in SKLearn?
        
       | kiranaxonator wrote:
       | I would also like to recommend Axoantor Low Code Platform
       | https://axonator.com/low-code-platform
        
         | rush86999 wrote:
         | this is not relevant
        
         | nidhaloff wrote:
         | Hi, thanks but how does this link to igel or to AI/ML
        
       | elishay wrote:
       | Cool project! Can probably integrate with SimpleML (author) to
       | add persistence and versioning. Happy to help if you're
       | interested https://github.com/eyadgaran/SimpleML
        
       | th0ma5 wrote:
       | I unfortunately would not recommend running this in production. I
       | don't see anything like debugging or centralized or distributed
       | logging either way, no performance metrics either from a raw code
       | or stats performance measurements. I could go on, but there are
       | many software projects out there that try to productionalize ML
       | and know nothing about stats, but this is the opposite, it seems
       | knowledgeable about statistical methods more than most of these
       | kinds of things, but doesn't seem to have a production mindset.
        
         | nidhaloff wrote:
         | Thanks for the feedback! When I first started the project, it
         | was not thought for production. Just for fast prototyping and
         | experimenting with no efforts at all. However, users liked the
         | tool and started requesting more features including support for
         | serving models and eventually deploying (e.g this issue
         | https://github.com/nidhaloff/igel/issues/62)
         | 
         | I agree with your point of vue. However, igel is fairly new and
         | evolving fast. Using igel to serve trained model is a new
         | feature that was implemented in the new release so igel has a
         | long way to go in order to be a solid product for production
         | use.It will surely get more mature with time.
         | 
         | Finally, notice that I didn't recommend running it in
         | production. Just mentioned that it is possible and takes no
         | efforts at all. However, if the user generated a trained model
         | then anything can be done with it from there. Technically, the
         | user can implement his/her own server and use the model as
         | wanted. Obviously, users should do that if they want more
         | control.
        
       ___________________________________________________________________
       (page generated 2021-06-25 23:01 UTC)