Feature Request: Improve model search functionality

[anamika-yadav99]

Is your feature request related to a problem? Please describe.

yes, every once in a while there's a typo or user enters a wrong spelling. model search function in such cases return null table which could be confusing for the user.

Describe the solution you'd like.

I'm still exploring for a good solution. perhaps, a simple approximate string matching algorithm which would return the closest matching model name in the ModelSearch function and also on fetch command.

Describe alternatives you've considered

No response

Additional context.

No response

[anamika-yadav99]

@GemmaTuron would this be a feature you would like to have?

[dauinh]

Hi @miquelduranfrigola @anamika-yadav99, can i work on this feature as well?

[dauinh]

I found this article to be helpful. Do you think we can do a 'did you mean' and implement a Trie to find the correct command?

[anamika-yadav99]

Hi @dauinh Thanks for the help. The article was insightful. I'm also thinking of somewhere along the similar lines. Compare the distance between the input model name and the model names in the hub or local and then return the closest matching model name. This is basically string matching algorithm only. Let's wait for what @miquelduranfrigola and @GemmaTuron have to say on this.

[dauinh]

Hi @dauinh Thanks for the help. The article was insightful. I'm also thinking of somewhere along the similar lines. Compare the distance between the input model name and the model names in the hub or local and then return the closest matching model name. This is basically string matching algorithm only. Let's wait for what @miquelduranfrigola and @GemmaTuron have to say on this.

Oh, i see what you mean... Thank you for clarifying!

[GemmaTuron]

Hi @anamika-yadav99

This sounds fantastic. My only concern is if the search function for close models would slow down the package too much, but we can certainly explore options. Can you add this to your application as a task?

[miquelduranfrigola]

This is interesting overall. Because we are using AirTable to store our models, my immediate guess is that there may be some AirTable functionality for fuzzy string matching or something like this in a text-based search. If this is the case, we should certainly take this avenue.

[anamika-yadav99]

Hi @miquelduranfrigola I looked into Airtable API. Unfortunately it doesn't support Fuzzy Search. Fuzzy search can easily be achieved using fuzzywuzzy python library but I guess, you wouldn't want any more dependency in ersilia. So, I have modified the Search function in Ersilia to include an algorithm which calculates the distance between the input from cli and data which is imported from hub in model catalog. It doesn't take a lot of time to perform the search through the table. The time complexity of algorithm is mn. I think similar solution would work for model fetch command as well. Shall I proceed with my approach so far?

[miquelduranfrigola]

Hi, @anamika-yadav99 thanks. This sounds like a good approach. Let's go for it. I am more interested in the catalog command than in the fetch command. So let's try with catalog, to start with?

[anamika-yadav99]

@miquelduranfrigola I'm almost done with the catalog command. I'll raise the PR by tomorrow.

[anamika-yadav99]

Hi @miquelduranfrigola I'm sorry for the delay. I was caught in some college work which took longer than expected. I have raised a pr for catalog command. Looking forward to your feedback.

[miquelduranfrigola]

Hi @anamika-yadav99 this is great. I have approved PR #262 and closing the issue now! Many thanks.