Best practice for using Julius for data analysis

I am trying to introduce Julius as a tool for doing simple data preparation and analysis to non-technical people.

However I have concerns over the consistency of the analysis. For eg. If I have datasets of 5 different brands of shoes, each with the same schema and I want to do sentiment analysis on the reviews of these brands using ensemble techniques with NLP models. Telling Julius a prompt like “Do sentiment analysis of these 5 datasets using ensemble techniques with NLP model” will result in 3-5 different models being selected for use. This is fine since all brands will be processed using the same sets of models.

But in the future, if I want to include a 6th brand for analysis and runs the same prompt again, it may result in different models being used and thus giving different results.

I am wondering if it would be a better approach to use Julius to generate the Python code for my initial run and save it so that I can run the code separately in a different environment for consistency sake

1 Like

It is definitely a good idea to save the python code as it would keep consistency in the figure/analysis output. Additionally, informing Julius on what model you wan to run can sometimes circumvent this issue. But, directly copying and pasting the python code is probably the best method in your situation.