nestauk / afs_early_years_labour_market_analysis Goto Github PK

• generate training dataset of EYP jobs - random sample of 6k
• create prodigy recipe to train model to extract qualifications

Join, clean and standardise curated dataset at the local authority level and getters including:

• # of children in each local authority
• # of early years providers per local authority
• rural_urban classification per local authority
• levels of deprivation per local authority
• % of children currently using EYP per local authority by age
• proportion of families with children aged 0-4 using any childcare
• proportion of families using childcare and paying it for themselves

this dataset will power the streamlit app to analyse how many more providers we will need given the new 30 hour policy.

based on one LA (or region, data permitting), using @Jack-Vines's initial analysis plan

Add a really good README to the root of the project

similar to EYP jobs, get shop assistant job adverts

Using dummy data where the data is in the format:

LSOA | age | additional_staff_needed

generate streamlit app for future shortages.

Add location, salary and skills information to the relevant jobs extract

• display future shortages in the streamlit app

Write an initial flow to try and find relevant early years jobs from OJO

As Finnula mentioned, she's keen to have additional information on job locations, including (but not limited to):

• urban/rural classification
• levels of deprivation
• birth rate (if possible - she isn't sure if it is)

I scoped out urban/rural classification + levels of deprivation at the local authority level. The data on the first two points exist in a our s3 bucket at the LA level (/inputs/loc_metadata/local_insight_January_2023_LSOA_download_1.csv). The main tasks could be to:

• scope out datasets at the ITL level that we could use instead and just merge at the ITL level
• use the 'coordinates' field to map onto LAs (you can use ONS shapefiles) then merge LA metadata
• update enrich_revelant_jobs.py to also include location metadata (can do it in the same PR as #15 I reckon)

I think whichever is most complete is likely the one to opt for.

• generate a list of similar jobs to transition into using Karlis's transitions algo (output = .txt file)
• Query OJO based on similar jobs
• Update flows to extract job adverts based on initial list