Version 12 (modified by 15 years ago) ( diff ) | ,
---|
Spreadsheet Importer
We have access to a lot of relatively unstructured data in Spreadsheets which we need to import into Sahana. Currently this is done via a laborious Data Entry process.
We really need to be able to do Bulk Imports.
Example Data sources:
- http://spreadsheets.google.com/ccc?key=0Aq_3OYelM4ZUdHJyc0lUekZLUmEwa3ZSU2dJS3YyWWc&hl=en_GB (Data Entry tab)
- The OCHA contacts list is a task we need to do regularly, so need to be able to handle duplicates effectively. We can influence them to standardise their format to make it easier.
Options:
CSV Import
Here is a process that works, although is a bit cumbersome, so we should look to automate as much as possible & improve:
- Get a spreadsheet of good Data together (e.g. using Google Docs)
- Save as CSV
- Open in Open Office (MS Excel destroys data - do NOT use!)
- Export copies of the tables you wish to import to in CSV format, e.g.:
- /gis/location.csv
- /hms/hospital.csv
- Plan how you are going to deal with duplicates between these datasets.
- e.g. copy exported data into Spreadsheet & then clear table:
db.module_resource.truncate()
- e.g. copy exported data into Spreadsheet & then clear table:
- Plan the order of imports:
- Need to import referenced tables before those which refer to them, e.g.:
- pr_person, gis_location, or_organisation, or_office, or_contact
- Need to import referenced tables before those which refer to them, e.g.:
- Copy the data from the datasheet into the one with the column headers
- NB All column headers need to be present, named correctly, in the correct order
- No blank rows in the sheet
- Upload the 1st file to the server: /home/haiti/web2py
- Do a Test Import on Dev
- Check that data looks good
- Repeat on Prod:
vim /etc/crontab #0-59/1 * * * * www-data cd /home/haiti/web2py/ && python web2py.py -C -D 1 >> /tmp/cron.output 2>&1 vim /home/haiti/prod/models/00_db.py migrate=True rm /etc/apache2/sites-enabled/prod-ssl /etc/init.d/apache2 force-reload /home/haiti/update-prod /home/haiti/shell-prod db(db.gis_location.id>0).select().last().id db.gis_location.import_from_csv_file(open('hospital_locations.csv','rb')) rows = db(db.gis_location.id>0).select() for row in rows: if row.lon and row.lat and not row.wkt: db.gis_location[row.id] = dict(wkt = 'POINT(%f %f)' % (row.lon, row.lat)) if not row.deleted: db.gis_location[row.id] = dict(deleted = False) if not row.created_on: db.gis_location[row.id] = dict(created_on = request.utcnow) if not row.modified_on: db.gis_location[row.id] = dict(modified_on = request.utcnow) db.commit() Ctrl+D
- Set the initial ID in the 2nd import sheet to be the last+1 & then continue on
- Upload the 2nd sheet to the server:
/home/haiti/shell-prod db.hms_hospital.import_from_csv_file(open('hospitals_import.csv','rb')) rows = db(db.hms_hospital.id>0).select() for row in rows: if not row.deleted: db.hms_hospital[row.id] = dict(deleted = False) if not row.created_on: db.hms_hospital[row.id] = dict(created_on = request.utcnow) if not row.modified_on: db.hms_hospital[row.id] = dict(modified_on = request.utcnow) db.commit() Ctrl+D ln -s /etc/apache2/sites-available/prod-ssl /etc/apache2/sites-enabled /etc/init.d/apache2 force-reload vim /home/haiti/prod/models/00_db.py migrate=False vim /etc/crontab 0-59/1 * * * * www-data cd /home/haiti/web2py/ && python web2py.py -C -D 1 >> /tmp/cron.output 2>&1
ToDo
- Do data-de-duplication
- hook into the Model's validators
- Provide a Controller to allow web-based access at: /admin/import_data (The current implementation broke at some point)
- Use functions imported from a new
modules/s3gis.py
for the WKT-creation
Proposed Automation for Spreadsheet (csv) Imports
Requirements (as per ToDo above)
- Allow admins to import data for a model via the Web interface
- Hook into the Model's validations
- Provide ability to de-duplicate (phase 2)
- Keep track of when and where data was imported from (maybe just for a few days to help with conflict resolution, etc).
Outline
- Allow an admin to start an import 'Job' for a particular Model
- Jobs are described by a free-form description field, and an uploaded .csv file.
- Extract csv column names, store in job. Set job state to 'new'
- Attempt to match csv column names to model column names
- Present list to user, allow them to change csv column <-> model column mapping
- Once user confirms column mappings, set job state to 'processing'
- Background cronjob picks up all import jobs in state 'processing', for each data row:
- Parse into dictionary of {'db_column': value, ..} using column name mapping in Job record
- Validate data using SQLFORM accepts method with dbio=False (no database activity)
- Store row into ImportLine model:
- job_id
- line_no
- valid (true, false)
- status (ignore, import, imported)
- data (pickled dictionary from above)
- Background job status set to 'processed'
- Ajax keeps the job page updated with the background job status, until completed
- Display table of rows and validation status to user, default status to ignore for rows that fail validation, import for ones that pass.
- Future enhancements will most likely occur at this step
- Allow user to edit/fix fields that failed validation and re-validate
- Present list of likely duplicate entries alongside each imported entry
- Future enhancements will most likely occur at this step
- User can de-select any row to ignore it, ImportLine status set to 'ignore'
- Once user has checked all rows, click 'Import', set job state to 'import'
- Background cronjob picks up all import jobs in state 'import', for each ImportLine with status=import:
- Insert data via SQLFORM.accepts method (dbio defaults to True this time, put into DB)
- pass special onvalidation method to accepts, to covert lat, long to wkt as shown above.
- Update ImportLine status to imported.
- Insert data via SQLFORM.accepts method (dbio defaults to True this time, put into DB)
- Background job status set to 'completed'
Google Spreadsheets API
Excel
- http://pypi.python.org/pypi/xlrd xlrd]
- Example Code - not the cleanest
Other
Note:
See TracWiki
for help on using the wiki.