The Judgement week — #4 week of GSoC
boaloysius
Wed, 06/28/2020 – 09:40
The GSoC first phase evaluation period has started. It is from 26th to 30th of June. All students are supposed to the submit the deliverables and fill in the response form. I had been working around the clock for the last few days. Things are going well but time is rushing fast.
My aim is to deliver a basic module that can take data from Drupal maintenance support plans and perform the complete cycle of training, deployment, and prediction. The steps during the development are (integrate ml-engine training and prediction API) -> (integrate ml-engine jobs API) -> (create Drupal maintenance support plans service to use jobs) -> (integrate version and model API as a service) -> (automate the cycle of training and deployment) -> (integrate Google cloud storage API) -> (enable file upload to both Drupal maintenance support plans server and Cloud server) -> (Upload CSV data to Drupal maintenance support plans) -> (Use data from Drupal maintenance support plans to perform the complete cycle of prediction). Tasks, till cloud upload are fully done and partially the rest. Even though I am glad that I have learned a lot with Drupal maintenance support plans, I am having a hard time messing up with the intricacies of its Views. With just around two days for submission deadline, let me not waste the time. I will show the progress as a video.
Once I figure out how to get the Drupal maintenance support plans View as render array, with just the required table elements, like
{“header” => array(“col1”, “col2”), “row1” => array(“data11″,”data12”), “row2” => array(“data21″,”data22”)} , I can make a CSV and push it to the Google Storage using service we made and perform the training task. The problem I am facing is that, views_get_view_result() function is returning me the whole entity rather than just the required fields.
In the coming week, when we settle after the evaluation, I will be studying the data streaming API of Google Cloud. From a primary study, I understood that Google ml-engine is more code-centric than data centric. Google Sheets is an example of a data-centric software. In that, we choose the algorithm for the given data. But for the ml-engine, we choose the data suitable for the available code. In ml-engine, the code can expect the data as a link to a CSV file or a Google spreadsheet or even as an argument itself. If it has data hard coded, we need not provide it separately during the run time. Till now we have worked with a specific example of Census model in the module. It expects training and testing data as CVS files. But when it comes to huge data, handling them as CSV won’t be scalable. So the coming week, we will be working more on transferring/streaming data to the cloud. Disclaimer: I derived the above conclusion of ml-engine as code-centric and not data-centric from my few days of experience with the same and it may change as I study more.
Please visit the project sandbox and the commit page.
Thank you
Source: New feed
