Predict with batch prediction in Vertex AI

[@rupjitchakraborty8012]

Thank you so much. This was really helpful.

[@ludelaire]

Thanks a lot, your tutorials have been of help to me. Keep it up! :)

[@cloud4datascience772]

Thanks for the kind words :)

[@linkawaken]

Nice tutorial. Given a batch prediction job, I would want to convert the instances payload to a DataFrame. Do you know if we have any control over the instances format (for example, including the column names from BigQuery) that allows one to construct a DataFrame with named columns e.g. for preprocessing in a custom model?

[@cloud4datascience772]

Thanks, I am glad you liked it!
There isn't much of a way that we can affect how the predictions are delivered to us. According to the documentation (cloud.google.com/python/docs/reference/aiplatform/latest/google.cloud.aiplatform.BatchPredictionJob) there are only two destinations where your results can be saved: BigQuery and Cloud Storage, if you want to do anything additional with the predictions you need to first grab it from one of those and do all the transformations starting from BigQuery table or file in Cloud Storage.

[@Videodecumple]

How do you know how much is going to cost using Vertex AI for training data (Text Classification) ? Is there any prices for privat users (researchers) ? Thanks!

[@cloud4datascience772]

If you register first time to Google Cloud Platform you will get $300 credits to experiment. You can use this to try to assess the potential costs in the future. Generally, the batch prediction cost depends on the machine which is used for the scoring and the time needed to score all the data. Usually, it's quite cheap compared to online predictions. Feel free to review the linked documentation for the specific price numbers => cloud.google.com/vertex-ai/pricing#prediction-prices

[@Just_lirycs]

is a good tutorial but i have a question how can I configurate the input_storage_format , because i just have one option jsonl

[@cloud4datascience772]

If you have just jsonl format I am guessing that you've trained your model on dataset which was not tabular (e.g. image data) is that correct?

[@Just_lirycs]

@@cloud4datascience772 yes sr , I was reading the library and this is my mistake , thanks again

[@cloud4datascience772]

@@Just_lirycs Although I haven't recorded the tutorial yet I already have a code on my github where I do the batch prediction on image dataset. There are 3 notebooks: for creating input file, making batch prediction and tranforming the output from the batch prediction job. Maybe you will find something from there useful :)
github.com/rafaello9472/c4ds/tree/main/Predict%20with%20batch%20prediction%20in%20Vertex%20AI%20-%20Image%20Classification

[@Just_lirycs]

@@cloud4datascience772 Thanks sr