What inputs parameters is good for sagemaker estimator?

Question

I am using R on Sagemaker notebook to train a xgboost model. I've proper permissions for s3_data, that saved in s3 as s3_train.csv and s3_test.csv.

It seems expect str, csv but I could not convert dataframe read in from s3 as csv files. I used paste0(toJSON(csv_train)), it says payload too large: 413, use TraingInput, it says not serializable.

J.