Build an AWS Lambda layer or Lambda docker container that performs time-series forecasting (ARIMA, SARIMA or other)
Budget: $30 – $250 USD
I'm working on a project that is forecasting the future values of Google Trends items, and I am in need of a robust statistical solution for performing time series prediction on hundreds of columns.
The inputs will be various columns from roughly 5 years of weekly Google Trends scores and some other social media data, the output will be a 26-week forecast from those trends scores. We're not too concerned about error, but we would like to put forth the best possible model for predicting future trends from data.
(Note: I do anticipate that most of the columns in the dataset won't have much predictive power in terms of time-series forecasting, but we want to put forward a best effort for those columns that do and come up with a solution for the columns that don't. It isn't detrimental to our process if some columns simply do not have predictive power in terms of time-series forecasting. )
I'm indifferent whether the solution is Arima, Sarima or another time-series forecasting function. Whatever you end up using, just send me a line of why that is the solution you chose. We're currently using Sagemaker DeepAR, but that is a bit slow and expensive for what we need, but it's the only scalable solution we've been able to use effectively.
The anterior column work is being done using Pandas Lambda layers saved in S3. You can see an example of the data being used in the attached csv. (note that the first column is the date column)
Let me know if you have any questions.
The inputs will be various columns from roughly 5 years of weekly Google Trends scores and some other social media data, the output will be a 26-week forecast from those trends scores. We're not too concerned about error, but we would like to put forth the best possible model for predicting future trends from data.
(Note: I do anticipate that most of the columns in the dataset won't have much predictive power in terms of time-series forecasting, but we want to put forward a best effort for those columns that do and come up with a solution for the columns that don't. It isn't detrimental to our process if some columns simply do not have predictive power in terms of time-series forecasting. )
I'm indifferent whether the solution is Arima, Sarima or another time-series forecasting function. Whatever you end up using, just send me a line of why that is the solution you chose. We're currently using Sagemaker DeepAR, but that is a bit slow and expensive for what we need, but it's the only scalable solution we've been able to use effectively.
The anterior column work is being done using Pandas Lambda layers saved in S3. You can see an example of the data being used in the attached csv. (note that the first column is the date column)
Let me know if you have any questions.