TSDF | CUDA Based implementation of a Truncated Signed Distance | GPU library

I'm trying to pre-load a DynamoDB table with records. I have about 1500 records to do. I've tried various ways to loop through only 5 but only one gets entered each time. Here is what I have so far.

The problem I have is as follows:

• Several processes are being observed and changes are recorded
• We only have the time of the change and observation value
• Since neither of the time series will have identical set of timestamps, direct comparison between those is not possible
• But we do know, that unless a change was recorded, the observation value stayed the same

How can we check what values were observed at a certain point of time across all processes? How can we produce periodical aggregates(e.g. a mean daily value) based on known information?

Here is a mockup of the data:

I want to forecast product' sales_index by using multiple features in the monthly time series. in the beginning, I started to use ARMA, ARIMA to do this but the output is not very satisfying to me. In my attempt, I just used dates and sales column to do forecasting, and output is not realistic to me. I think I should include more features column to predict sales_index column. However, I was wondering is there any way to do this prediction by using multiple features from the monthly time series. I haven't done much of time series using scikit-learn. Can anyone point me out any possible way of doing this? Any possible thoughts?

my attempt using ARMA/ARIMA:

Here is reproducible monthly time series data on this gist and here is my current attempt:

Though the common sense and literature is clear about the behaviour of strcmp():

...

Suppose I have a pandas table, with one column the stock ticker, another the date, and I want to, for each date, rescale the returns to follow the uniform distribution. Now, sklearn.preprocessing has a perfectly fine quantile_transform function for this, but I can't seem to shoehorn it into the pandas tranform or apply functionality,

The obligatory example:

What is the best way to aggregate a data frame across columns (axis=1) applying multiple functions?

Applying a list of functions works as expected with the default axis=0: