Problem with big Data in TraMineR

Question

I would like to analyse the trajectories of patients hospitalized for CVA (cerebrovascular accident) with the package TraMineR. I have a large dataset with around 25,000 trajectories observed during 365 days. There are 5 states in my data. When I try to use the seqdist function which computes the distances between sequences, I have problems with computer's memory. Some people suggested to aggregate identical sequences, compute the distance and cluster the sequences using weights. But this solution is not sufficient since I have 20000 unique sequences in my data. An other solution proposed in the web is to analyze a sample of the data. So, I have analyzed a sample of 5000 trajectories in my data. Is there another solution by using all the trajectories ? Thank you in advance for your answer.

Christelle FERREIRA

Answer 1

You can do the following, but none of the options is perfect.

• Aggregate identical sequences (as proposed), see here https://stackoverflow.com/questions/15929936/problem-with-big-data-during-computation-of-sequence-distances-using-tramine/15929937
• Compute only the lower half of the distance matrix (use the argument "full.matrix=FALSE")
• Reduce the granularity of the sequences, see here Modifying the time granularity of a state sequence . By using say weeks instead of days, you will lower the number of distinct sequences, and, hence, the impact of proposition 1 (aggregating sequence).
• Use a computer with more RAM.