Save DR and Evaluation Results

After running DR and evaluating them, one of the things that you should consider is to save them for later use. Admittedly, these results are often only an intermediate step in your research or workflow, and also, some of these computations are quite inefficient to rerun every time. Thus, this short tutorial shows you how to save your results to dist for both your reductions and your evaluations.

---

Save Reductions

There are two ways to save your embeddings. You can either you the hands-off approach and save all embeddings at once, or you can save them yourself. The former is really easy (as shown in the Quickstart Guide):

>>> results.save_all_reductions(save_dir="YOUR_DIR_PATH", delimiter=",")

For the purpose of this tutorials, results is a Reductions object with existing reductions and evaluations. This quick method will save all your reductions with their embedding names as the file names. However, you may wish to save them on your own for reasons such as needing only a few of the embeddings after ranking the DR methods or wanting to name DR in your own ways, you can do the following:

>>> results.save_reduction(name="umap", path="./my_umap.txt", delimiter=",")

This saves your UMAP embedding to your specified path with a different name.

There are a couple of other customizations you can do. One is the delimiter. If you prefer a tsv or other deliminters, simply change that parameter will help you:

>>> results.save_all_reductions(save_dir="YOUR_DIR_PATH", delimiter="\t")

Further, if a file already exists in the path specified, we don’t overwrite files by default. Rather, we throw an FileExistsError to minimize any mistakes. However, if you do wish to overwrite old files, you can specify the overwrite parameter:

>>> results.save_all_reductions(save_dir="YOUR_DIR_PATH", overwrite=True, delimiter="\t")

Overwriting will be handy sometimes without having to specifically delete files, but it’s not the default behavior, which can obviously cause issues.

---

Save Evaluations

After evaluating your DR, it’s also possible to save your results for future analyses. The interface is quite similar to save embeddings, except that you can currently save it to a csv file:

>>> results.save_evaluations(path="YOUR_FILE_PATH")

Here, you do need to give it a full path. The resulting csv file will result in a table that looks like this:

Category	Metric	Method	Value
global	spearman	PCA	0.37249992312498076
global	spearman	ICA	0.2901695885423971
global	emd	PCA	0.41181293853758305
global	emd	ICA	0.7883300728571819
local	knn	PCA	0.126
local	knn	ICA	0.126
local	npe	PCA	0.76
local	npe	ICA	0.76

You can easily run any downstream analyses using pandas or even R if you prefer.