summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorJSDurand <mmemmew@gmail.com>2023-06-02 14:49:53 +0800
committerJSDurand <mmemmew@gmail.com>2023-06-02 14:49:53 +0800
commit83c66eb77c6affaa9ac4fabd808556613c5bf973 (patch)
tree87aa01730f07e2010746163f35c0a010e569727a
parent662817e6367a865a2d86a99581172cc45f585807 (diff)
diary
* DIARY: Added a diary that might serve as a record of my thoughts.
-rw-r--r--DIARY25
1 files changed, 25 insertions, 0 deletions
diff --git a/DIARY b/DIARY
new file mode 100644
index 0000000..13c9c6e
--- /dev/null
+++ b/DIARY
@@ -0,0 +1,25 @@
+This is a "diary" that records my thoughts when trying to debug the
+forest manipulations. Basically the forests are wrongly "merged", and
+produce weird clones.
+
+Right now I have finished refactoring most of the codes, but still
+fail to perceive why the forests behave in that way. I even finished
+a proto-type of my procedural macros which help me reduce
+boiler-plates but are not necessary.
+
+Right now I think I need more refined control over the details of the
+forests manipulations. To be more precise, I need to be able to
+observe, step by step, how the chain-rule machine manipulated the
+forests. This means in particular to be able to produce graphs of
+half-way forests whenever I want. I have not yet figured out how to
+do this exactly.
+
+The rough plan is to compress all steps of graphs into an "incremental
+format" that stores not just forests information, but how they were
+created, and allow the users to re-create the forests step by step.
+
+Simply put, this is just another feature that I would like to have for
+the end-users, but postponed to later since that was not essential for
+my developments. Now this seems to be quite important for me to
+properly observe the forests, so it is perhaps time to implement this
+feature first.