WIP: Syntree visualisation with graphviz #19

chrulm · 2022-12-19T09:49:41Z

Visualize syntree with graphviz instead of mermaid.js.

Works by plotting all chemicals as pngs and then uses graphviz to create a single image of the entire syntree.

Reason for change:

Found it too clunky to render graphviz correctly in VS Code with mermaid preview extension and export them.

To do:

update readme
write (some) tests
figure out how to plot in higher resolution, ideally svg
add target node to plot (if present/available as smiles)
add wrapper or script to plot multiple syntrees with sane defaults
add direct display in jupyter notebooks

Inspiration from: https://github.com/MolecularAI/aizynthfinder/blob/9e44989213c11f1bb647a00b8756e0c76a8f4b52/aizynthfinder/utils/image.py

Rt1 order invariance

Fix rxnmask for merging

- moves hparam config into respective file to `get_args()` - hence deletes yaml files - disable `TQDMProgressBar` on HPC - allows class weights for `act` - save checkpoints/logs/etc to new dir set by `init_save_dir()`

Wandb sweep

- also bumps rdkit to 2022.03 for compatibility with datamol

Revive some unittests

Refactor inference

add: test iclr models can be loaded from ckpts

Misc fixes

chrulm added 30 commits October 17, 2022 19:03

adds wandb

1edbc9d

separate file IO from splitting into Xy

557218d

extract Xy data with 2nd reactant for rt1

14003a2

adds tqdm progress bar

aafb728

allows to disable rdkit logs

e460216

adds cosine distance as loss fct

638c58a

bug fix: do not raise exception when checking for rxn mask

43cfe46

Merge pull request #8 from chrulm/rt1-order-invariance

0a85d78

Rt1 order invariance

Merge pull request #7 from chrulm/fix-rxnmask-for-merging

770cda3

Fix rxnmask for merging

delete unused constants

f6287e3

correctly detect number of available cpus on HPC

6a15ce6

adds: input validation for mlp

9054f82

clean up type hints + comments

e306082

refactor config to yaml files

13237b9

unifiy code to train act, rt1, rxn, rt2:

5979806

- moves hparam config into respective file to `get_args()` - hence deletes yaml files - disable `TQDMProgressBar` on HPC - allows class weights for `act` - save checkpoints/logs/etc to new dir set by `init_save_dir()`

Merge pull request #9 from chrulm/wandb-sweep

1b308e6

Wandb sweep

adds numba ; speeds up BallTree

c2efa2c

format

3b13df3

deprecate mol_fp() for fp_embedding(), is faster

242d6a1

delete default args

ce8524d

allows to change logger level to DEBUG for syntree generation

604e163

implement reading chembl data

060f464

delete default args

a4c0518

fix seed

cb74be9

genearte + save summary on syntree filtering

5c8c538

new dependency: add datamol

f8cff56

- also bumps rdkit to 2022.03 for compatibility with datamol

update comments

6da7e6e

adds input validation

09e20c0

update print methods for SyntheticTree

99d4dda

update type hints

ab5cae7

chrulm added 29 commits November 25, 2022 20:17

refactor tests for genetic algorithm, use pytest

a3d0797

delete old unittest tests (unused and outdated)

96b042b

add: basic sanity tests for syntrees and fingerprints

5ede95e

Merge pull request #11 from chrulm/revive-some-unittests

69509f2

Revive some unittests

force install gpu version of pytorch

a864dac

update env

bcbc4bd

add: clean chembl smiles and delete invalid smiles

d2e17dd

add: split SyntheticTreeSet by depth

562f582

adds basic fcts for parallelisation

e8723aa

refactor: re-write as SynTreeDecoder (OOP)

92598bc

add: simplify interface

f234b6d

Merge pull request #12 from chrulm/refactor-inference

16b76be

Refactor inference

fix template

4428d0d

change: max_depth of syntree 8

1a45202

fix: enforce pytdc<0.3.8 due to bug on their end

e6e812e

change: use chunked_parallelfor all mp

ddb7ce6

fix: ensure we create output folder

58bd398

temporary fix for dimenions

dbedda1

fix: correct wandb group for rxn network

f5df0e7

fix: use as many cpus as we allow

4609393

fix: dont save index with pd.to_csv

9cf18d0

fix: adds missing keyword-only arguments to load iclr checkpoints

6af4ac6

add: test iclr models can be loaded from ckpts

adds comments

29d9d0f

fix: unify name for knn validation loss

2326b89

explicit about where to add dropout

dba58af

add: number of action as better proxy for complexity then depth

5d1e503

format

f9ef890

Merge pull request #13 from chrulm/misc-fixes

a5e9c13

Misc fixes

new: wip: plot syntree with graphviz

906d18a

chrulm closed this Dec 19, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

WIP: Syntree visualisation with graphviz #19

WIP: Syntree visualisation with graphviz #19

chrulm commented Dec 19, 2022

WIP: Syntree visualisation with graphviz #19

WIP: Syntree visualisation with graphviz #19

Conversation

chrulm commented Dec 19, 2022