Resumable Training #1130
Merged
Resumable Training #1130
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
jimzers
commented
Jan 18, 2023
sleap/nn/training.py
Outdated
| # and load it into the current model | ||
| previous_model_path = os.path.join( | ||
| self.config.model.base_checkpoint, "best_model.h5" | ||
| ) # TODO: add flexibilty to resume from any checkpoint (e.g. latest_model, specific epoch, etc.) |
There was a problem hiding this comment.
TODO: add flexibilty to resume from any checkpoint (e.g. latest_model, specific epoch, etc.)
There was a problem hiding this comment.
check if base checkpoint is a file and just use that
|
TODO: How does one set |
Codecov Report
@@ Coverage Diff @@
## develop #1130 +/- ##
===========================================
+ Coverage 70.10% 71.94% +1.84%
===========================================
Files 131 131
Lines 22984 23026 +42
===========================================
+ Hits 16113 16567 +454
+ Misses 6871 6459 -412
📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
* add cli options for resumable training * add test for cli resume training * black linting for cli resumable training * simplify resumable checkpoint CLI fn to a single CLI arg (#1132) * simplify resumable checkpoint CLI fn to a single CLI arg * Adam/resumable training 3 (#1150) * correct path of labels_path for test_training * add resume training to gui * add train from scratch message * Add finishing touches to resumable training PR (#1150) (#1168) * Refactor/update 'use trained' and 'resume training' checkbox logic * Simplify checkbox logic and reset model field when resume training * Reset checkboxes upon changing config selection * Handle case for updating TrainingEditor when sender is not a checkbox * Add complete state space GUI test for checkboxes * Finish combobox test * Test that form is reset * Remove straggling TODO --------- Co-authored-by: roomrys <[email protected]> Co-authored-by: jimzers <[email protected]> --------- Co-authored-by: jimzers <[email protected]> Co-authored-by: Liezl Maree <[email protected]> * Return trainer from sleap-train and check that trainer configured correctly * Add CLI documentation for website --------- Co-authored-by: jimzers <[email protected]> Co-authored-by: Liezl Maree <[email protected]>
talmo
approved these changes
Feb 14, 2023
roomrys
added a commit
that referenced
this pull request
Feb 24, 2023
* GUI Training: Use hidden params from loaded config (#1053) * Add optional unragging arg to model export (#1054) * Fix config option to `split_by_inds` (#1060) * Convert training, validation, and test to Labels object * Add test for split_by_inds * Use Labels.extract instead of Labels(List[LabeledFrames]) * Tracking: robust assignment of the best score to an instance (#1062) * Set max instances for top down models (#1070) * Add optional unragging arg to model export * Add option to set max instances for multi-instance models * Fix test * Don't create instances during inference if no points were found (#1073) * Don't create instances during inference if no points were found * Add points check for all predictors * Fix single instance predictor logic and test * Add tests for all predictors Co-authored-by: roomrys <[email protected]> * Add one-line fix to VideoWriterSkyvideo (#1082) * Fix parser for sleap-export (#1085) * Refactor commands to load project as `AppCommand`s (#1098) * Add working Proof of Concept * Create command class for loading project * Split `LoadProjectFile` as a subclass of `LoadLabelsObject` * Reroute last existing reference * Remove debugging code * Flexibly resize input layer of `tf.keras.Model` upon loading trained model (#1084) * Add initial implementation (auto output stride problematic) * Add to load_predictor test (error when auto-compute output stride) * Use output stride from config instead of auto-computing * Fix output-stride/padding modulo error and do not resize on export * Fix resizing bug in multi-class predictors * Non-functional clean-up * Rename new input layer to original name * Add inference integration test * Minimize config surgery, generalize layer iteration Co-authored-by: Talmo Pereira <[email protected]> * Add Option to Make Trail Shade Darker/Lighter (#1103) * Make trails 60% darker than track color * Add menu option for shade of trails * Remove unexpected indent (fat-fingered) * Create signal that updates plot instead of removing and replotting items (#1134) * Create signal that updates plot instead of redrawing * Remove debug code * Non-functional self-review changes * Fix symmetric skeletons (via table input) (#1136) Ensure variable initialized before calling it * Nix export of tracking results (#1068) * [io] export tracking results to NIX file * [io] nix added to export filter only if available * [nixio] refactor, add scores link data as mtag * [nixio] speeding up export by chunked writing * [nixio] rename point score to node score * [nixio] fix missing dimension descriptor for node scores * [export analysis] support multiple formats also for bulk export * [nixio] export centroid, some documentation * [nixio] fix double dot before filename suffix * [nixio] fix bug when not all nodes were found * [nixio] housekeeping * [nix] add nix analyis output format to convert * [nix] tiny fix, catch file write error and properly close file * [inference] main takes optional args. Can be imported to run inference form scripts * [convert] simplify if else structure and outfile handling for analysis export * [nix] use pathlib instead of os * [nix] catch if there are instances with a None frame_idx ... not sure why this occurred. The nix adaptor cannot save instances that are not related to a frame. * [nix] move checks to top of write function * [nix] use absolute imports * [nix] use black to reformat * [commands] revert qtpy import and apply code style * [convert] use absolute imports, apply code style * [commands]fix imports * [inference/nix]fix linter complaint, adjust nix types for scores * [nix] add test case for nix export format * [nix] extended testing, some modifications of adaptor * [skeleton] add __eq__ to Skeleton ... make Node.name and Node.weight instance variables instead of class variables * [nix] add nixio to requirements, remove unused nix_available, ... allow for non-unique entries in node, track and skeleton. Extend node map to store the skeleton it is part of * [nix] make the linter happy * [Node] force definition of a name Co-authored-by: Liezl Maree <[email protected]> * [nix] use getattr for getting grayscale information Co-authored-by: Liezl Maree <[email protected]> * [nix] fixes according to review * [convert] break out of loop upon finding the video Co-authored-by: Liezl Maree <[email protected]> * [commands.py] use pathilb instead of splitting filename Co-authored-by: Liezl Maree <[email protected]> * [dev requirements] remove linebreak at last line * [skeleton] revert attribute creation back to original * [nix] break lines in class documentation * Ensure all file references are closed * Make the linter happy * Add tests for ExportAnalysis and (docs for) sleap-convert Co-authored-by: Liezl Maree <[email protected]> * Fix body vs symmetry subgraph filtering (#1142) Co-authored-by: Liezl Maree <[email protected]> * Handle changing backbones in training editor GUI (#1140) Co-authored-by: Liezl Maree <[email protected]> * Added scaling functionality for both the instances and bounding box. (#1133) * Create VisibleBoundingBox class. * Added instance scaling functionality in addition to bounding box scaling functionality. * Update sleap/gui/widgets/video.py Co-authored-by: Talmo Pereira <[email protected]> * Update sleap/gui/widgets/video.py Co-authored-by: Talmo Pereira <[email protected]> * Update sleap/gui/widgets/video.py Co-authored-by: Talmo Pereira <[email protected]> * Update sleap/gui/widgets/video.py Co-authored-by: Talmo Pereira <[email protected]> * Update sleap/gui/widgets/video.py Co-authored-by: Talmo Pereira <[email protected]> * Added new testing for scaling operation and simplified VisibleBoundingBox class code. * Added type hinting to the scaling padding and removed erroneous bounding rect initialization. Co-authored-by: Talmo Pereira <[email protected]> Co-authored-by: Liezl Maree <[email protected]> * Add better error message for top down (#1121) * Add better error message for top down * Add test for error message * Raise different error, fix test * Hotfix for video save #1098 (#1148) * Add a hotfix for #1098 * WIP: Add test for detecting changes on load * Finialize change on load test * Remove unused imports * Skip test if on windows since files are being used in parallel * Add central padding to SizeMatcher (#1129) * add center padding to size matcher * add test for center padding * add ensure_float option to inference layer * reformat resizing and test_resizing * Remove redundant operation * Replace existing constants with fixtures --------- Co-authored-by: Liezl Maree <[email protected]> * Added MoveNet as an external model reference (#1141) * add center padding to size matcher * add test for center padding * add ensure_float option to inference layer * reformat resizing and test_resizing * add MoveNet as an external model inference * add the movenet to the from_model_paths * add tests * add comments to movenet predictor * add tensorflow_hub to the requirements.txt * modified default video path * resolved most of the comments except expanding the predictor * expanded Predictor.from_model_paths function to include any pre-trained models. * add test_load_model * added from_trained_models in class Predictor and modified test_load_model for it. * modified test_load_model to be more generalized. * moved pretrained model from Predictor.from_trained_model to Predictor.from_model_paths and added a test for it. * Fix Predictor.from_model_paths and tests * Rename load_movenet_model to make_model_movenet * minor clean-up * Remove redundant operation * Replace existing constants with fixtures * Handle loading movenet models via load_model API * Clean-up doc strings --------- Co-authored-by: Liezl Maree <[email protected]> * Resumable Training (#1130) * add resume training functionality * add testing function for resume training functionality * linting black * Resumable Training 2 - CLI Options (#1131) * add cli options for resumable training * add test for cli resume training * black linting for cli resumable training * simplify resumable checkpoint CLI fn to a single CLI arg (#1132) * simplify resumable checkpoint CLI fn to a single CLI arg * Adam/resumable training 3 (#1150) * correct path of labels_path for test_training * add resume training to gui * add train from scratch message * Add finishing touches to resumable training PR (#1150) (#1168) * Refactor/update 'use trained' and 'resume training' checkbox logic * Simplify checkbox logic and reset model field when resume training * Reset checkboxes upon changing config selection * Handle case for updating TrainingEditor when sender is not a checkbox * Add complete state space GUI test for checkboxes * Finish combobox test * Test that form is reset * Remove straggling TODO --------- Co-authored-by: roomrys <[email protected]> Co-authored-by: jimzers <[email protected]> --------- Co-authored-by: jimzers <[email protected]> Co-authored-by: Liezl Maree <[email protected]> * Return trainer from sleap-train and check that trainer configured correctly * Add CLI documentation for website --------- Co-authored-by: jimzers <[email protected]> Co-authored-by: Liezl Maree <[email protected]> * Small (final?) revisions and fix test * Revert changes to fixture --------- Co-authored-by: jimzers <[email protected]> Co-authored-by: Liezl Maree <[email protected]> * GenericTableModel/View improvements (#1163) * [dataviews] GenericTableModel/View improvements ... * GenericTableView got a new argument specifying whether the ellipsis for long cell content should be right (old behavior, default) or left useful for long content such as the filenames in the video table. * GenericTableView uses all the space that available to the table. * The model's data function returns the full cell content to be shown as tool tip text. * [gui/app] set the ellipsis to be on the left for long table contents --------- Co-authored-by: Liezl Maree <[email protected]> * Add Skeleton Templates (#1122) * Update docs: change 'M1' to 'Apple Silicon' (#1188) * Bump to 1.3.0a0 (#1189) --------- Co-authored-by: sheridana <[email protected]> Co-authored-by: getzze <[email protected]> Co-authored-by: Talmo Pereira <[email protected]> Co-authored-by: Jan Grewe <[email protected]> Co-authored-by: Sean Afshar <[email protected]> Co-authored-by: Jiaying Hsu <[email protected]> Co-authored-by: Adam Lee <[email protected]> Co-authored-by: jimzers <[email protected]> Co-authored-by: Jan Grewe <[email protected]> Co-authored-by: Aaditya Prasad <[email protected]>
11 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
#1117 PR 1: Add functionality to resume training
Types of changes
Does this address any currently open issues?
Outside contributors checklist
Thank you for contributing to SLEAP!
❤️