This repository has been archived by the owner on Jan 22, 2024. It is now read-only.
Generate Synthetic dataset. #88
Open
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
dbieber
approved these changes
Dec 31, 2021
scripts/setup-tpu.sh
Outdated
| @@ -5,6 +5,8 @@ git clone https://[email protected]/googleprivate/compressive-ip | |||
|
|
|||
| @@ -53,19 +57,31 @@ def generate_dataset( | |||
| else: | |||
| test_file_writer.write(record_bytes) | |||
|
|
|||
| def get_target_index(target, keep_errors_only): | |||
| error_idx_offset = 1 if keep_errors_only else 1000 | |||
There was a problem hiding this comment.
Q: Is it OK having error indexes hardcoded here, or will it make maintenance hard later?
core/distributed/sweep.py
Outdated
| @@ -180,18 +180,19 @@ def main(experiment_id=None, study_id=None, dataset_path=None, skip_create=False | |||
| if experiment_id is None: | |||
There was a problem hiding this comment.
merge conflict. don't check in.
- Adds edge_* features to dataset, 6 edge types - Sweeps - GGNN implementation (tests timing out still) - Adds run_test (option for no subsampling in run_test) to eval 1 epoch - Adds inspect_edges to analyze_data
A dry-run sweep generates the commands for the sweep without running them. This is useful for resuming old runs on different machines than they were originally run on, or for resuming just a subset of old runs.
Allows setting number of training steps and seed, so we can run multiple runs of a single model to compute variance of the metrics. Colab for generating commands is here: https://colab.research.google.com/drive/1axwI8dGJ1_wTLIKJsLEx0FazHaPXEu72#scrollTo=IroRFMZyl6kR&uniqifier=3
- Configs for GGNN: - config.ggnn_use_fixed_num_layers = True - config.ggnn_layers = 3 - new dataset with edge info for GGNN - generates edge_sources_shape on the fly - We're filtering the same examples as before - Sweep for ggnn experiments - Overwriting top checkpoints after preemption (better would be a new checkpoints dir) to avoid failure on restart - Supports both fixed num layers and num_steps num layers for ggnns - Code for generating sampled test set with roughly equal error and no error examples
The sgd optimizer state has changed, so our naive existing method of loading old checkpoints doesn't always work. This works around that for test. The restore logic now skips init (was unnecessary and slow anyway), loads the old checkpoint state, but then only keeps the params, dropping opt_state. Also in this commit: the ability to restore from an LSTM into an Exception IPA-GNN or regular IPA-GNN. To do this, set --config.finetune=LSTM
Merged assert error generation
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
No description provided.