Commit graph

Select branches

Hide pull requests

main

#1

Mono Color

• fe57d6aab2
  
  rainfall_stats: initial implementation this might reveal why we are having problems. If most/all the rainfall radar data is v small numbers, normalising might help. Starbeamrainbowlabs 2022-11-24 18:58:16 +0000
• 3131b4f7b3
  
  debug2 Starbeamrainbowlabs 2022-11-24 18:25:32 +0000
• d55a13f536
  
  debug Starbeamrainbowlabs 2022-11-24 18:24:03 +0000
• 1f60f2a580
  
  do_argmax Starbeamrainbowlabs 2022-11-24 18:11:03 +0000
• 6c09d5254d
  
  fixup Starbeamrainbowlabs 2022-11-24 17:57:48 +0000
• 54a841efe9
  
  train_mono_predict: convert to correct format Starbeamrainbowlabs 2022-11-24 17:56:07 +0000
• 105dc5bc56
  
  missing kwargs Starbeamrainbowlabs 2022-11-24 17:51:29 +0000
• 1e1d6dd273
  
  fixup Starbeamrainbowlabs 2022-11-24 17:48:19 +0000
• 011e0aef78
  
  update cli docs Starbeamrainbowlabs 2022-11-24 16:38:07 +0000
• 773944f9fa
  
  train_mono_predict: initial implementation Starbeamrainbowlabs 2022-11-24 16:33:50 +0000
• d31326cb30
  
  slurm train mono: fix partition name Starbeamrainbowlabs 2022-11-22 17:02:02 +0000
• ce28ac4013
  
  slurm: add job for train_mono Starbeamrainbowlabs 2022-11-22 16:58:46 +0000
• 3a0356929c
  
  mono: drop the sparse Starbeamrainbowlabs 2022-11-22 16:20:56 +0000
• 7e8f63f8ba
  
  fixup Starbeamrainbowlabs 2022-11-21 19:38:24 +0000
• ace4c8b246
  
  dataset_mono: debug Starbeamrainbowlabs 2022-11-21 18:46:21 +0000
• 527b34942d
  
  convnext_inverse: kernel_size 4→2 Starbeamrainbowlabs 2022-11-11 19:29:37 +0000
• 0662d0854b
  
  model_mono: fix bottleneck Starbeamrainbowlabs 2022-11-11 19:11:40 +0000
• 73acda6d9a
  
  fix debug logging Starbeamrainbowlabs 2022-11-11 19:08:38 +0000
• 9da059d738
  
  model shape logging Starbeamrainbowlabs 2022-11-11 19:03:37 +0000
• 00917b2698
  
  dataset_mono: log shapes Starbeamrainbowlabs 2022-11-11 19:02:43 +0000
• 54ae88b1b4
  
  in this entire blasted project I have yet to get the rotation of anything correct....! Starbeamrainbowlabs 2022-11-11 18:58:45 +0000
• 4b2e418ddc
  
  finish comment ...oops Starbeamrainbowlabs 2022-11-11 18:55:35 +0000
• a7a475dcd1
  
  debug 2 Starbeamrainbowlabs 2022-11-11 18:38:07 +0000
• bf2f6e9b64
  
  debug logging it begins again Starbeamrainbowlabs 2022-11-11 18:31:40 +0000
• 481eeb3759
  
  mono: fix dataset preprocessing rogue dimension Starbeamrainbowlabs 2022-11-11 18:31:27 +0000
• 9035450213
  
  mono: instantiate right model Starbeamrainbowlabs 2022-11-11 18:28:29 +0000
• 69a2d0cf04
  
  fixup Starbeamrainbowlabs 2022-11-11 18:27:01 +0000
• 65e801cf28
  
  train_mono: fix crash Starbeamrainbowlabs 2022-11-11 18:26:25 +0000
• 8ac5159adc
  
  dataset_mono: simplify param passing, onehot+threshold water depth data Starbeamrainbowlabs 2022-11-11 18:23:50 +0000
• 3a3f7e85da
  
  typo Starbeamrainbowlabs 2022-11-11 18:03:09 +0000
• 3313f77c88
  
  Add (untested) mono rainfall → water depth model * sighs * Unfortunately I can't seem to get contrastive learning to work..... Starbeamrainbowlabs 2022-11-10 22:36:11 +0000
• ce194d9227
  
  slurm: customise log file names Starbeamrainbowlabs 2022-11-10 21:09:34 +0000
• 9384b89165
  
  model_segmentation: spare → normal crossentropy, activation functions at end Starbeamrainbowlabs 2022-11-10 20:53:37 +0000
• b6676e7361
  
  switch from sparse to normal crossentropy Starbeamrainbowlabs 2022-11-10 20:50:56 +0000
• d8be26d476
  
  Merge branch 'main' of git.starbeamrainbowlabs.com:sbrl/PhD-Rainfall-Radar Starbeamrainbowlabs 2022-11-10 20:49:01 +0000
• b03388de60
  
  dataset_segmenter: DEBUG: fix water shape Starbeamrainbowlabs 2022-11-10 20:48:21 +0000
• daf691bf43
  
  typo Starbeamrainbowlabs 2022-11-10 19:55:00 +0000
• 0aa2ce19f5
  
  read_metadata: support file inputs as well as dirs Starbeamrainbowlabs 2022-11-10 19:53:30 +0000
• aa7d9b8cf6
  
  fixup Starbeamrainbowlabs 2022-11-10 19:46:09 +0000
• 0894bd09e8
  
  train_predict: add error message for parrams.json not found Starbeamrainbowlabs 2022-11-10 19:45:41 +0000
• e519b0adb3
  
  GzipChildProcess: spawn-stream is buggy IIRC Starbeamrainbowlabs 2022-11-09 16:43:05 +0000
• ddbf2cb734
  
  slurm-process: -n28 → --exclusive Starbeamrainbowlabs 2022-11-04 17:42:50 +0000
• a762664063
  
  slurm-process: -n28 fo uniq call Starbeamrainbowlabs 2022-11-04 17:23:20 +0000
• 0166b4d09e
  
  slurm-process: change log file names Starbeamrainbowlabs 2022-11-04 17:11:10 +0000
• 0353072d15
  
  allow pretrain to run on gpu we've slashed the size of the 2nd encoder, so ti should fit naow? Starbeamrainbowlabs 2022-11-04 17:02:07 +0000
• 44ad51f483
  
  CallbackNBatchCsv: bugfix .sort() → sorted() Starbeamrainbowlabs 2022-11-04 16:40:21 +0000
• 4dddcfcb42
  
  pretrain_predict: missing \n Starbeamrainbowlabs 2022-11-04 16:01:28 +0000
• 1375201c5f
  
  CallbackNBatchCsv: open_handle mode Starbeamrainbowlabs 2022-11-03 18:29:00 +0000
• 3206d6b7e7
  
  slurm: rename segmenter job name Starbeamrainbowlabs 2022-11-03 17:12:27 +0000
• f2ae74ce7b
  
  how could I be so stupid..... round 2 Starbeamrainbowlabs 2022-11-02 17:38:26 +0000
• 441ad92b12
  
  slurm: fixup Starbeamrainbowlabs 2022-11-01 19:57:15 +0000
• bc0e5f05a8
  
  slurm: fixup Starbeamrainbowlabs 2022-11-01 19:55:04 +0000
• 784b8ed35c
  
  recordify: catch NaN --count-file Starbeamrainbowlabs 2022-11-01 19:53:21 +0000
• c17a4ca05a
  
  slurm: fix sanity logic Starbeamrainbowlabs 2022-11-01 19:38:04 +0000
• 79b231198f
  
  slurm-process: check input files are readable Starbeamrainbowlabs 2022-11-01 19:03:37 +0000
• a69fa9f0f3
  
  slurm: rename Starbeamrainbowlabs 2022-11-01 18:59:55 +0000
• f8341e7d89
  
  slurm: add .log Starbeamrainbowlabs 2022-11-01 18:59:15 +0000
• fecc63b6a2
  
  wrangler: write high-level job file Starbeamrainbowlabs 2022-11-01 18:56:27 +0000
• 91152ebb1c
  
  wrangler:recordify update cli help we only output .jsonl.gz to a DIRECTORY, so update cli help to reflect this Starbeamrainbowlabs 2022-11-01 18:29:47 +0000
• 5f8d6dc6ea
  
  Add metrics every 64 batches this is important, because with large batches it can be difficult to tell what's happening inside each epoch. Starbeamrainbowlabs 2022-10-31 19:26:10 +0000
• cf872ef739
  
  how could I be so *stupid*...... Starbeamrainbowlabs 2022-10-31 18:40:58 +0000
• da32d75778
  
  make_callbacks: display steps, not samples Starbeamrainbowlabs 2022-10-31 18:36:28 +0000
• dfef7db421
  
  moar debugging Starbeamrainbowlabs 2022-10-31 18:26:34 +0000
• 172cf9d8ce
  
  tweak Starbeamrainbowlabs 2022-10-31 18:19:43 +0000
• dbe35ee943
  
  loss: comment l2 norm Starbeamrainbowlabs 2022-10-31 18:09:03 +0000
• 5e60319024
  
  fixup Starbeamrainbowlabs 2022-10-31 17:56:49 +0000
• b986b069e2
  
  debug party time Starbeamrainbowlabs 2022-10-31 17:50:29 +0000
• 458faa96d2
  
  loss: fixup Starbeamrainbowlabs 2022-10-31 17:18:21 +0000
• 55dc05e8ce
  
  contrastive: comment weights that aren't needed Starbeamrainbowlabs 2022-10-31 16:26:48 +0000
• 33391eaf16
  
  train_predict/jsonl: don't argmax I'm interested inthe raw values Starbeamrainbowlabs 2022-10-26 17:21:19 +0100
• 74f2cdb900
  
  train_predict: .list() → .tolist() Starbeamrainbowlabs 2022-10-26 17:12:36 +0100
• 4f9d543695
  
  train_predict: don't pass model_code it's redundant Starbeamrainbowlabs 2022-10-26 17:11:36 +0100
• 1b489518d0
  
  segmenter: add LayerStack2Image to custom_objects Starbeamrainbowlabs 2022-10-26 17:05:50 +0100
• 48ae8a5c20
  
  LossContrastive: normalise features as per the paper Starbeamrainbowlabs 2022-10-26 16:52:56 +0100
• 843cc8dc7b
  
  contrastive: rewrite the loss function. The CLIP paper *does* kinda make sense I think Starbeamrainbowlabs 2022-10-26 16:45:45 +0100
• fad1399c2d
  
  convnext: whitespace Starbeamrainbowlabs 2022-10-26 16:45:20 +0100
• 1d872cb962
  
  contrastive: fix initial temperature value It should be 1/0.07, but we had it set to 0.07...... Starbeamrainbowlabs 2022-10-26 16:45:01 +0100
• f994d449f1
  
  Layer2Image: fix Starbeamrainbowlabs 2022-10-25 21:32:17 +0100
• 6a29105f56
  
  model_segmentation: stack not reshape Starbeamrainbowlabs 2022-10-25 21:25:15 +0100
• 98417a3e06
  
  prepare for NCE loss .....but Tensorflow's implementation looks to be for supervised models :-( Starbeamrainbowlabs 2022-10-25 21:15:05 +0100
• bb0679a509
  
  model_segmentation: don't softmax twice Starbeamrainbowlabs 2022-10-25 21:11:48 +0100
• f2e2ca1484
  
  model_contrastive: make water encoder significantly shallower Starbeamrainbowlabs 2022-10-24 20:52:31 +0100
• a6b07a49cb
  
  count water/nowater pixels in Jupyter Notebook Starbeamrainbowlabs 2022-10-24 18:05:34 +0100
• a8b101bdae
  
  dataset_predict: add shape_water_desired Starbeamrainbowlabs 2022-10-24 18:05:13 +0100
• 587c1dfafa
  
  train_predict: revamp jsonl handling Starbeamrainbowlabs 2022-10-21 16:53:08 +0100
• 8195318a42
  
  SparseCategoricalAccuracy: losses → metrics Starbeamrainbowlabs 2022-10-21 16:51:20 +0100
• 612735aaae
  
  rename shuffle arg Starbeamrainbowlabs 2022-10-21 16:35:45 +0100
• c98d8d05dd
  
  segmentation: use the right accuracy Starbeamrainbowlabs 2022-10-21 16:17:05 +0100
• bb0258f5cd
  
  flip squeeze operator ordering Starbeamrainbowlabs 2022-10-21 15:38:57 +0100
• af26964c6a
  
  batched_iterator: reset i_item after every time Starbeamrainbowlabs 2022-10-21 15:35:43 +0100
• c5b1501dba
  
  train-predict fixup Starbeamrainbowlabs 2022-10-21 15:27:39 +0100
• 42aea7a0cc
  
  plt.close() fixup Starbeamrainbowlabs 2022-10-21 15:23:54 +0100
• 12dad3bc87
  
  vis/segmentation: fix titles Starbeamrainbowlabs 2022-10-21 15:22:35 +0100
• 0cb2de5d06
  
  train-preedict: close matplotlib after we've finished they act like file handles Starbeamrainbowlabs 2022-10-21 15:19:31 +0100
• 81e53efd9c
  
  PNG: create output dir if doesn't exist Starbeamrainbowlabs 2022-10-21 15:17:39 +0100
• 3f7db6fa78
  
  fix embedding confusion Starbeamrainbowlabs 2022-10-21 15:15:59 +0100
• 847cd97ec4
  
  fixup Starbeamrainbowlabs 2022-10-21 14:26:58 +0100
• 0e814b7e98
  
  Contraster → Segmenter Starbeamrainbowlabs 2022-10-21 14:25:43 +0100
• 1b658a1b7c
  
  train-predict: can't destructure array when iterating generator ....it seems to lead to undefined behaviour or something Starbeamrainbowlabs 2022-10-20 19:34:04 +0100
• aed2348a95
  
  train_predict: fixup Starbeamrainbowlabs 2022-10-20 15:42:33 +0100