rasa icon indicating copy to clipboard operation
rasa copied to clipboard

Swap non-deterministic ops with possibly deterministic ones

Open dakshvar22 opened this issue 3 years ago • 20 comments

Proposed changes:

  • ...

Status (please check what you already did):

  • [ ] added some tests for the functionality
  • [ ] updated the documentation
  • [ ] updated the changelog (please check changelog for instructions)
  • [ ] reformat files using black (please check Readme for instructions)

dakshvar22 avatar Mar 03 '22 18:03 dakshvar22

Status of the run: Failed

Commit: 911ad2ae00c0bbff9539620136d1886063cc70c3, The full report is available as an artifact.

Datadog dashboard link

github-actions[bot] avatar Mar 03 '22 18:03 github-actions[bot]

Status of the run: Succeeded

Commit: 911ad2ae00c0bbff9539620136d1886063cc70c3, The full report is available as an artifact.

Datadog dashboard link

Dataset: Customer 1, Dataset repository branch: mr-tests (external repository), commit: c36f81a7b7a012e8b42c8286d3fee8c0a3e3b896 Configuration repository branch: nib-test

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 1m42s, train: 6m36s, total: 8m17s
0.7816 (no data) 0.9967 (no data) no data

github-actions[bot] avatar Mar 03 '22 18:03 github-actions[bot]

Status of the run: Failed

Commit: 911ad2ae00c0bbff9539620136d1886063cc70c3, The full report is available as an artifact.

Datadog dashboard link

Dataset: Carbon Bot, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + BERT + DIET(bow) + ResponseSelector(bow)
test: 2m7s, train: 5m45s, total: 7m51s
0.7922 (-0.00) 0.7529 (0.00) 0.5430 (-0.01)
Sparse + BERT + DIET(seq) + ResponseSelector(t2t)
test: 3m9s, train: 6m26s, total: 9m35s
0.7806 (0.00) 0.7880 (0.00) 0.5563 (0.00)
Sparse + DIET(bow) + ResponseSelector(bow)
test: 50s, train: 3m37s, total: 4m27s
0.7456 (-0.00) 0.7529 (0.00) 0.5232 (0.01)
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 1m50s, train: 5m15s, total: 7m5s
0.7398 (0.00) 0.7022 (0.00) 0.5364 (0.01)

Dataset: Hermit, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + BERT + DIET(bow) + ResponseSelector(bow)
test: 3m46s, train: 31m0s, total: 34m45s
0.8717 (-0.00) 0.7504 (0.00) no data
Sparse + DIET(bow) + ResponseSelector(bow)
test: 1m15s, train: 24m10s, total: 25m24s
0.8271 (0.00) 0.7504 (0.00) no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 1m53s, train: 18m42s, total: 20m35s
0.8346 (0.00) 0.7585 (0.00) no data

Dataset: Private 1, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 37s, train: 4m35s, total: 5m12s
0.9002 (0.00) 0.9612 (0.00) no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 1m17s, train: 4m15s, total: 5m31s
0.9096 (0.00) 0.9735 (0.01) no data

Dataset: Private 2, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 43s, train: 6m48s, total: 7m31s
0.8498 (0.01) no data no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 49s, train: 5m42s, total: 6m30s
0.8530 (0.00) no data no data

Dataset: Private 3, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 36s, train: 1m17s, total: 1m52s
0.8683 (0.00) no data no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 41s, train: 53s, total: 1m34s
0.8642 (0.00) no data no data

github-actions[bot] avatar Mar 03 '22 20:03 github-actions[bot]

Status of the run: Failed

Commit: 911ad2ae00c0bbff9539620136d1886063cc70c3, The full report is available as an artifact.

Datadog dashboard link

Dataset: Carbon Bot, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + BERT + DIET(bow) + ResponseSelector(bow)
test: 2m8s, train: 5m49s, total: 7m56s
0.7883 (-0.01) 0.7529 (0.00) 0.5563 (0.00)
Sparse + BERT + DIET(seq) + ResponseSelector(t2t)
test: 3m13s, train: 6m29s, total: 9m41s
0.7806 (0.00) 0.7880 (0.00) 0.5563 (0.00)
Sparse + DIET(bow) + ResponseSelector(bow)
test: 53s, train: 4m12s, total: 5m5s
0.7573 (0.01) 0.7529 (0.00) 0.5249 (0.01)

github-actions[bot] avatar Mar 04 '22 03:03 github-actions[bot]

Status of the run: Failed

Commit: 911ad2ae00c0bbff9539620136d1886063cc70c3, The full report is available as an artifact.

Datadog dashboard link

Dataset: Carbon Bot, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + BERT + DIET(bow) + ResponseSelector(bow)
test: 2m10s, train: 5m50s, total: 8m0s
0.7922 (-0.00) 0.7529 (0.00) 0.5629 (0.01)
Sparse + BERT + DIET(seq) + ResponseSelector(t2t)
test: 3m6s, train: 6m13s, total: 9m19s
0.7806 (0.00) 0.7880 (0.00) 0.5364 (-0.02)
Sparse + DIET(bow) + ResponseSelector(bow)
test: 51s, train: 3m37s, total: 4m28s
0.7476 (0.00) 0.7529 (0.00) 0.5364 (0.02)
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 1m54s, train: 5m6s, total: 6m59s
0.7398 (0.00) 0.7022 (0.00) 0.5364 (0.01)

Dataset: Hermit, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + BERT + DIET(bow) + ResponseSelector(bow)
test: 3m48s, train: 31m4s, total: 34m51s
0.8690 (-0.00) 0.7504 (0.00) no data
Sparse + DIET(bow) + ResponseSelector(bow)
test: 1m15s, train: 24m7s, total: 25m22s
0.8262 (0.00) 0.7504 (0.00) no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 2m1s, train: 18m38s, total: 20m39s
0.8364 (0.00) 0.7571 (0.00) no data

Dataset: Private 1, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 38s, train: 4m48s, total: 5m26s
0.8940 (-0.00) 0.9612 (0.00) no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 1m16s, train: 4m17s, total: 5m33s
0.9075 (0.00) 0.9717 (0.01) no data

Dataset: Private 2, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 44s, train: 6m51s, total: 7m34s
0.8455 (0.01) no data no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 49s, train: 5m43s, total: 6m32s
0.8530 (0.00) no data no data

Dataset: Private 3, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 38s, train: 1m20s, total: 1m57s
0.8683 (0.00) no data no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 42s, train: 55s, total: 1m37s
0.8642 (0.00) no data no data

github-actions[bot] avatar Mar 04 '22 04:03 github-actions[bot]

Status of the run: Succeeded

Commit: 911ad2ae00c0bbff9539620136d1886063cc70c3, The full report is available as an artifact.

Datadog dashboard link

Dataset: Sara, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + BERT + DIET(bow) + ResponseSelector(bow)
test: 6m24s, train: 12m7s, total: 18m31s
0.6967 (0.01) 0.7949 (0.00) 0.7907 (-0.01)
Sparse + BERT + DIET(seq) + ResponseSelector(t2t)
test: 8m16s, train: 9m45s, total: 18m1s
0.7039 (-0.00) 0.7848 (-0.01) 0.7891 (-0.01)
Sparse + DIET(bow) + ResponseSelector(bow)
test: 2m6s, train: 9m37s, total: 11m42s
0.6630 (-0.01) 0.7949 (0.00) 0.7736 (-0.01)
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 2m47s, train: 6m42s, total: 9m30s
0.6794 (-0.00) 0.7692 (-0.00) 0.7907 (-0.01)

github-actions[bot] avatar Mar 04 '22 07:03 github-actions[bot]

Status of the run: Failed

Commit: c2b0bdc9bba09e0779b549cd420631c8ec4acbaf, The full report is available as an artifact.

Datadog dashboard link

github-actions[bot] avatar Mar 04 '22 07:03 github-actions[bot]

Status of the run: Failed

Commit: c2b0bdc9bba09e0779b549cd420631c8ec4acbaf, The full report is available as an artifact.

Datadog dashboard link

github-actions[bot] avatar Mar 04 '22 07:03 github-actions[bot]

Status of the run: Succeeded

Commit: c2b0bdc9bba09e0779b549cd420631c8ec4acbaf, The full report is available as an artifact.

Datadog dashboard link

Dataset: Private 1, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 39s, train: 5m59s, total: 6m38s
0.7495 (-0.15) 0.9612 (0.00) no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 1m17s, train: 4m27s, total: 5m44s
0.8170 (-0.09) 0.9709 (0.00) no data

Dataset: Private 2, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 45s, train: 6m53s, total: 7m38s
0.2650 (-0.57) no data no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 54s, train: 5m54s, total: 6m48s
0.5923 (-0.26) no data no data

github-actions[bot] avatar Mar 04 '22 08:03 github-actions[bot]

Status of the run: Failed

Commit: 8bd612eadadfecd9d616f066bc32fc3a73485ae4, The full report is available as an artifact.

Datadog dashboard link

github-actions[bot] avatar Mar 04 '22 09:03 github-actions[bot]

Status of the run: Succeeded

Commit: 5a78a9fddd7966886546eca346cce126b7bc477c, The full report is available as an artifact.

Datadog dashboard link

Dataset: Private 1, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 42s, train: 5m5s, total: 5m47s
0.8919 (-0.01) 0.9612 (0.00) no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 1m25s, train: 4m16s, total: 5m40s
0.9012 (-0.01) 0.9663 (0.00) no data

Dataset: Private 2, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 47s, train: 7m1s, total: 7m47s
0.8423 (0.00) no data no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 52s, train: 5m49s, total: 6m41s
0.8552 (0.00) no data no data

github-actions[bot] avatar Mar 04 '22 10:03 github-actions[bot]

Status of the run: Failed

Commit: 63d18b65a0ae76ba98b161a418f3c9dee0e90497, The full report is available as an artifact.

Datadog dashboard link

github-actions[bot] avatar Mar 04 '22 11:03 github-actions[bot]

Status of the run: Failed

Commit: 63d18b65a0ae76ba98b161a418f3c9dee0e90497, The full report is available as an artifact.

Datadog dashboard link

Dataset: Private 1, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 42s, train: 4m51s, total: 5m33s
0.8992 (0.00) 0.9612 (0.00) no data

github-actions[bot] avatar Mar 04 '22 11:03 github-actions[bot]

Status of the run: Succeeded

Commit: e9e8b437a1238a78a4638b9819fb0aaf0f78d06d, The full report is available as an artifact.

Datadog dashboard link

Dataset: Private 1, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 42s, train: 4m50s, total: 5m32s
0.8950 (-0.00) 0.9612 (0.00) no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 1m25s, train: 4m32s, total: 5m57s
0.9023 (-0.00) 0.9655 (-0.00) no data

Dataset: Private 2, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + DIET(bow) + ResponseSelector(bow)
test: 45s, train: 6m59s, total: 7m44s
0.8573 (0.02) no data no data
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 55s, train: 5m53s, total: 6m47s
0.8552 (0.00) no data no data

github-actions[bot] avatar Mar 04 '22 12:03 github-actions[bot]

Hey @dakshvar22! :wave: To run model regression tests, comment with the /modeltest command and a configuration.

Tips :bulb:: The model regression test will be run on push events. You can re-run the tests by re-add status:model-regression-tests label or use a Re-run jobs button in Github Actions workflow.

Tips :bulb:: Every time when you want to change a configuration you should edit the comment with the previous configuration.

You can copy this in your comment and customize:

/modeltest

```yml
##########
## Available datasets
##########
# - "Carbon Bot" (NLU)
# - "Customer 1" (NLU, Core)
# - "Hermit" (NLU)
# - "Private 1" (NLU)
# - "Private 2" (NLU)
# - "Private 3" (NLU)
# - "Sara" (NLU, Core)
# - "financial-demo" (NLU, Core)
# - "helpdesk-assistant" (NLU, Core)
# - "insurance-demo" (NLU, Core)
# - "retail-demo" (NLU, Core)

##########
## Available NLU configurations
##########
# - "BERT + DIET(bow) + ResponseSelector(bow)"
# - "BERT + DIET(seq) + ResponseSelector(t2t)"
# - "Spacy + DIET(bow) + ResponseSelector(bow)"
# - "Spacy + DIET(seq) + ResponseSelector(t2t)"
# - "Sparse + BERT + DIET(bow) + ResponseSelector(bow)"
# - "Sparse + BERT + DIET(seq) + ResponseSelector(t2t)"
# - "Sparse + DIET(bow) + ResponseSelector(bow)"
# - "Sparse + DIET(seq) + ResponseSelector(t2t)"
# - "Sparse + Spacy + DIET(bow) + ResponseSelector(bow)"
# - "Sparse + Spacy + DIET(seq) + ResponseSelector(t2t)"

##########
## Available Core configurations
##########
# - "Rules"
# - "Rules + AugMemo"
# - "Rules + AugMemo + TED"
# - "Rules + Memo"
# - "Rules + Memo + TED"
# - "Rules + TED"

## Example configuration
#################### syntax #################
## include:
##   - dataset: ["<dataset_name>"]
##     config: ["<configuration_name>"]
#
## Example:
## include:
##  - dataset: ["Carbon Bot"]
##    config: ["Sparse + DIET(bow) + ResponseSelector(bow)"]
#
## Shortcut:
## You can use the "all" shortcut to include all available configurations or datasets
#
## Example: Use the "Sparse + EmbeddingIntent + ResponseSelector(bow)" configuration
## for all available datasets
## include:
##  - dataset: ["all"]
##    config: ["Sparse + DIET(bow) + ResponseSelector(bow)"]
#
## Example: Use all available configurations for the "Carbon Bot" and "Sara" datasets
## and for the "Hermit" dataset use the "Sparse + DIET + ResponseSelector(T2T)" and
## "BERT + DIET + ResponseSelector(T2T)" configurations:
## include:
##  - dataset: ["Carbon Bot", "Sara"]
##    config: ["all"]
##  - dataset: ["Hermit"]
##    config: ["Sparse + DIET(seq) + ResponseSelector(t2t)", "BERT + DIET(seq) + ResponseSelector(t2t)"]
#
## Example: Define a branch name to check-out for a dataset repository. Default branch is 'main'
## dataset_branch: "test-branch"
## include:
##  - dataset: ["Carbon Bot", "Sara"]
##    config: ["all"]
#
## Example: Define number of repetitions. This will inform how often to repeat all runs defined in the include section. Default is 1
## num_repetitions: 2
## include:
##  - dataset: ["Carbon Bot", "Sara"]
##    config: ["Sparse + DIET(seq) + ResponseSelector(t2t)"]
##
## Shortcuts:
## You can use the "all" shortcut to include all available configurations or datasets.
## You can use the "all-nlu" shortcut to include all available NLU configurations or datasets.
## You can use the "all-core" shortcut to include all available core configurations or datasets.

include:
 - dataset: ["Carbon Bot"]
   config: ["Sparse + DIET(bow) + ResponseSelector(bow)"]

```

github-actions[bot] avatar Mar 04 '22 13:03 github-actions[bot]

/modeltest

dataset_branch: "nib-test"
include:
 - dataset: ["Sara", "Sara", "Sara", "Sara"]
   config: ["Sparse + DIET(seq) + ResponseSelector(t2t)", "Sparse + DIET(bow) + ResponseSelector(bow)", "Sparse + BERT + DIET(bow) + ResponseSelector(bow)", "Sparse + BERT + DIET(seq) + ResponseSelector(t2t)"]

github-actions[bot] avatar Mar 04 '22 13:03 github-actions[bot]

The model regression tests have started. It might take a while, please be patient. As soon as results are ready you'll see a new comment with the results.

Used configuration can be found in the comment.

github-actions[bot] avatar Mar 04 '22 13:03 github-actions[bot]

Status of the run: Succeeded

Commit: e9e8b437a1238a78a4638b9819fb0aaf0f78d06d, The full report is available as an artifact.

Datadog dashboard link

Dataset: Sara, Dataset repository branch: nib-test, commit: 1434bff267f2be514caea55c94638ecd8c4bd864

Configuration Intent Classification Micro F1 Entity Recognition Micro F1 Response Selection Micro F1
Sparse + BERT + DIET(bow) + ResponseSelector(bow)
test: 6m22s, train: 11m57s, total: 18m18s
0.7010 (0.01) 0.7949 (0.00) 0.7907 (-0.01)
Sparse + BERT + DIET(seq) + ResponseSelector(t2t)
test: 7m27s, train: 9m6s, total: 16m32s
0.6948 (-0.01) 0.7913 (-0.00) 0.8031 (0.01)
Sparse + DIET(bow) + ResponseSelector(bow)
test: 1m57s, train: 8m37s, total: 10m34s
0.6688 (0.00) 0.7949 (0.00) 0.7804 (-0.01)
Sparse + DIET(seq) + ResponseSelector(t2t)
test: 2m42s, train: 6m30s, total: 9m11s
0.6774 (-0.00) 0.7629 (-0.01) 0.7876 (-0.01)

github-actions[bot] avatar Mar 04 '22 15:03 github-actions[bot]

Status: Even with the above changes, the model training is not deterministic. Will need to try a different approach to first determine what's the first point of non-determinism in the model architecture. Clearly its not sparse.sparse_dense_matmul.

dakshvar22 avatar Mar 10 '22 10:03 dakshvar22

CLA assistant check
Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.


Daksh seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
You have signed the CLA already but the status is still pending? Let us recheck it.

CLAassistant avatar Jul 18 '22 22:07 CLAassistant

Closing this as it does not look like this line of work yielded improvements.

twerkmeister avatar Sep 14 '22 07:09 twerkmeister