# UDTube (beta)
[](https://dl.circleci.com/status-badge/redirect/gh/CUNY-CL/udtube/tree/master)
UDTube is a neural morphological analyzer based on
[PyTorch](https://pytorch.org/), [Lightning](https://lightning.ai/), and
[Hugging Face transformers](https://huggingface.co/docs/transformers/en/index).
## Philosophy
Named in homage to the venerable
[UDPipe](https://lindat.mff.cuni.cz/services/udpipe/), UDTube is focused on
incremental inference, allowing it to be used to label large text collections.
## Design
The UDTube model consists of a pre-trained (and possibly, fine-tuned)
transformer encoder which feeds into a classifier layer with many as four heads
handling the different morphological tasks.
Lightning is used to generate the [training, validation, inference, and
evaluation
loops](https://lightning.ai/docs/pytorch/latest/common/lightning_module.html#hooks).
The [LightningCLI
interface](https://lightning.ai/docs/pytorch/stable/cli/lightning_cli.html#lightning-cli)
is used to provide a user interface and manage configuration.
Below, we use [YAML](https://yaml.org/) to specify configuration options, and we
strongly recommend users do the same. However, most configuration options can
also be specified using POSIX-style command-line flags.
## Installation
To install UDTube and its dependencies, run the following command:
pip install .
## File formats
Other than YAML configuration files, most operations use files in
[CoNLL-U](https://universaldependencies.org/format.html) format. This is a
10-column tab-separated format with a blank line between each sentence and `#`
used for comments. In all cases, the `ID` and `FORM` field must be fully
populated; the `_` blank tag can be used for unknown fields.
Many of our experiments are performed using CoNLL-U data from the [Universal
Dependencies project](https://universaldependencies.org/).
## Tasks
UDTube can perform up to four morphological tasks simultaneously:
- Lemmatization is performed using the `LEMMA` field and [edit
scripts](https://aclanthology.org/P14-2111/).
- [Universal part-of-speech
tagging](https://universaldependencies.org/u/pos/index.html) is performed
using the `UPOS` field: enable with `data: use_upos: true`.
- Language-specific part-of-speech tagging is performed using the `XPOS`
field: enable with `data: use_xpos: true`.
- Morphological feature tagging is performed using the `FEATS` field: enable
with `data: use_feats: true`.
The following caveats apply:
- Note that many newer Universal Dependencies datasets do not have
language-specific part-of-speech-tags.
- The `FEATS` field is treated as a single unit and is not segmented in any
way.
- One can convert from [Universal Dependencies morphological
features](https://universaldependencies.org/u/feat/index.html) to [UniMorph
features](https://unimorph.github.io/schema/) using
[`scripts/convert_to_um.py`](scripts/convert_to_um.py).
- UDTube does not perform dependency parsing at present, so the `HEAD`,
`DEPREL`, and `DEPS` fields are ignored and should be specified as `_`.
## Usage
The `udtube` command-line tool uses a subcommand interface, with the four
following modes. To see the full set of options available with each subcommand,
use the `--print_config` flag. For example:
udtube fit --print_config
will show all configuration options (and their default values) for the `fit`
subcommand.
### Training (`fit`)
In `fit` mode, one trains a UDTube model from scratch. Naturally, most
configuration options need to be set at training time. E.g., it is not possible
to switch between different pre-trained encoders or enable new tasks after
training.
This mode is invoked using the `fit` subcommand, like so:
udtube fit --config path/to/config.yaml
#### Seeding
Setting the `seed_everything:` argument to some value ensures a reproducible
experiment.
#### Encoder
The encoder layer consists of a pre-trained BERT-style transformer model. By
default, UDTube uses multilingual cased BERT
(`model: encoder: google-bert/bert-base-multilingual-cased`). In theory, UDTube
can use any Hugging Face pre-trained encoder so long as it provides a
`AutoTokenizer` and has been exposed to the target language. We [list all the
Hugging Face encoders we have tested thus far](udtube/encoders.py), and warn
users when selecting an untested encoder. Since there is no standard for
referring to the between-layer dropout probability parameter, it is in some
cases also necessary to specify what this argument is called for a given model.
We welcome pull requests from users who successfully make use of encoders not
listed here.
So-called "tokenizer-free" pre-trained encoders like ByT5 are not currently
supported as they lack an `AutoTokenizer`.
#### Classifier
The classifier layer contains up to four sequential linear heads for the four
tasks described above. By default all four are enabled.
#### Optimization
UDTube uses separate optimizers and LR schedulers for the encoder and
classifier. The intuition behind this is that we may wish to make slow, small
changes (or possibly, no changes at all) to the pre-trained encoder, whereas we
wish to make more rapid and larger changes to the classifier.
The following YAML snippet shows a simple configuration that encapsulates this
principle. It uses the Adam optimizer for both encoder and classifier, but uses
a lower learning rate for the encoder with a linear warm-up and a higher
learning rate for the classifier.
...
model:
encoder_optimizer:
class_path: torch.optim.Adam
init_args:
lr: 1e-5
encoder_scheduler:
class_path: udtube.schedulers.WarmupInverseSquareRoot
init_args:
warmup_epochs: 5
classifier_optimizer:
class_path: torch.optim.Adam
init_args:
lr: 1e-3
classifier_scheduler:
class_path: lightning.pytorch.cli.ReduceLROnPlateau
init_args:
monitor: val_loss
factor: 0.1
...
The default scheduler is `udtube.schedulers.DummyScheduler`, which keeps
learning rate fixed to its initial value.
#### Checkpointing
The
[`ModelCheckpoint`](https://lightning.ai/docs/pytorch/stable/api/lightning.pytorch.callbacks.ModelCheckpoint.html)
is used to control the generation of checkpoint files. A sample YAML snippet is
given below.
...
checkpoint:
filename: "model-{epoch:03d}-{val_loss:.4f}"
monitor: val_loss
verbose: true
...
Without some specification under `checkpoint:` UDTube will not generate
checkpoints!
#### Callbacks
The user will likely want to configure additional callbacks. Some useful
examples are given below.
The
[`LearningRateMonitor`](https://lightning.ai/docs/pytorch/stable/api/lightning.pytorch.callbacks.LearningRateMonitor.html)
callback records learning rates; this is useful when working with multiple
optimizers and/or schedulers, as we do here. A sample YAML snippet is given
below.
...
trainer:
callbacks:
- class_path: lightning.pytorch.callbacks.LearningRateMonitor
init_args:
logging_interval: epoch
...
The
[`EarlyStopping`](https://lightning.ai/docs/pytorch/stable/common/early_stopping.html)
callback enables early stopping based on a monitored quantity and a fixed
"patience". A sample YAML snipppet with a patience of 10 is given below.
...
trainer:
callbacks:
- class_path: lightning.pytorch.callbacks.EarlyStopping
init_args:
monitor: val_loss
patience: 10
verbose: true
...
Adjust the `patience` parameter as needed.
All three of these features are enabled in the [sample configuration
files](configs) we provide.
#### Logging
By default, UDTube performs some minimal logging to standard error and uses
progress bars to keep track of progress during each epoch. However, one can
enable additional logging faculties during training, using a similar syntax to
the one we saw above for callbacks.
The
[`CSVLogger`](https://lightning.ai/docs/pytorch/stable/extensions/generated/lightning.pytorch.loggers.CSVLogger.html)
logs all monitored quantities to a CSV file. A sample configuration is given
below.
...
trainer:
logger:
- class_path: lightning.pytorch.loggers.CSVLogger
init_args:
save_dir: /Users/Shinji/models
...
Adjust the `save_dir` argument as needed.
The
[`WandbLogger`](https://lightning.ai/docs/pytorch/stable/extensions/generated/lightning.pytorch.loggers.WandbLogger.html)
works similarly to the `CSVLogger`, but sends the data to the third-party
website [Weights & Biases](https://wandb.ai/site), where it can be used to
generate charts or share artifacts. A sample configuration is given below.
...
trainer:
logger:
- class_path: lightning.pytorch.loggers.wandb.WandbLogger
init_args:
entity: NERV
project: unit1
save_dir: /Users/Shinji/models
...
Adjust the `entity`, `project`, and `save_dir` arguments as needed; note that
this functionality requires a working account with Weights & Biases.
#### Other options
By default, UDTube attempts to model all four tasks; one can disable the
language-specific tagging task using `model: use_xpos: false`, and so on.
Dropout probability is specified using `model: dropout: ...`.
The encoder has multiple layers. The input to the classifier consists of just
the last few layers mean-pooled together. The number of layers used for
mean-pooling is specified using `model: pooling_layers: ...`.
By default, lemmatization uses reverse-edit scripts. This is appropriate for
predominantly suffixal languages, which are thought to represent the majority of
the world's languages. If working with a predominantly prefixal language,
disable this with `model: reverse_edits: false`.
The following YAML snippet shows the default architectural arguments.
...
model:
dropout: 0.5
encoder: google-bert/bert-base-multilingual-cased
pooling_layers: 4
reverse_edits: true
use_upos: true
use_xpos: true
use_lemma: true
use_feats: true
...
Batch size is specified using `data: batch_size: ...` and defaults to 32.
There are a number of ways to specify how long a model should train for. For
example, the following YAML snippet specifies that training should run for 100
epochs or 6 wall-clock hours, whichever comes first.
...
trainer:
max_epochs: 100
max_time: 00:06:00:00
...
### Validation (`validate`)
In `validation` mode, one runs the validation step over labeled validation data
(specified as `data: val: path/to/validation.conllu`) using a previously trained
checkpoint (`--ckpt_path path/to/checkpoint.ckpt` from the command line),
recording total loss and per-task accuracies. In practice this is mostly usefulf
or debugging.
This mode is invoked using the `validate` subcommand, like so:
udtube validate --config path/to/config.yaml --ckpt_path path/to/checkpoint.ckpt
### Evaluation (`test`)
In `test` mode, we compute accuracy over held-out test data (specified as
`data: test: path/to/test.conllu`) using a previously trained checkpoint
(`--ckpt_path path/to/checkpoint.ckpt` from the command line); it differs from
`validation` mode in that it uses the `test` file rather than the `val` file and
it does not compute loss.
This mode is invoked using the `test` subcommand, like so:
udtube test --config path/to/config.yaml --ckpt_path path/to/checkpoint.ckpt
### Inference (`predict`)
In `predict` mode, a previously trained model checkpoint
(`--ckpt_path path/to/checkpoint.ckpt` from the command line) is used to label a
CoNLL-U file. One must also specify the path where the predictions will be
written.
...
predict:
path: /Users/Shinji/predictions.conllu
...
Here are some additional details:
- In `predict` mode UDTube loads the file to be labeled incrementally (i.e.,
one sentence at a time) so this can be used with very large files.
- In `predict` mode, if no path for the predictions is specified, stdout will
be used. If using this in conjunction with \> or \|, add
`--trainer.enable_progress_bar false` on the command line.
- The target task fields are overriden if their heads are active.
- Use [`scripts/pretokenize.py`](scripts/pretokenize.py) to convert raw text
files to CoNLL-U input files.
This mode is invoked using the `predict` subcommand, like so:
udtube predict --config path/to/config.yaml --ckpt_path path/to/checkpoint.ckpt
## Examples
See [`examples`](examples/README.md) for some worked examples including
hyperparameter sweeping with [Weights & Biases](https://wandb.ai/site).
## Additional scripts
See [`scripts/README.md`](scripts/README.md) for details on provided scripts not
mention above.
## License
UDTube is distributed under an [Apache 2.0 license](LICENSE.txt).
## Contribution
We welcome contributions using the fork-and-pull model.
## References
If you use UDTube in your research, we would appreciate it if you cited the
following document, which describes the model:
Yakubov, D. 2024. [How do we learn what we cannot
say?](https://academicworks.cuny.edu/gc_etds/5622/) Master's thesis, CUNY
Graduate Center.
Raw data
{
"_id": null,
"home_page": null,
"name": "udtube",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.9",
"maintainer_email": null,
"keywords": "computational linguistics, morphology, natural language processing, language",
"author": "Daniel Yakubov, Kyle Gorman",
"author_email": null,
"download_url": "https://files.pythonhosted.org/packages/38/be/b8be57725a348274e992ec1e3a88c3d71efbcd2d1b5d2b8c4f6837b3d668/udtube-0.1.0.tar.gz",
"platform": null,
"description": "# UDTube (beta)\n\n[](https://dl.circleci.com/status-badge/redirect/gh/CUNY-CL/udtube/tree/master)\n\nUDTube is a neural morphological analyzer based on\n[PyTorch](https://pytorch.org/), [Lightning](https://lightning.ai/), and\n[Hugging Face transformers](https://huggingface.co/docs/transformers/en/index).\n\n## Philosophy\n\nNamed in homage to the venerable\n[UDPipe](https://lindat.mff.cuni.cz/services/udpipe/), UDTube is focused on\nincremental inference, allowing it to be used to label large text collections.\n\n## Design\n\nThe UDTube model consists of a pre-trained (and possibly, fine-tuned)\ntransformer encoder which feeds into a classifier layer with many as four heads\nhandling the different morphological tasks.\n\nLightning is used to generate the [training, validation, inference, and\nevaluation\nloops](https://lightning.ai/docs/pytorch/latest/common/lightning_module.html#hooks).\nThe [LightningCLI\ninterface](https://lightning.ai/docs/pytorch/stable/cli/lightning_cli.html#lightning-cli)\nis used to provide a user interface and manage configuration.\n\nBelow, we use [YAML](https://yaml.org/) to specify configuration options, and we\nstrongly recommend users do the same. However, most configuration options can\nalso be specified using POSIX-style command-line flags.\n\n## Installation\n\nTo install UDTube and its dependencies, run the following command:\n\n pip install .\n\n## File formats\n\nOther than YAML configuration files, most operations use files in\n[CoNLL-U](https://universaldependencies.org/format.html) format. This is a\n10-column tab-separated format with a blank line between each sentence and `#`\nused for comments. In all cases, the `ID` and `FORM` field must be fully\npopulated; the `_` blank tag can be used for unknown fields.\n\nMany of our experiments are performed using CoNLL-U data from the [Universal\nDependencies project](https://universaldependencies.org/).\n\n## Tasks\n\nUDTube can perform up to four morphological tasks simultaneously:\n\n- Lemmatization is performed using the `LEMMA` field and [edit\n scripts](https://aclanthology.org/P14-2111/).\n\n- [Universal part-of-speech\n tagging](https://universaldependencies.org/u/pos/index.html) is performed\n using the `UPOS` field: enable with `data: use_upos: true`.\n\n- Language-specific part-of-speech tagging is performed using the `XPOS`\n field: enable with `data: use_xpos: true`.\n\n- Morphological feature tagging is performed using the `FEATS` field: enable\n with `data: use_feats: true`.\n\nThe following caveats apply:\n\n- Note that many newer Universal Dependencies datasets do not have\n language-specific part-of-speech-tags.\n- The `FEATS` field is treated as a single unit and is not segmented in any\n way.\n- One can convert from [Universal Dependencies morphological\n features](https://universaldependencies.org/u/feat/index.html) to [UniMorph\n features](https://unimorph.github.io/schema/) using\n [`scripts/convert_to_um.py`](scripts/convert_to_um.py).\n- UDTube does not perform dependency parsing at present, so the `HEAD`,\n `DEPREL`, and `DEPS` fields are ignored and should be specified as `_`.\n\n## Usage\n\nThe `udtube` command-line tool uses a subcommand interface, with the four\nfollowing modes. To see the full set of options available with each subcommand,\nuse the `--print_config` flag. For example:\n\n udtube fit --print_config\n\nwill show all configuration options (and their default values) for the `fit`\nsubcommand.\n\n### Training (`fit`)\n\nIn `fit` mode, one trains a UDTube model from scratch. Naturally, most\nconfiguration options need to be set at training time. E.g., it is not possible\nto switch between different pre-trained encoders or enable new tasks after\ntraining.\n\nThis mode is invoked using the `fit` subcommand, like so:\n\n udtube fit --config path/to/config.yaml\n\n#### Seeding\n\nSetting the `seed_everything:` argument to some value ensures a reproducible\nexperiment.\n\n#### Encoder\n\nThe encoder layer consists of a pre-trained BERT-style transformer model. By\ndefault, UDTube uses multilingual cased BERT\n(`model: encoder: google-bert/bert-base-multilingual-cased`). In theory, UDTube\ncan use any Hugging Face pre-trained encoder so long as it provides a\n`AutoTokenizer` and has been exposed to the target language. We [list all the\nHugging Face encoders we have tested thus far](udtube/encoders.py), and warn\nusers when selecting an untested encoder. Since there is no standard for\nreferring to the between-layer dropout probability parameter, it is in some\ncases also necessary to specify what this argument is called for a given model.\nWe welcome pull requests from users who successfully make use of encoders not\nlisted here.\n\nSo-called \"tokenizer-free\" pre-trained encoders like ByT5 are not currently\nsupported as they lack an `AutoTokenizer`.\n\n#### Classifier\n\nThe classifier layer contains up to four sequential linear heads for the four\ntasks described above. By default all four are enabled.\n\n#### Optimization\n\nUDTube uses separate optimizers and LR schedulers for the encoder and\nclassifier. The intuition behind this is that we may wish to make slow, small\nchanges (or possibly, no changes at all) to the pre-trained encoder, whereas we\nwish to make more rapid and larger changes to the classifier.\n\nThe following YAML snippet shows a simple configuration that encapsulates this\nprinciple. It uses the Adam optimizer for both encoder and classifier, but uses\na lower learning rate for the encoder with a linear warm-up and a higher\nlearning rate for the classifier.\n\n ...\n model:\n encoder_optimizer:\n class_path: torch.optim.Adam\n init_args:\n lr: 1e-5\n encoder_scheduler:\n class_path: udtube.schedulers.WarmupInverseSquareRoot\n init_args:\n warmup_epochs: 5\n classifier_optimizer:\n class_path: torch.optim.Adam\n init_args:\n lr: 1e-3\n classifier_scheduler:\n class_path: lightning.pytorch.cli.ReduceLROnPlateau\n init_args:\n monitor: val_loss\n factor: 0.1\n ...\n\nThe default scheduler is `udtube.schedulers.DummyScheduler`, which keeps\nlearning rate fixed to its initial value.\n\n#### Checkpointing\n\nThe\n[`ModelCheckpoint`](https://lightning.ai/docs/pytorch/stable/api/lightning.pytorch.callbacks.ModelCheckpoint.html)\nis used to control the generation of checkpoint files. A sample YAML snippet is\ngiven below.\n\n ...\n checkpoint:\n filename: \"model-{epoch:03d}-{val_loss:.4f}\"\n monitor: val_loss\n verbose: true\n ...\n\nWithout some specification under `checkpoint:` UDTube will not generate\ncheckpoints!\n\n#### Callbacks\n\nThe user will likely want to configure additional callbacks. Some useful\nexamples are given below.\n\nThe\n[`LearningRateMonitor`](https://lightning.ai/docs/pytorch/stable/api/lightning.pytorch.callbacks.LearningRateMonitor.html)\ncallback records learning rates; this is useful when working with multiple\noptimizers and/or schedulers, as we do here. A sample YAML snippet is given\nbelow.\n\n ...\n trainer:\n callbacks:\n - class_path: lightning.pytorch.callbacks.LearningRateMonitor\n init_args:\n logging_interval: epoch\n ...\n\nThe\n[`EarlyStopping`](https://lightning.ai/docs/pytorch/stable/common/early_stopping.html)\ncallback enables early stopping based on a monitored quantity and a fixed\n\"patience\". A sample YAML snipppet with a patience of 10 is given below.\n\n ...\n trainer:\n callbacks:\n - class_path: lightning.pytorch.callbacks.EarlyStopping\n init_args:\n monitor: val_loss\n patience: 10\n verbose: true\n ...\n\nAdjust the `patience` parameter as needed.\n\nAll three of these features are enabled in the [sample configuration\nfiles](configs) we provide.\n\n#### Logging\n\nBy default, UDTube performs some minimal logging to standard error and uses\nprogress bars to keep track of progress during each epoch. However, one can\nenable additional logging faculties during training, using a similar syntax to\nthe one we saw above for callbacks.\n\nThe\n[`CSVLogger`](https://lightning.ai/docs/pytorch/stable/extensions/generated/lightning.pytorch.loggers.CSVLogger.html)\nlogs all monitored quantities to a CSV file. A sample configuration is given\nbelow.\n\n ...\n trainer:\n logger:\n - class_path: lightning.pytorch.loggers.CSVLogger\n init_args:\n save_dir: /Users/Shinji/models\n ...\n \n\nAdjust the `save_dir` argument as needed.\n\nThe\n[`WandbLogger`](https://lightning.ai/docs/pytorch/stable/extensions/generated/lightning.pytorch.loggers.WandbLogger.html)\nworks similarly to the `CSVLogger`, but sends the data to the third-party\nwebsite [Weights & Biases](https://wandb.ai/site), where it can be used to\ngenerate charts or share artifacts. A sample configuration is given below.\n\n ...\n trainer:\n logger:\n - class_path: lightning.pytorch.loggers.wandb.WandbLogger\n init_args:\n entity: NERV\n project: unit1\n save_dir: /Users/Shinji/models\n ...\n\nAdjust the `entity`, `project`, and `save_dir` arguments as needed; note that\nthis functionality requires a working account with Weights & Biases.\n\n#### Other options\n\nBy default, UDTube attempts to model all four tasks; one can disable the\nlanguage-specific tagging task using `model: use_xpos: false`, and so on.\n\nDropout probability is specified using `model: dropout: ...`.\n\nThe encoder has multiple layers. The input to the classifier consists of just\nthe last few layers mean-pooled together. The number of layers used for\nmean-pooling is specified using `model: pooling_layers: ...`.\n\nBy default, lemmatization uses reverse-edit scripts. This is appropriate for\npredominantly suffixal languages, which are thought to represent the majority of\nthe world's languages. If working with a predominantly prefixal language,\ndisable this with `model: reverse_edits: false`.\n\nThe following YAML snippet shows the default architectural arguments.\n\n ...\n model:\n dropout: 0.5\n encoder: google-bert/bert-base-multilingual-cased\n pooling_layers: 4\n reverse_edits: true\n use_upos: true\n use_xpos: true\n use_lemma: true\n use_feats: true\n ...\n \n\nBatch size is specified using `data: batch_size: ...` and defaults to 32.\n\nThere are a number of ways to specify how long a model should train for. For\nexample, the following YAML snippet specifies that training should run for 100\nepochs or 6 wall-clock hours, whichever comes first.\n\n ...\n trainer:\n max_epochs: 100\n max_time: 00:06:00:00\n ...\n\n### Validation (`validate`)\n\nIn `validation` mode, one runs the validation step over labeled validation data\n(specified as `data: val: path/to/validation.conllu`) using a previously trained\ncheckpoint (`--ckpt_path path/to/checkpoint.ckpt` from the command line),\nrecording total loss and per-task accuracies. In practice this is mostly usefulf\nor debugging.\n\nThis mode is invoked using the `validate` subcommand, like so:\n\n udtube validate --config path/to/config.yaml --ckpt_path path/to/checkpoint.ckpt\n\n### Evaluation (`test`)\n\nIn `test` mode, we compute accuracy over held-out test data (specified as\n`data: test: path/to/test.conllu`) using a previously trained checkpoint\n(`--ckpt_path path/to/checkpoint.ckpt` from the command line); it differs from\n`validation` mode in that it uses the `test` file rather than the `val` file and\nit does not compute loss.\n\nThis mode is invoked using the `test` subcommand, like so:\n\n udtube test --config path/to/config.yaml --ckpt_path path/to/checkpoint.ckpt\n\n### Inference (`predict`)\n\nIn `predict` mode, a previously trained model checkpoint\n(`--ckpt_path path/to/checkpoint.ckpt` from the command line) is used to label a\nCoNLL-U file. One must also specify the path where the predictions will be\nwritten.\n\n ...\n predict:\n path: /Users/Shinji/predictions.conllu\n ...\n\nHere are some additional details:\n\n- In `predict` mode UDTube loads the file to be labeled incrementally (i.e.,\n one sentence at a time) so this can be used with very large files.\n- In `predict` mode, if no path for the predictions is specified, stdout will\n be used. If using this in conjunction with \\> or \\|, add\n `--trainer.enable_progress_bar false` on the command line.\n- The target task fields are overriden if their heads are active.\n- Use [`scripts/pretokenize.py`](scripts/pretokenize.py) to convert raw text\n files to CoNLL-U input files.\n\nThis mode is invoked using the `predict` subcommand, like so:\n\n udtube predict --config path/to/config.yaml --ckpt_path path/to/checkpoint.ckpt\n\n## Examples\n\nSee [`examples`](examples/README.md) for some worked examples including\nhyperparameter sweeping with [Weights & Biases](https://wandb.ai/site).\n\n## Additional scripts\n\nSee [`scripts/README.md`](scripts/README.md) for details on provided scripts not\nmention above.\n\n## License\n\nUDTube is distributed under an [Apache 2.0 license](LICENSE.txt).\n\n## Contribution\n\nWe welcome contributions using the fork-and-pull model.\n\n## References\n\nIf you use UDTube in your research, we would appreciate it if you cited the\nfollowing document, which describes the model:\n\nYakubov, D. 2024. [How do we learn what we cannot\nsay?](https://academicworks.cuny.edu/gc_etds/5622/) Master's thesis, CUNY\nGraduate Center.\n",
"bugtrack_url": null,
"license": "Apache 2.0",
"summary": "Neural morphological analysis",
"version": "0.1.0",
"project_urls": {
"homepage": "https://github.com/CUNY-CL/udtube"
},
"split_keywords": [
"computational linguistics",
" morphology",
" natural language processing",
" language"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "dbfe30801f2afee5c38732023a357c2b2f463e906d7c4074eb7962e065b69e84",
"md5": "d6282f6579b42b956357c39b0465057a",
"sha256": "b84fdab58d8b8686b2576e5395519a85196c933f54640a47f599beeb0301a4a0"
},
"downloads": -1,
"filename": "udtube-0.1.0-py3-none-any.whl",
"has_sig": false,
"md5_digest": "d6282f6579b42b956357c39b0465057a",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.9",
"size": 33567,
"upload_time": "2024-10-28T21:47:07",
"upload_time_iso_8601": "2024-10-28T21:47:07.069685Z",
"url": "https://files.pythonhosted.org/packages/db/fe/30801f2afee5c38732023a357c2b2f463e906d7c4074eb7962e065b69e84/udtube-0.1.0-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "38beb8be57725a348274e992ec1e3a88c3d71efbcd2d1b5d2b8c4f6837b3d668",
"md5": "a08f0d4203bd8909f60bd904d96cba85",
"sha256": "c8d936844860650950ed1dd9e146ac512bbedd9d38f268beeaf89dd5479628f1"
},
"downloads": -1,
"filename": "udtube-0.1.0.tar.gz",
"has_sig": false,
"md5_digest": "a08f0d4203bd8909f60bd904d96cba85",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.9",
"size": 31776,
"upload_time": "2024-10-28T21:47:09",
"upload_time_iso_8601": "2024-10-28T21:47:09.014470Z",
"url": "https://files.pythonhosted.org/packages/38/be/b8be57725a348274e992ec1e3a88c3d71efbcd2d1b5d2b8c4f6837b3d668/udtube-0.1.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-10-28 21:47:09",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "CUNY-CL",
"github_project": "udtube",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"circle": true,
"requirements": [],
"lcname": "udtube"
}
JSON
