ecfr-textcat / README.md
DagimB's picture
Update README.md (#7)
124a776 verified
metadata
language: en
tags:
  - machine learning
  - natural language processing
  - huggingface

πŸͺ Weasel Project: Citations of ECFR Banking Regulation in a spaCy pipeline.

Custom text classification project for spaCy v3 adapted from the spaCy v3

πŸ“‹ project.yml

The project.yml defines the data assets required by the project, as well as the available commands and workflows. For details, see the Weasel documentation.

⏯ Commands

The following commands are defined by the project. They can be executed using weasel run [name]. Commands are only re-run if their inputs have changed.

Command Description
format-script Execute the Python script firstStep-format.py, which performs the initial formatting of a dataset file for the first step of the project. This script extracts text and labels from a dataset file in JSONL format and writes them to a new JSONL file in a specific format.

Usage:

spacy project run execute-first-step-format-script

Explanation:

  • The script firstStep-format.py reads data from the file specified in the dataset_file variable (data/train200.jsonl by default).
  • It extracts text and labels from each JSON object in the dataset file.
  • If both text and at least one label are available, it writes a new JSON object to the output file specified in the output_file variable (data/firstStep_file.jsonl by default) with the extracted text and label.
  • If either text or label is missing in a JSON object, a warning message is printed.
  • Upon completion, the script prints a message confirming the processing and the path to the output file. | | train-text-classification-model | Train the text classification model for the second step of the project using the secondStep-score.py script. This script loads a blank English spaCy model and adds a text classification pipeline to it. It then trains the model using the processed data from the first step.

Usage:

spacy project run train-text-classification-model

Explanation:

  • The script secondStep-score.py loads a blank English spaCy model and adds a text classification pipeline to it.
  • It reads processed data from the file specified in the processed_data_file variable (data/firstStep_file.jsonl by default).
  • The processed data is converted to spaCy format for training the model.
  • The model is trained using the converted data for a specified number of iterations (n_iter).
  • Losses are printed for each iteration during training.
  • Upon completion, the trained model is saved to the specified output directory (./my_trained_model by default). | | classify-unlabeled-data | Classify the unlabeled data for the third step of the project using the thirdStep-label.py script. This script loads the trained spaCy model from the previous step and classifies each record in the unlabeled dataset.

Usage:

spacy project run classify-unlabeled-data

Explanation:

  • The script thirdStep-label.py loads the trained spaCy model from the specified model directory (./my_trained_model by default).
  • It reads the unlabeled data from the file specified in the unlabeled_data_file variable (data/train.jsonl by default).
  • Each record in the unlabeled data is classified using the loaded model.
  • The predicted labels for each record are extracted and stored along with the text.
  • The classified data is optionally saved to a file specified in the output_file variable (data/thirdStep_file.jsonl by default). | | format-labeled-data | Format the labeled data for the final step of the project using the finalStep-formatLabel.py script. This script processes the classified data from the third step and transforms it into a specific format, considering a threshold for label acceptance.

Usage:

spacy project run format-labeled-data

Explanation:

  • The script finalStep-formatLabel.py reads classified data from the file specified in the input_file variable (data/thirdStep_file.jsonl by default).
  • For each record, it determines accepted categories based on a specified threshold.
  • It constructs an output record containing the text, predicted labels, accepted categories, answer (accept/reject), and options with meta information.
  • The transformed data is written to the file specified in the output_file variable (data/train4465.jsonl by default). | | setup-environment | Set up the Python virtual environment. | | review-evaluation-data | Review the evaluation data in Prodigy and automatically accept annotations.

Usage:

spacy project run review-evaluation-data

Explanation:

  • The command reviews the evaluation data in Prodigy.
  • It automatically accepts annotations made during the review process.
  • Only sessions allowed by the environment variable PRODIGY_ALLOWED_SESSIONS are permitted to review data. In this case, the session 'reviwer' is allowed. | | export-reviewed-evaluation-data | Export the reviewed evaluation data from Prodigy to a JSONL file named 'goldenEval.jsonl'.

Usage:

spacy project run export-reviewed-evaluation-data

Explanation:

  • The command exports the reviewed evaluation data from Prodigy to a JSONL file.
  • The data is exported from the Prodigy database associated with the project named 'project3eval-review'.
  • The exported data is saved to the file 'goldenEval.jsonl'.
  • This command helps in preserving the reviewed annotations for further analysis or processing. | | import-training-data | Import the training data into Prodigy from a JSONL file named 'train200.jsonl'.

Usage:

spacy project run import-training-data

Explanation:

  • The command imports the training data into Prodigy from the specified JSONL file.
  • The data is imported into the Prodigy database associated with the project named 'prodigy3train'.
  • This command prepares the training data for annotation and model training in Prodigy. | | import-golden-evaluation-data | Import the golden evaluation data into Prodigy from a JSONL file named 'goldeneval.jsonl'.

Usage:

spacy project run import-golden-evaluation-data

Explanation:

  • The command imports the golden evaluation data into Prodigy from the specified JSONL file.
  • The data is imported into the Prodigy database associated with the project named 'golden3'.
  • This command prepares the golden evaluation data for further analysis and model evaluation in Prodigy. | | train-model-experiment1 | Train a text classification model using Prodigy with the 'prodigy3train' dataset and evaluating on 'golden3'.

Usage:

spacy project run train-model-experiment1

Explanation:

  • The command trains a text classification model using Prodigy.
  • It uses the 'prodigy3train' dataset for training and evaluates the model on the 'golden3' dataset.
  • The trained model is saved to the './output/experiment1' directory. | | download-model | Download the English language model 'en_core_web_lg' from spaCy.

Usage:

spacy project run download-model

Explanation:

  • The command downloads the English language model 'en_core_web_lg' from spaCy.
  • This model is used as the base model for further data processing and training in the project. | | convert-data-to-spacy-format | Convert the annotated data from Prodigy to spaCy format using the 'prodigy3train' and 'golden3' datasets.

Usage:

spacy project run convert-data-to-spacy-format

Explanation:

  • The command converts the annotated data from Prodigy to spaCy format.
  • It uses the 'prodigy3train' and 'golden3' datasets for conversion.
  • The converted data is saved to the './corpus' directory with the base model 'en_core_web_lg'. | | train-custom-model | Train a custom text classification model using spaCy with the converted data in spaCy format.

Usage:

spacy project run train-custom-model

Explanation:

  • The command trains a custom text classification model using spaCy.
  • It uses the converted data in spaCy format located in the './corpus' directory.
  • The model is trained using the configuration defined in 'corpus/config.cfg'. |

⏭ Workflows

The following workflows are defined by the project. They can be executed using weasel run [name] and will run the specified commands in order. Commands are only re-run if their inputs have changed.

Workflow Steps
all format-script β†’ train-text-classification-model β†’ classify-unlabeled-data β†’ format-labeled-data β†’ setup-environment β†’ review-evaluation-data β†’ export-reviewed-evaluation-data β†’ import-training-data β†’ import-golden-evaluation-data β†’ train-model-experiment1 β†’ download-model β†’ convert-data-to-spacy-format β†’ train-custom-model

πŸ—‚ Assets

The following assets are defined by the project. They can be fetched by running weasel assets in the project directory.

File Source Description
corpus/labels/ner.json Local JSON file containing NER labels
corpus/labels/parser.json Local JSON file containing parser labels
corpus/labels/tagger.json Local JSON file containing tagger labels
corpus/labels/textcat_multilabel.json Local JSON file containing multilabel text classification labels
data/eval.jsonl Local JSONL file containing evaluation data
data/firstStep_file.jsonl Local JSONL file containing formatted data from the first step
data/five_examples_annotated5.jsonl Local JSONL file containing five annotated examples
data/goldenEval.jsonl Local JSONL file containing golden evaluation data
data/thirdStep_file.jsonl Local JSONL file containing classified data from the third step
data/train.jsonl Local JSONL file containing training data
data/train200.jsonl Local JSONL file containing initial training data
data/train4465.jsonl Local JSONL file containing formatted and labeled training data
my_trained_model/textcat_multilabel/cfg Local Configuration files for the text classification model
my_trained_model/textcat_multilabel/model Local Trained model files for the text classification model
my_trained_model/vocab/key2row Local Mapping from keys to row indices in the vocabulary
my_trained_model/vocab/lookups.bin Local Binary lookups file for the vocabulary
my_trained_model/vocab/strings.json Local JSON file containing string representations of the vocabulary
my_trained_model/vocab/vectors Local Directory containing vector files for the vocabulary
my_trained_model/vocab/vectors.cfg Local Configuration file for vectors in the vocabulary
my_trained_model/config.cfg Local Configuration file for the trained model
my_trained_model/meta.json Local JSON file containing metadata for the trained model
my_trained_model/tokenizer Local Tokenizer files for the trained model
output/experiment1/model-best/textcat_multilabel/cfg Local Configuration files for the best model in experiment 1
output/experiment1/model-best/textcat_multilabel/model Local Trained model files for the best model in experiment 1
output/experiment1/model-best/vocab/key2row Local Mapping from keys to row indices in the vocabulary for the best model in experiment 1
output/experiment1/model-best/vocab/lookups.bin Local Binary lookups file for the vocabulary for the best model in experiment 1
output/experiment1/model-best/vocab/strings.json Local JSON file containing string representations of the vocabulary for the best model in experiment 1
output/experiment1/model-best/vocab/vectors Local Directory containing vector files for the vocabulary for the best model in experiment 1
output/experiment1/model-best/vocab/vectors.cfg Local Configuration file for vectors in the vocabulary for the best model in experiment 1
output/experiment1/model-best/config.cfg Local Configuration file for the best model in experiment 1
output/experiment1/model-best/meta.json Local JSON file containing metadata for the best model in experiment 1
output/experiment1/model-best/tokenizer Local Tokenizer files for the best model in experiment 1
output/experiment1/model-last/textcat_multilabel/cfg Local Configuration files for the last model in experiment 1
output/experiment1/model-last/textcat_multilabel/model Local Trained model files for the last model in experiment 1
output/experiment1/model-last/vocab/key2row Local Mapping from keys to row indices in the vocabulary for the last model in experiment 1
output/experiment1/model-last/vocab/lookups.bin Local Binary lookups file for the vocabulary for the last model in experiment 1
output/experiment1/model-last/vocab/strings.json Local JSON file containing string representations of the vocabulary for the last model in experiment 1
output/experiment1/model-last/vocab/vectors Local Directory containing vector files for the vocabulary for the last model in experiment 1
output/experiment1/model-last/vocab/vectors.cfg Local Configuration file for vectors in the vocabulary for the last model in experiment 1
output/experiment1/model-last/config.cfg Local Configuration file for the last model in experiment 1
output/experiment1/model-last/meta.json Local JSON file containing metadata for the last model in experiment 1
output/experiment1/model-last/tokenizer Local Tokenizer files for the last model in experiment 1
output/experiment3/model-best/textcat_multilabel/cfg Local Configuration files for the best model in experiment 3
output/experiment3/model-best/textcat_multilabel/model Local Trained model files for the best model in experiment 3
output/experiment3/model-best/vocab/key2row Local Mapping from keys to row indices in the vocabulary for the best model in experiment 3
output/experiment3/model-best/vocab/lookups.bin Local Binary lookups file for the vocabulary for the best model in experiment 3
output/experiment3/model-best/vocab/strings.json Local JSON file containing string representations of the vocabulary for the best model in experiment 3
output/experiment3/model-best/vocab/vectors Local Directory containing vector files for the vocabulary for the best model in experiment 3
output/experiment3/model-best/vocab/vectors.cfg Local Configuration file for vectors in the vocabulary for the best model in experiment 3
output/experiment3/model-best/config.cfg Local Configuration file for the best model in experiment 3
output/experiment3/model-best/meta.json Local JSON file containing metadata for the best model in experiment 3
output/experiment3/model-best/tokenizer Local Tokenizer files for the best model in experiment 3
output/experiment3/model-last/textcat_multilabel/cfg Local Configuration files for the last model in experiment 3
output/experiment3/model-last/textcat_multilabel/model Local Trained model files for the last model in experiment 3
output/experiment3/model-last/vocab/key2row Local Mapping from keys to row indices in the vocabulary for the last model in experiment 3
output/experiment3/model-last/vocab/lookups.bin Local Binary lookups file for the vocabulary for the last model in experiment 3
output/experiment3/model-last/vocab/strings.json Local JSON file containing string representations of the vocabulary for the last model in experiment 3
output/experiment3/model-last/vocab/vectors Local Directory containing vector files for the vocabulary for the last model in experiment 3
output/experiment3/model-last/vocab/vectors.cfg Local Configuration file for vectors in the vocabulary for the last model in experiment 3
output/experiment3/model-last/config.cfg Local Configuration file for the last model in experiment 3
output/experiment3/model-last/meta.json Local JSON file containing metadata for the last model in experiment 3
output/experiment3/model-last/tokenizer Local Tokenizer files for the last model in experiment 3
python_Code/finalStep-formatLabel.py Local Python script for formatting labeled data in the final step
python_Code/firstStep-format.py Local Python script for formatting data in the first step
python_Code/five_examples_annotated.ipynb Local Jupyter notebook containing five annotated examples
python_Code/secondStep-score.py Local Python script for scoring data in the second step
python_Code/thirdStep-label.py Local Python script for labeling data in the third step
python_Code/train_eval_split.ipynb Local Jupyter notebook for training and evaluation data splitting
TerminalCode.txt Local Text file containing terminal code
README.md Local Markdown file containing project documentation
prodigy.json Local JSON file containing Prodigy configuration