Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update recogniser #201

Merged
merged 6 commits into from
Mar 23, 2023
Merged

Update recogniser #201

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
d93c8c1
Update recogniser
mcollardanuy Mar 22, 2023
b95feff
Remove deezy exception
mcollardanuy Mar 22, 2023
72c5df3
Merge pull request #202 from Living-with-machines/106-remove-try-deezy
fedenanni Mar 23, 2023
3ba38d2
Add test for n-dash fix
fedenanni Mar 23, 2023
05bae6a
Merge branch '196-update-recogniser' of https://github.com/Living-wit…
fedenanni Mar 23, 2023
6529533
Add test for loading model from hum
fedenanni Mar 23, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
144 changes: 144 additions & 0 deletions examples/load_use_ner_model.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,144 @@
{
"cells": [
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"# Loading and using a NER model\n",
"\n",
"This notebook shows how to load an existing named entity recognition (NER) model from the HuggingFace hub.\n",
"\n",
"We start by importing some libraries, and the `recogniser` script from the `geoparser` folder:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"import sys\n",
"\n",
"sys.path.insert(0, os.path.abspath(os.path.pardir))\n",
"from geoparser import recogniser"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"Create a `myner` object of the `Recogniser` class.\n",
"\n",
"We only need to pass the path to the model in `model` and set `load_from_hum` to True, as follows:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"myner = recogniser.Recogniser(\n",
" model=\"dslim/bert-base-NER\", # Path to the HuggingFace model\n",
" load_from_hub=True, # Set this to True\n",
")"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"Print the Recogniser (see that most fields are empty, because they are not needed):"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"print(myner)"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"If we try to train the model, nothing happens, because we're loading an existing model:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"myner.train()"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"Now, to use the model you want to use, you'll need to load it into a NER pipeline:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"myner.model, myner.pipe = myner.create_pipeline()"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"And, finally, use the newly trained model to predict the named entities in a sentence."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"sentence = \"A remarkable case of rattening has just occurred in the building trade at Sheffield.\"\n",
"\n",
"predictions = myner.ner_predict(sentence)\n",
"print(predictions) # Note that, if you've trained the model in the test mode, the model will probably not identify \"Sheffield\" as a location."
]
}
],
"metadata": {
"kernelspec": {
"display_name": "resolution-cNmUJBkC-py3.9",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.7"
},
"orig_nbformat": 4
},
"nbformat": 4,
"nbformat_minor": 2
}
40 changes: 27 additions & 13 deletions examples/train_use_ner_model.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
"source": [
"# Training and using a NER model\n",
"\n",
"This notebook shows how to train a new toponym recognition (NER) model.\n",
"This notebook shows how to train a new toponym recognition (NER) model. You can see how to use a model from the HuggingFace hub in the `load_use_ner_model.ipynb` notebook.\n",
"\n",
"We start by importing some libraries, and the `recogniser` script from the `geoparser` folder:"
]
Expand Down Expand Up @@ -36,9 +36,7 @@
"> ```json\n",
"> {\"id\":\"3896239_29\",\"ner_tags\":[\"O\",\"B-STREET\",\"I-STREET\",\"O\",\"O\",\"O\",\"B-BUILDING\",\"I-BUILDING\",\"O\",\"O\",\"O\",\"O\",\"O\",\"O\",\"O\",\"O\",\"O\",\"O\"],\"tokens\":[\",\",\"Old\",\"Millgate\",\",\",\"to\",\"the\",\"Collegiate\",\"Church\",\",\",\"where\",\"they\",\"arrived\",\"a\",\"little\",\"after\",\"ten\",\"oclock\",\".\"]}\n",
"> {\"id\":\"8262498_11\",\"ner_tags\":[\"O\",\"O\",\"O\",\"O\",\"O\",\"O\",\"O\",\"O\",\"O\",\"O\",\"O\",\"B-LOC\",\"O\",\"B-LOC\",\"O\",\"O\",\"O\",\"O\",\"O\",\"O\"],\"tokens\":[\"On\",\"the\",\"\\u2018\",\"JSth\",\"November\",\"the\",\"ship\",\"Santo\",\"Christo\",\",\",\"from\",\"Monteveido\",\"to\",\"Cadiz\",\",\",\"with\",\"hides\",\"and\",\"copper\",\".\"]}\n",
"> ```\n",
">\n",
"> **Note:** If your tagset is different from the one we use (`[\"O\", \"B-LOC\", \"I-LOC\", \"B-STREET\", \"I-STREET\", \"B-BUILDING\", \"I-BUILDING\", \"B-OTHER\", \"I-OTHER\", \"B-FICTION\", \"I-FICTION\"]`), you will need to manually modify the mapping dictionaries `map_tag_label()` and `encode_dict()` in the script [utils/ner.py](https://github.com/Living-with-machines/toponym-resolution/blob/main/utils/ner.py) accordingly. In the future, we plan to improve this bit so it's automatically done, but for the time being you will need to do this manually."
"> ```"
]
},
{
Expand All @@ -48,30 +46,46 @@
"outputs": [],
"source": [
"myner = recogniser.Recogniser(\n",
" model_name=\"your_chosen_model_name\", # The name of your NER model (note: we may append suffixes to it).\n",
" model=None, # We'll store the NER model here, leave this empty.\n",
" model=\"blb_lwm-ner-fine\",\n",
" train_dataset=\"../experiments/outputs/data/lwm/ner_fine_train.json\", # Path to the json file containing the training set (see note above).\n",
" test_dataset=\"../experiments/outputs/data/lwm/ner_fine_dev.json\", # Path to the json file containing the test set (see note above).\n",
" pipe=None, # We'll store the NER pipeline here, leave this empty.\n",
" base_model=\"khosseini/bert_1760_1900\", # Base model to fine-tune for NER. The value can be: either \n",
" # your local path to a model or the huggingface path.\n",
" # In this case, we use the huggingface path:\n",
" # https://huggingface.co/khosseini/bert_1760_1900). You can\n",
" # chose any other model from the HuggingFace hub, as long as it's\n",
" # trained on the \"Fill-Mask\" objective (filter by task).\n",
" train_dataset=\"../experiments/outputs/data/lwm/ner_df_train.json\", # Path to the json file containing the training set (see note above).\n",
" test_dataset=\"../experiments/outputs/data/lwm/ner_df_dev.json\", # Path to the json file containing the test set (see note above).\n",
" output_model_path=\"../resources/models/\", # Path where the NER model will be stored.\n",
" model_path=\"../resources/models/\", # Path where the NER model will be stored.\n",
" training_args={\n",
" \"learning_rate\": 5e-5,\n",
" \"batch_size\": 16,\n",
" \"num_train_epochs\": 4,\n",
" \"weight_decay\": 0.01,\n",
" }, # Training arguments: you can change them.\n",
" overwrite_training=True, # Set to True if you want to overwrite an existing model with the same name.\n",
" do_test=True, # Set to True if you want to perform the training on test mode (the string \"_test\" will be appended to your model name).\n",
" training_tagset=\"fine\", # Options are: \"coarse\" (will cluster all tags under LOC) or \"fine\" (will keep the different tags). See note above.\n",
" overwrite_training=False, # Set to True if you want to overwrite an existing model with the same name.\n",
" do_test=False, # Set to True if you want to perform the training on test mode (the string \"_test\" will be appended to your model name).\n",
" load_from_hub=False, # Whether the model should be loaded from the HuggingFace hub\n",
")"
]
},
{
"attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"Print the Recogniser:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"print(myner)"
]
},
{
"cell_type": "markdown",
"metadata": {},
Expand Down Expand Up @@ -122,7 +136,7 @@
"sentence = \"A remarkable case of rattening has just occurred in the building trade at Sheffield.\"\n",
"\n",
"predictions = myner.ner_predict(sentence)\n",
"print(predictions) # Note that, if you've trained the model in the test mode, the model will probably not identify \"Sheffield\" as a location."
"print([pred for pred in predictions if pred[\"entity\"] != \"O\"]) # Note that, if you've trained the model in the test mode, the model will probably not identify \"Sheffield\" as a location."
]
}
],
Expand Down
22 changes: 20 additions & 2 deletions experiments/prepare_data.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,10 +79,28 @@
lwm_train_ner, lwm_dev_ner = train_test_split(
lwm_df, test_size=0.2, random_state=RANDOM_SEED
)

# Store LwM with fine-grained tags:
lwm_train_ner.to_json(
output_path_lwm + "ner_fine_train.json", orient="records", lines=True
)
lwm_dev_ner.to_json(output_path_lwm + "ner_fine_dev.json", orient="records", lines=True)

# Convert fine-grained tags to coarse:
lwm_train_ner["ner_tags"] = lwm_train_ner["ner_tags"].apply(
lambda x: preprocess_data.fine_to_coarse(x)
)
lwm_dev_ner["ner_tags"] = lwm_dev_ner["ner_tags"].apply(
lambda x: preprocess_data.fine_to_coarse(x)
)

# Store LwM with coarse-grained tags:
lwm_train_ner.to_json(
output_path_lwm + "ner_df_train.json", orient="records", lines=True
output_path_lwm + "ner_coarse_train.json", orient="records", lines=True
)
lwm_dev_ner.to_json(
output_path_lwm + "ner_coarse_dev.json", orient="records", lines=True
)
lwm_dev_ner.to_json(output_path_lwm + "ner_df_dev.json", orient="records", lines=True)

# Process data for the resolution experiments:
lwm_train_df = preprocess_data.process_lwm_for_linking(topres_path_train, gazetteer_ids)
Expand Down
Loading