README.md

---
language: en
license: other
library_name: span-marker
tags:
- span-marker
- token-classification
- ner
- named-entity-recognition
- generated_from_span_marker_trainer
datasets:
- tner/bionlp2004
metrics:
- precision
- recall
- f1
widget:
- text: Coexpression of HMG I/Y and Oct-2 in cell lines lacking Oct-2 results in high
    levels of HLA-DRA gene expression, and in vitro DNA-binding studies reveal that
    HMG I/Y stimulates Oct-2A binding to the HLA-DRA promoter.
- text: In erythroid cells most of the transcription activity was contained in a 150
    bp promoter fragment with binding sites for transcription factors AP2, Sp1 and
    the erythroid-specific GATA-1.
- text: 'Synergy between signal transduction pathways is obligatory for expression
    of c-fos in B and T cell lines: implication for c-fos control via surface immunoglobulin
    and T cell antigen receptors.'
- text: CIITA mRNA is normally inducible by IFN-gamma in class II non-inducible,
    RB-defective lines, and in one line, re-expression of RB has no effect on CIITA
    mRNA induction levels.
- text: As we reported previously, MNDA mRNA level in adherent monocytes is elevated
    by IFN-alpha; in this study, we further assessed MNDA expression in in vitro
    monocyte-derived macrophages.
pipeline_tag: token-classification
co2_eq_emissions:
  emissions: 45.104
  source: codecarbon
  training_type: fine-tuning
  on_cloud: false
  gpu_model: 1 x NVIDIA GeForce RTX 3090
  cpu_model: 13th Gen Intel(R) Core(TM) i7-13700K
  ram_total_size: 31.777088165283203
  hours_used: 0.296
model-index:
- name: SpanMarker with bert-base-uncased on BioNLP2004
  results:
  - task:
      type: token-classification
      name: Named Entity Recognition
    dataset:
      name: BioNLP2004
      type: tner/bionlp2004
      split: test
    metrics:
    - type: f1
      value: 0.7620637836032726
      name: F1
    - type: precision
      value: 0.7289958470876371
      name: Precision
    - type: recall
      value: 0.7982742537313433
      name: Recall
---

# SpanMarker with bert-base-uncased on BioNLP2004

This is a [SpanMarker](https://github.com/tomaarsen/SpanMarkerNER) model trained on the [BioNLP2004](https://huggingface.co/datasets/tner/bionlp2004) dataset that can be used for Named Entity Recognition. This SpanMarker model uses [bert-base-uncased](https://huggingface.co/bert-base-uncased) as the underlying encoder. See [train.py](train.py) for the training script.

## Model Details

### Model Description

- **Model Type:** SpanMarker
- **Encoder:** [bert-base-uncased](https://huggingface.co/bert-base-uncased)
- **Maximum Sequence Length:** 256 tokens
- **Maximum Entity Length:** 8 words
- **Training Dataset:** [BioNLP2004](https://huggingface.co/datasets/tner/bionlp2004)
- **Language:** en
- **License:** other

### Model Sources

- **Repository:** [SpanMarker on GitHub](https://github.com/tomaarsen/SpanMarkerNER)
- **Thesis:** [SpanMarker For Named Entity Recognition](https://raw.githubusercontent.com/tomaarsen/SpanMarkerNER/main/thesis.pdf)

### Model Labels
| Label     | Examples                                                                                         |
|:----------|:-------------------------------------------------------------------------------------------------|
| DNA       | "immunoglobulin heavy-chain enhancer", "enhancer", "immunoglobulin heavy-chain ( IgH ) enhancer" |
| RNA       | "GATA-1 mRNA", "c-myb mRNA", "antisense myb RNA"                                                 |
| cell_line | "monocytic U937 cells", "TNF-treated HUVECs", "HUVECs"                                           |
| cell_type | "B cells", "non-B cells", "human red blood cells"                                                |
| protein   | "ICAM-1", "VCAM-1", "NADPH oxidase"                                                              |

## Evaluation

### Metrics
| Label     | Precision | Recall | F1     |
|:----------|:----------|:-------|:-------|
| **all**   | 0.7290    | 0.7983 | 0.7621 |
| DNA       | 0.7174    | 0.7505 | 0.7336 |
| RNA       | 0.6977    | 0.7692 | 0.7317 |
| cell_line | 0.5831    | 0.7020 | 0.6370 |
| cell_type | 0.8222    | 0.7381 | 0.7779 |
| protein   | 0.7196    | 0.8407 | 0.7755 |

## Uses

### Direct Use

```python
from span_marker import SpanMarkerModel

# Download from the 🤗 Hub
model = SpanMarkerModel.from_pretrained("tomaarsen/span-marker-bert-base-uncased-bionlp")
# Run inference
entities = model.predict("In erythroid cells most of the transcription activity was contained in a 150 bp promoter fragment with binding sites for transcription factors AP2, Sp1 and the erythroid-specific GATA-1.")
```

### Downstream Use
You can finetune this model on your own dataset.

<details><summary>Click to expand</summary>

```python
from span_marker import SpanMarkerModel, Trainer

# Download from the 🤗 Hub
model = SpanMarkerModel.from_pretrained("tomaarsen/span-marker-bert-base-uncased-bionlp")

# Specify a Dataset with "tokens" and "ner_tag" columns
dataset = load_dataset("conll2003") # For example CoNLL2003

# Initialize a Trainer using the pretrained model & dataset
trainer = Trainer(
    model=model,
    train_dataset=dataset["train"],
    eval_dataset=dataset["validation"],
)
trainer.train()
trainer.save_model("tomaarsen/span-marker-bert-base-uncased-bionlp-finetuned")
```
</details>

## Training Details

### Training Set Metrics
| Training set          | Min | Median  | Max |
|:----------------------|:----|:--------|:----|
| Sentence length       | 2   | 26.5790 | 166 |
| Entities per sentence | 0   | 2.7528  | 23  |

### Training Hyperparameters
- learning_rate: 5e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 3

### Training Results
| Epoch  | Step | Validation Loss | Validation Precision | Validation Recall | Validation F1 | Validation Accuracy |
|:------:|:----:|:---------------:|:--------------------:|:-----------------:|:-------------:|:-------------------:|
| 0.4505 | 300  | 0.0210          | 0.7497               | 0.7659            | 0.7577        | 0.9254              |
| 0.9009 | 600  | 0.0162          | 0.8048               | 0.8217            | 0.8131        | 0.9432              |
| 1.3514 | 900  | 0.0154          | 0.8126               | 0.8249            | 0.8187        | 0.9434              |
| 1.8018 | 1200 | 0.0149          | 0.8148               | 0.8451            | 0.8296        | 0.9481              |
| 2.2523 | 1500 | 0.0150          | 0.8297               | 0.8438            | 0.8367        | 0.9501              |
| 2.7027 | 1800 | 0.0145          | 0.8280               | 0.8443            | 0.8361        | 0.9501              |

### Environmental Impact
Carbon emissions were measured using [CodeCarbon](https://github.com/mlco2/codecarbon).
- **Carbon Emitted**: 0.045 kg of CO2
- **Hours Used**: 0.296 hours

### Training Hardware
- **On Cloud**: No
- **GPU Model**: 1 x NVIDIA GeForce RTX 3090
- **CPU Model**: 13th Gen Intel(R) Core(TM) i7-13700K
- **RAM Size**: 31.78 GB

### Framework Versions

- Python: 3.9.16
- SpanMarker: 1.3.1.dev
- Transformers : 4.29.2
- PyTorch: 2.0.1+cu118
- Datasets: 2.14.3
- Tokenizers: 0.13.2