Merge pull request #22 from bioscan-ml/fix/pin-datasets-version

fix: pin dependency versions for Python and datasets compatibility

Author: millanp95

README.md

@@ -37,10 +37,21 @@ features = output.mean(1)
 
 ### Reproducing the results from the paper
 
-0. Clone this repository and install the required libraries by running
+0. Clone this repository and install the required libraries.
+The instructions below assume a working `pip`- or
+[`uv`](https://docs.astral.sh/uv/)-managed Python environment.
+Requires **Python 3.11 or 3.12**
+(`torchtext`, a deprecated dependency,
+does not provide wheels for 3.13+).
+See [`pyproject.toml`](pyproject.toml) for the full list of
+pinned dependencies.
 ```shell
 pip install -e .
 ```
+Or, using `uv`:
+```shell
+uv sync
+```
 
 1. Download the data from our Hugging Face Dataset [repository](https://huggingface.co/datasets/bioscan-ml/CanadianInvertebrates-ML)
 ```shell

data/download_HF_CanInv.py

@@ -1,5 +1,8 @@
 from datasets import load_dataset
 
+# trust_remote_code=True is required because the HF dataset uses a custom
+# loading script. This requires datasets<4 (v4+ removed script support).
+# See: https://github.com/bioscan-ml/BarcodeBERT/issues/21
 ds = load_dataset("bioscan-ml/CanadianInvertebrates-ML", trust_remote_code=True)
 print("Formatting the data into CSV files ...")
 

pyproject.toml

@@ -9,14 +9,16 @@ authors = [
 ]
 description = "repository for training and evaluating BarcodeBERT"
 readme = "README.md"
-requires-python = ">=3.11"
+requires-python = ">=3.11,<3.13"
 dependencies = [
     "accelerate>=0.23.0",
     "boto3>=1.28.51",
     "botocore>=1.31.51",
     "cmake>=3.27.5",
+    "datasets>=2.16,<4",
     "einops>=0.6.1",
     "matplotlib",
+    "numba>=0.59",
     "numpy>=1.25.0",
     "omegaconf>=2.3.0",
     "opt-einsum>=3.3.0",