Update README.md

README.md

@@ -21,7 +21,7 @@ This dataset contains the Materials Project-derived reference table used by Crys
 
 ## Dataset Fields
 
-The current Parquet file contains 200290 rows. The columns are:
+The current Parquet file contains 200290 rows. These columns are from Materials Project:
 
 ```text
 material_id
@@ -32,11 +32,16 @@ is_stable
 band_gap
 is_metal
 efermi
-bulk_modulus
-shear_modulus
 homogeneous_poisson
 description
 formula_pretty
+```
+
+and these columns are from our MLIP:
+
+```text
+bulk_modulus
+shear_modulus
 thermal_expansion_300k
 ```
 
@@ -46,9 +51,7 @@ The `structure` column is serialized in the project simple-crystal text format.
 
 Download the Parquet file and place it at the expected project path:
 
-```bash
-mkdir -p assets/MP
-python - <<'PY_DOWNLOAD'
+```python
 from huggingface_hub import hf_hub_download
 
 path = hf_hub_download(
@@ -58,13 +61,12 @@ path = hf_hub_download(
     local_dir="assets/MP",
 )
 print(path)
-PY_DOWNLOAD
 ```
 
 Then recreate the shelve database used by the training and evaluation code:
 
 ```bash
-python scripts/parquet_to_shelve.py   --source assets/MP/MP_shelve.parquet   --output assets/MP/MP_shelve   --write   --overwrite
+python scripts/parquet_to_shelve.py
 ```
 
 The reconstructed `assets/MP/MP_shelve` database is used by the training datasets, generation code, and ground-truth evaluation metrics.