Improve readmes.

Update version.

CHANGES.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## 0.5.1
+- Model versioning scheme implemented.
+- Added convenience method to clear cache.
+
 ## 0.5.0
 - Model serialization has been completely rewritten to make it easier to use models out of the box.
 - Convenience method `matgl.load_model` is now the default way to load models.

README.md

@@ -33,11 +33,13 @@ This effort is a collaboration between the [Materials Virtual Lab](http://materi
 
 ## Status
 
-- Jun 8 2023: Simplified saving and loading of models. Now models can be loaded with one line of code!
-- Jun 7 2023: Near feature parity with original TF implementations. Re-trained M3Gnet universal potential now
+Major milestones are summarized below. Full change log is provided [here](https://materialsvirtuallab.github.io/matgl/changes).
+- v0.5.1 (Jun 9 2023): Model versioning implemented.
+- v0.5.0 (Jun 8 2023): Simplified saving and loading of models. Now models can be loaded with one line of code!
+- v0.4.0 (Jun 7 2023): Near feature parity with original TF implementations. Re-trained M3Gnet universal potential now
   available.
-- Apr 26 2023: Pre-trained MEGNet models now available for formation energies and band gaps!
-- Feb 16 2023: Both initial implementations of M3GNet and MEGNet architectures have been completed. Expect bugs!
+- v0.1.0 (Feb 16 2023): Initial implementations of M3GNet and MEGNet architectures have been completed. Expect
+  bugs!
 
 ## Architectures
 
@@ -113,7 +115,9 @@ eform = model.predict_structure(struct)
 print(f"The predicted formation energy for CsCl is {float(eform.numpy()):.3f} eV/atom.")
 ```
 
-More examples are available [here](examples).
+## Example notebooks
+
+Primary usage documentation will be done via Jupyter notes, which are available [here](examples).
 
 ## Docs
 

pretrained_models/M3GNet-MP-2021.2.8-PES/README.md

@@ -8,19 +8,16 @@ in the dataset.
 # Training dataset
 
 MP-2021.2.8: Materials Project structure relaxations as of 2021.2.8.
-
-Number of structures for training: 167237
-
-Number of structures for validation: 18584
+- Training set size: 167237
+- Validation set size: 18584
 
 # Performance metrics
 
 ## Training and validation errors
-The reported numbers are mean absolute error of energies, forces and stresses.
-
-Train: 19.977 eV/atom, 0.063 eV/A, 0.272 GPa
 
-Valid: 23.713 eV/atom, 0.070 eV/A, 0.394 GPa
+MAEs of energies, forces and stresses, respectively
+- Training: 19.977 eV/atom, 0.063 eV/A, 0.272 GPa
+- Validation: 23.713 eV/atom, 0.070 eV/A, 0.394 GPa
 
 ## Cubic crystals benchmark
 

pretrained_models/MEGNet-MP-2018.6.1-Eform/README.md

@@ -7,22 +7,16 @@ originally implemented in tensorflow.
 # Training dataset
 
 MP-2018.6.1: Materials Project formation energy as of 2018.6.1.
-
-Number of structures for training: 62315
-
-Number of structures for validation: 3461
-
-Number of structures for testing: 3463
+- Training set size: : 62315
+- Validation set size: 3461
+- Test set size: 3463
 
 # Performance metrics
 
-The reported numbers are mean absolute error of formation energy in eV/atom.
-
-Train: 0.010 eV/atom
-
-Valid: 0.029 eV/atom
-
-Test: 0.028 eV/atom
+MAE of formation energy in eV/atom.
+- Training: 0.010 eV/atom
+- Validation: 0.029 eV/atom
+- Test: 0.028 eV/atom
 
 # References
 

pretrained_models/MEGNet-MP-2019.4.1-BandGap-mfi/README.md

@@ -6,19 +6,15 @@ version of the MEGNet band gap model originally implemented in tensorflow.
 
 # Training dataset
 
-MP-2019.4.1: Materials Project formation energy as of 2019.4.1.
-
-Number of structures for training: 48912
-
-Number of structures for validation: 879
+MP-2019.4.1: Materials Project band gap as of 2019.4.1.
+- Training set size: 48912
+- Validation set size: 879
 
 # Performance metrics
 
-The reported numbers are mean absolute error of formation energy in eV
-
-Train: 0.075 eV
-
-Valid: 0.314 eV
+MAE of band gap in eV
+- Training: 0.075 eV
+- Validation: 0.314 eV
 
 # References
 

setup.py

@@ -12,7 +12,7 @@
 
 setup(
     name="matgl",
-    version="0.5.0",
+    version="0.5.1",
     author="Tsz Wai Ko, Marcel Nassar, Ji Qi, Santiago Miret, Shyue Ping Ong",
     author_email="[email protected], [email protected]",
     maintainer="Shyue Ping Ong",