Update code for new numpy/pyg/cuda dependencies

.gitignore

@@ -1,8 +1,8 @@
-01-benchmark_surfaces/
-01-benchmark_surfaces_npy/
-01-benchmark_pdbs_npy/
-01-benchmark_pdbs/
-01-benchmark_pdbs/
+
+# dMaSIF
+surface_data
+.vscode
+runs
 shape_index/
 masif_preds/
 runs/
@@ -13,6 +13,179 @@ NeurIPS_2020_benchmarks/
 *.out
 figures/
 timings/
-data_analysis/roc_curves
-data_analysis/.ipynb_checkpoints/
-.ipynb_checkpoints/
+data_analysis
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+Footer
+© 2023 GitHub, Inc.
+Footer navigation
+Terms
+Privacy
+Security
+Status
+Docs
+Contact GitHub
+Pricing
+API
+Training
+Blog
+About

README.md

@@ -44,16 +44,16 @@ Models have been trained on either a single NVIDIA RTX 2080 Ti or a single Tesla
 
 Scripts have been tested using the following two sets of core dependencies:
 
-| Dependency | First Option  | Second Option |
-| ------------- | ------------- | ------------- |
-| GCC | 7.5.0 | 8.4.0 |
-| CMAKE | 3.10.2 | 3.16.5 |
-| CUDA | 10.0.130 | 10.2.89  |
-| cuDNN | 7.6.4.38  | 7.6.5.32  |
-| Python | 3.6.9  | 3.7.7  |
-| PyTorch | 1.4.0  | 1.6.0  |
-| PyKeops | 1.4  | 1.4.1  |
-| PyTorch Geometric | 1.5.0  | 1.6.1  |
+| Dependency | First Option  | Second Option | Updated Version |
+| ------------- | ------------- | ------------- | ------------- |
+| GCC | 7.5.0 | 8.4.0 | 9.2.0 |
+| CMAKE | 3.10.2 | 3.16.5 | 3.22.2 |
+| CUDA | 10.0.130 | 10.2.89  | 11.7 |
+| cuDNN | 7.6.4.38  | 7.6.5.32  | 7.6.x |
+| Python | 3.6.9  | 3.7.7  | 3.8.16 |
+| PyTorch | 1.4.0  | 1.6.0  | 1.13.1 |
+| PyKeops | 1.4  | 1.4.1  | 2.1.1 |
+| PyTorch Geometric | 1.5.0  | 1.6.1  | 2.2.0 |
 
 
 ## Code overview

benchmark_models.py

@@ -320,12 +320,12 @@ def load_mesh(self, xyz, triangles=None, normals=None, weights=None, batch=None)
 
         # 3.b) Pseudo-geodesic window:
         # Pseudo-geodesic squared distance:
-        rho2_ij = ((x_j - x_i) ** 2).sum(-1) * ((2 - (n_i | n_j)) ** 2)  # (N, N, 1)
+        rho2_ij = ((x_j - x_i) ** 2).sum(-1) * ((2 - (n_i | n_j)) ** 2)  # (N, N, 1) # eq (5) in paper
         # Gaussian window:
         window_ij = (-rho2_ij).exp()  # (N, N, 1)
 
         # 3.c) Coordinates in the (u, v) basis - not oriented yet:
-        X_ij = uv_i.matvecmult(x_j - x_i)  # (N, N, 2)
+        X_ij = uv_i.matvecmult(x_j - x_i)  # (N, 1, 6)*(N, N, 3) -> (N, N, 2)
 
         # 3.d) Local average in the tangent plane:
         orientation_weight_ij = window_ij * weights_j  # (N, N, 1)

data.py

@@ -217,15 +217,15 @@ def __init__(
         self.rand_rot1 = rand_rot1
         self.rand_rot2 = rand_rot2
 
-    def __inc__(self, key, value):
+    def __inc__(self, key, value, *args, **kwargs):
         if key == "face_p1":
             return self.xyz_p1.size(0)
         if key == "face_p2":
             return self.xyz_p2.size(0)
         else:
             return super(PairData, self).__inc__(key, value)
 
-    def __cat_dim__(self, key, value):
+    def __cat_dim__(self, key, value, *args, **kwargs):
         if ("index" in key) or ("face" in key):
             return 1
         else:

data_iteration.py

@@ -321,7 +321,7 @@ def iterate(
                 if not args.single_protein:
                     P2["rand_rot"] = torch.eye(3, device=P2["xyz"].device)
                     P2["atom_center"] = torch.zeros((1, 3), device=P2["xyz"].device)
-                    
+
             torch.cuda.synchronize()
             prediction_time = time.time()
             outputs = net(P1, P2)

data_preprocessing/convert_pdb2npy.py

@@ -34,6 +34,9 @@ def load_structure_np(fname, center):
 def convert_pdbs(pdb_dir, npy_dir):
     print("Converting PDBs")
     for p in tqdm(pdb_dir.glob("*.pdb")):
-        protein = load_structure_np(p, center=False)
+        try:
+            protein = load_structure_np(p, center=False)
+        except:
+            print(p)
         np.save(npy_dir / (p.stem + "_atomxyz.npy"), protein["xyz"])
         np.save(npy_dir / (p.stem + "_atomtypes.npy"), protein["types"])

data_preprocessing/convert_ply2npy.py

@@ -42,7 +42,10 @@ def load_surface_np(fname, center):
 def convert_plys(ply_dir, npy_dir):
     print("Converting PLYs")
     for p in tqdm(ply_dir.glob("*.ply")):
-        protein = load_surface_np(p, center=False)
+        try:
+            protein = load_surface_np(p, center=False)
+        except:
+            print(p)
         np.save(npy_dir / (p.stem + "_xyz.npy"), protein["xyz"])
         np.save(npy_dir / (p.stem + "_triangles.npy"), protein["triangles"])
         np.save(npy_dir / (p.stem + "_features.npy"), protein["features"])

geometry_processing.py

@@ -541,13 +541,12 @@ def curvatures(
 
         # (minus) Shape operator, i.e. the differential of the Gauss map:
         # = (PPt^-1 @ PQt) : simple estimation through linear regression
-        S = torch.solve(PQt, PPt).solution
+        S = torch.linalg.solve(PQt, PPt)
         a, b, c, d = S[:, 0, 0], S[:, 0, 1], S[:, 1, 0], S[:, 1, 1]  # (N,)
-
         # Normalization
         mean_curvature = a + d
         gauss_curvature = a * d - b * c
-        features += [mean_curvature.clamp(-1, 1), gauss_curvature.clamp(-1, 1)]
+        features += [torch.nan_to_num(mean_curvature).clamp(-1, 1), torch.nan_to_num(gauss_curvature).clamp(-1, 1)]
 
     features = torch.stack(features, dim=-1)
     return features

main_training.py

@@ -15,6 +15,11 @@
 from helper import *
 from Arguments import parser
 
+import pykeops
+
+# Clean up the already compiled files
+pykeops.clean_pykeops()
+
 # Parse the arguments, prepare the TensorBoard writer:
 args = parser.parse_args()
 writer = SummaryWriter("runs/{}".format(args.experiment_name))
@@ -110,7 +115,7 @@
             dataloader = val_loader
         elif dataset_type == "Test":
             dataloader = test_loader
-
+        
         # Perform one pass through the data:
         info = iterate(
             net,

test.py

@@ -0,0 +1,52 @@
+# # Standard imports:
+# import numpy as np
+# import torch
+# from torch.utils.data import random_split
+# from torch_geometric.loader import DataLoader
+# from torch_geometric.transforms import Compose
+# from pathlib import Path
+
+# # Custom data loader and model:
+# from data import ProteinPairsSurfaces, PairData, CenterPairAtoms
+# from data import RandomRotationPairAtoms, NormalizeChemFeatures, iface_valid_filter
+# from helper import *
+# from Arguments import parser
+
+# # args
+# random_rotation = True
+# batch_size = 8
+# search = True
+# radius = 12.
+
+# # We load the train and test datasets.
+# # Random transforms, to ensure that no network/baseline overfits on pose parameters:
+# transformations = (
+#     Compose([NormalizeChemFeatures(), CenterPairAtoms(), RandomRotationPairAtoms()])
+#     if random_rotation
+#     else Compose([NormalizeChemFeatures()])
+# )
+
+# # PyTorch geometric expects an explicit list of "batched variables":
+# batch_vars = ["xyz_p1", "xyz_p2", "atom_coords_p1", "atom_coords_p2"]
+# # Load the train dataset:
+# train_dataset = ProteinPairsSurfaces(
+#     "surface_data", ppi=search, train=True, transform=transformations
+# )
+# # train_dataset = [data for data in train_dataset if iface_valid_filter(data)]
+# # train_loader = DataLoader(
+# #     train_dataset, batch_size=1, follow_batch=batch_vars, shuffle=True
+# # )
+# print("Preprocessing training dataset")
+
+# Testing PyKeops installation
+import pykeops
+
+# Changing verbose and mode
+pykeops.verbose = True
+pykeops.build_type = 'Debug'
+
+# Clean up the already compiled files
+pykeops.clean_pykeops()
+
+# Test Numpy integration
+pykeops.test_numpy_bindings()

test.sh

@@ -0,0 +1,5 @@
+#! /usr/bin/env
+
+module load gcc/9.2.0 cuda/11.7
+
+python -W ignore -u main_training.py --experiment_name dMaSIF_search_3layer_12A --batch_size 64 --embedding_layer dMaSIF --search True --device cuda:0 --random_rotation True --radius 12.0 --n_layers 3 --seed 0