Upload folder using huggingface_hub

.vscode/settings.json

@@ -0,0 +1,5 @@
+{
+    "python-envs.defaultEnvManager": "ms-python.python:conda",
+    "python-envs.defaultPackageManager": "ms-python.python:conda",
+    "python-envs.pythonProjects": []
+}

README.md

@@ -1,3 +1,41 @@
----
-license: mit
----
+---
+license: mit
+datasets:
+- PLAID-datasets/Tensile2d
+language:
+- en
+pipeline_tag: graph-ml
+---
+# PCA-GP model for VKILS59 dataset
+
+The code used to train this model is given in `train.py`.
+
+## Install
+
+```bash
+conda env create -n mmgp_tensile2d -f https://huggingface.co/fabiencasenave/mmgp_tensile2d/resolve/main/environment.yml
+conda mmgp_tensile2d
+pip install git+https://huggingface.co/fabiencasenave/mmgp_tensile2d
+```
+
+## Use
+
+```python
+from datasets import load_dataset
+from plaid.bridges.huggingface_bridge import huggingface_dataset_to_plaid
+import mmgp_tensile2d
+
+model = mmgp_tensile2d.load()
+
+hf_dataset = load_dataset("PLAID-datasets/Tensile2d", split="all_samples")
+ids_test = hf_dataset.description["split"]['test']
+
+dataset_test, _ = huggingface_dataset_to_plaid(hf_dataset, ids = ids_test, processes_number = 6, verbose = True)
+
+print("Check the 'U1' field is not present: dataset_test[0].get_field('U1') =", dataset_test[0].get_field('U1'))
+
+print("Run prediction...")
+dataset_pred = model.predict(dataset_test)
+
+print("Check the 'U1' field is not present: dataset_pred[0].get_field('U1') =", dataset_pred[0].get_field('U1'))
+```

config_pipeline.yml

@@ -0,0 +1,59 @@
+mmgp_nodes_transf:
+  in_features_identifiers:
+    - type: nodes
+
+mmgp_u1_transf:
+  in_features_identifiers:
+    - type: field
+      name: U1
+
+
+input_scalar_scaler:
+  in_features_identifiers:
+    - type: scalar
+      name: P
+    - type: scalar
+      name: p1
+    - type: scalar
+      name: p2
+    - type: scalar
+      name: p3
+    - type: scalar
+      name: p4
+    - type: scalar
+      name: p5
+
+pca_nodes:
+  in_features_identifiers:
+    - type: nodes
+  out_features_identifiers:
+    - type: scalar
+      name: reduced_nodes_*
+
+pca_u1:
+  in_features_identifiers:
+    - type: field
+      name: U1
+  out_features_identifiers:
+    - type: scalar
+      name: reduced_U1_*
+
+regressor_mach:
+  in_features_identifiers:
+    - type: scalar
+      name: P
+    - type: scalar
+      name: p1
+    - type: scalar
+      name: p2
+    - type: scalar
+      name: p3
+    - type: scalar
+      name: p4
+    - type: scalar
+      name: p5
+    - type: scalar
+      name: reduced_nodes_*
+  out_features_identifiers:
+    - type: scalar
+      name: reduced_U1_*

environment.yml

@@ -0,0 +1,11 @@
+name: mmgp_tensile2d
+channels:
+  - conda-forge
+  - nodefaults
+dependencies:
+  - python=3.11
+  - muscat=2.4.1
+  - pip
+  - pip:
+    - git+https://github.com/PLAID-lib/plaid.git@11f5912a2ade5dea8e708041d61dc7ab4a32baf4
+    - git+https://gitlab.com/drti/mmgp.git

pipeline.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9950a0b4b472fd87721f54fe271aec7da67cec3afef8045f01e319a6d24cbe3
+size 1996169

pyproject.toml

@@ -0,0 +1,23 @@
+[build-system]
+requires = [
+    "setuptools>=60",
+    "setuptools-scm>=8.0"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "mmgp_tensile2d"
+version = "0.1.0"
+description = "MMGP model for Tensile2d dataset"
+authors = [
+    { name = "Fabien Casenave", email = "[email protected]" }
+]
+readme = "README.md"
+requires-python = ">=3.9"
+
+[project.urls]
+Homepage = "https://huggingface.co/fabiencasenave/mmgp_tensile2d"
+Repository = "https://huggingface.co/fabiencasenave/mmgp_tensile2d"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+include = ["mmgp_tensile2d*"]

src/mmgp_tensile2d.egg-info/PKG-INFO

@@ -0,0 +1,49 @@
+Metadata-Version: 2.4
+Name: mmgp_tensile2d
+Version: 0.1.0
+Summary: MMGP model for Tensile2d dataset
+Author-email: Fabien Casenave <[email protected]>
+Project-URL: Homepage, https://huggingface.co/fabiencasenave/mmgp_tensile2d
+Project-URL: Repository, https://huggingface.co/fabiencasenave/mmgp_tensile2d
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+
+---
+license: mit
+datasets:
+- PLAID-datasets/VKI-LS59
+language:
+- en
+pipeline_tag: graph-ml
+---
+# PCA-GP model for VKILS59 dataset
+
+The code used to train this model is given in `train.py`.
+
+## Install
+
+```bash
+pip install git+https://huggingface.co/fabiencasenave/pca_gp_vkils59
+```
+
+## Use
+
+```python
+from datasets import load_dataset
+from plaid.bridges.huggingface_bridge import huggingface_dataset_to_plaid
+import pca_gp_vkils59
+
+model = pca_gp_vkils59.load()
+
+hf_dataset = load_dataset("PLAID-datasets/VKI-LS59", split="all_samples")
+ids_test = hf_dataset.description["split"]['test']
+
+dataset_test, _ = huggingface_dataset_to_plaid(hf_dataset, ids = ids_test, processes_number = 6, verbose = True)
+
+print("Check the 'mach' field is not present: dataset_test[0].get_field('mach', base_name='Base_2_2') =", dataset_test[0].get_field('mach', base_name='Base_2_2'))
+
+print("Run prediction...")
+dataset_pred = model.predict(dataset_test)
+
+print("Check the 'mach' field is now present: dataset_pred[0].get_field('mach', base_name='Base_2_2') =", dataset_pred[0].get_field('mach', base_name='Base_2_2'))
+```

src/mmgp_tensile2d.egg-info/SOURCES.txt

@@ -0,0 +1,8 @@
+README.md
+pyproject.toml
+src/mmgp_tensile2d/__init__.py
+src/mmgp_tensile2d/utils.py
+src/mmgp_tensile2d.egg-info/PKG-INFO
+src/mmgp_tensile2d.egg-info/SOURCES.txt
+src/mmgp_tensile2d.egg-info/dependency_links.txt
+src/mmgp_tensile2d.egg-info/top_level.txt

src/mmgp_tensile2d.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

src/mmgp_tensile2d.egg-info/top_level.txt

@@ -0,0 +1 @@
+mmgp_tensile2d

src/mmgp_tensile2d/__init__.py

@@ -0,0 +1 @@
+from .utils import load

src/mmgp_tensile2d/utils.py

@@ -0,0 +1,21 @@
+
+import numpy as np
+import joblib
+from huggingface_hub import hf_hub_download
+from Muscat.Containers.MeshGraphTools import FloaterMeshParametrization, RenumberMeshForParametrization
+
+def length_scale_init(X):
+    return np.ones(X.shape[1])
+
+
+def load():
+    path = hf_hub_download(repo_id="fabiencasenave/mmgp_tensile2d", filename="pipeline.joblib")
+    return joblib.load(path)
+
+def morphing(mesh):
+    mesh_renumb, _, n_boundary = RenumberMeshForParametrization(
+        mesh, inPlace=False)
+    mesh_renumb.elemFields = mesh_renumb.nodeFields = {}
+    morphed_mesh, _ = FloaterMeshParametrization(
+    mesh_renumb, n_boundary)
+    return morphed_mesh

test.py

@@ -0,0 +1,17 @@
+from datasets import load_dataset
+from plaid.bridges.huggingface_bridge import huggingface_dataset_to_plaid
+import mmgp_tensile2d
+
+model = mmgp_tensile2d.load()
+
+hf_dataset = load_dataset("PLAID-datasets/Tensile2d", split="all_samples")
+ids_test = hf_dataset.description["split"]['test']
+
+dataset_test, _ = huggingface_dataset_to_plaid(hf_dataset, ids = ids_test, processes_number = 6, verbose = True)
+
+print("Check the 'U1' field is not present: dataset_test[0].get_field('U1') =", dataset_test[0].get_field('U1'))
+
+print("Run prediction...")
+dataset_pred = model.predict(dataset_test)
+
+print("Check the 'U1' field is not present: dataset_pred[0].get_field('U1') =", dataset_pred[0].get_field('U1'))

train.py

@@ -0,0 +1,128 @@
+import os
+os.environ["OMP_PROC_BIND"] = "spread"
+os.environ["OMP_PLACES"]    = "threads"
+
+from pathlib import Path
+
+import yaml
+import numpy as np
+import joblib
+
+from datasets.utils.logging import disable_progress_bar
+from datasets import load_dataset
+
+from sklearn.base import clone
+from sklearn.pipeline import Pipeline
+
+from sklearn.decomposition import PCA
+from sklearn.preprocessing import MinMaxScaler
+from sklearn.gaussian_process import GaussianProcessRegressor
+from sklearn.gaussian_process.kernels import Matern
+from sklearn.multioutput import MultiOutputRegressor
+
+from sklearn.model_selection import KFold, GridSearchCV
+
+from plaid.bridges.huggingface_bridge import huggingface_dataset_to_plaid, huggingface_description_to_problem_definition
+from plaid.pipelines.sklearn_block_wrappers import WrappedPlaidSklearnTransformer, WrappedPlaidSklearnRegressor
+from plaid.pipelines.plaid_blocks import PlaidTransformedTargetRegressor, PlaidColumnTransformer
+from mmgp.pipelines.mmgp_blocks import MMGPPreparer, MMGPTransformer
+
+from mmgp_tensile2d.utils import length_scale_init, morphing
+
+
+n_processes = min(max(1, os.cpu_count()), 8)
+
+
+# load dataset
+hf_dataset = load_dataset("PLAID-datasets/Tensile2d", split="all_samples")
+ids_train = hf_dataset.description["split"]['train_500'][:8]
+
+dataset_train, _ = huggingface_dataset_to_plaid(hf_dataset, ids = ids_train, processes_number = n_processes, verbose = True)
+
+
+# load pipeline configuration and trim dataset
+with open("config_pipeline.yml", 'r') as f:
+    config = yaml.safe_load(f)
+
+all_feature_id = config['input_scalar_scaler']['in_features_identifiers'] +\
+    config['pca_nodes']['in_features_identifiers'] + config['pca_u1']['in_features_identifiers']
+
+dataset_train = dataset_train.from_features_identifier(all_feature_id)
+
+
+# define learning pipeline
+preparator = MMGPPreparer(common_mesh_id = 1, morphing = morphing)
+
+input_scalar_scaler = WrappedPlaidSklearnTransformer(MinMaxScaler(), **config['input_scalar_scaler'])
+
+nodes_preprocessor = Pipeline(
+    steps=[
+        ("mmgp_nodes_transf", MMGPTransformer(**config['mmgp_nodes_transf'])),
+        ('pca_nodes', WrappedPlaidSklearnTransformer(PCA(n_components=4), **config['pca_nodes'])),
+    ]
+)
+
+column_preprocessor = PlaidColumnTransformer(
+                [
+                    ('input_scalar_scaler', input_scalar_scaler),
+                    ('nodes_preprocessor', nodes_preprocessor),
+                ]
+            )
+
+preprocessor = Pipeline(
+    steps=[
+        ("preparator", preparator),
+        ('column_preprocessor', column_preprocessor),
+    ]
+)
+
+
+kernel = Matern(length_scale_bounds=(1e-8, 1e8), nu = 2.5)
+
+gpr = GaussianProcessRegressor(
+    kernel=kernel,
+    optimizer='fmin_l_bfgs_b',
+    n_restarts_optimizer=1,
+    random_state=42)
+
+reg = MultiOutputRegressor(gpr)
+
+
+dynamics_params_factory = {'estimator__kernel__length_scale':length_scale_init}
+
+regressor = WrappedPlaidSklearnRegressor(reg, **config['regressor_mach'], dynamics_params_factory = dynamics_params_factory)
+
+postprocessor = Pipeline(
+    steps=[
+        ("mmgp_u1_transf", MMGPTransformer(**config['mmgp_u1_transf'])),
+        ('pca_u1', WrappedPlaidSklearnTransformer(PCA(n_components=4), **config['pca_u1'])),
+    ]
+)
+
+
+target_regressor = PlaidTransformedTargetRegressor(
+    regressor=regressor,
+    transformer=postprocessor,
+    # out_features_identifiers = config['pca_u1']['in_features_identifiers']
+)
+
+pipeline = Pipeline(
+    steps=[
+        ("preprocessor", preprocessor),
+        ("regressor", target_regressor),
+    ]
+)
+
+
+
+# Set hyperameter that have been optimized by cross-valdiation on the training set
+optimized_pipeline = clone(pipeline).set_params(
+    preprocessor__column_preprocessor__nodes_preprocessor__pca_nodes__sklearn_block__n_components = 4,
+    regressor__transformer__pca_u1__sklearn_block__n_components = 8
+)
+
+# Train the model
+optimized_pipeline.fit(dataset_train)
+
+# Save model
+joblib.dump(optimized_pipeline, "pipeline.joblib")