Merge pull request #66 from LIHPC-Computational-Geometry/65-only-topological-constraints

New process to constrain actions

Author: ArzhelaR

.github/workflows/model-view-ci.yml

@@ -23,7 +23,7 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
-        pip install flake8 pytest numpy coverage pygame matplotlib
+        pip install flake8 pytest numpy coverage pygame matplotlib scipy shapely
         if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
     - name: Lint with flake8
       run: |

README.md

@@ -34,25 +34,27 @@ For training on triangular meshes, you can use an agent with all three actions:
 ##### 1. Using `tune/model_RL/PPO_model`
 
 - Configure the model and environment parameters in:  
-  `tune/training/train.py`
+  `tune/training/config/trimesh_config_PPO_perso.yaml`
 
 - Then run the following command from the `tune/` directory:
   ```bash
-  python main.py
+  python -m training.train_trimesh.py
   ```
-
-
+  
 ##### 2. Using PPO from Stable Baselines 3 (SB3)
 
 - Configure the model and environment parameters in:  
-    - `tune/environment/environment_config.json`
-    - `tune/model_RL/parameters/PPO_config.json`
+  `tune/training/config/trimesh_config_PPO_SB3.yaml`
 
-- Then run the training script in pycharm `tune/training/train_trimesh_SB3.py`
+- Then run the following command from the `tune/` directory:
+  ```bash
+  python -m training.train_trimesh_SB3.py
+  ```
 
 ###### Flip-Only Training (SB3 PPO)
 
 To train an agent using only the flip action with SB3 PPO, run the training script in pycharm `tune/training/train_trimesh_flip_SB3.py`
+> ❗ This environment may be deprecated.
 
 ---
 
@@ -65,21 +67,26 @@ For training on quadrangular meshes, you can use an agent with all four actions:
 
 #### 🚀 Starting Training
 
-##### 1. Configure the model and environment parameters in :
-  - `tune/environment/environment_config.json`
-  - `tune/model_RL/parameters/PPO_config.json`
-  
-##### 2. Using `tune/model_RL/PPO_model_pers`
-Run the following command from the `tune/` directory:
+##### 1. Using `tune/model_RL/PPO_model`
+
+- Configure the model and environment parameters in:  
+  `tune/training/config/quadmesh_config_PPO_perso.yaml`
+
+- Then run the following command from the `tune/` directory:
   ```bash
-  python -m training.train_quadmesh
+  python -m training.train_quadmesh.py
   ```
+  
+##### 2. Using PPO from Stable Baselines 3 (SB3)
 
-##### 3. Using PPO from Stable Baselines 3 (SB3)
-Run the following command from the `tune/` directory:
+- Configure the model and environment parameters in:  
+  `tune/training/config/quadmesh_config_PPO_SB3.yaml`
+
+- Then run the following command from the `tune/` directory:
   ```bash
-  python -m training.train_quadmesh_SB3
+  python -m training.train_quadmesh_SB3.py
   ```
+  
 #### 🧪 Testing a Saved SB3 Policy
 
 After training, the model is saved as a `.zip` file in the `tune/training/policy_saved/` directory. To evaluate the policy, follow these steps in `tune/training/exploit_SB3_policy.py` :

docs/actions.md

@@ -18,7 +18,7 @@ However, in some configurations, flip action can lead to problematic configurati
   <img src="img/actions/flip_c_after.png" width="350"/>
 
     * **triangular quad:** This configuration should be avoided, as it results in the creation of a degenerate (flat) triangle. As illustrated in the figures below, flipping dart 0 leads to a flattened face between nodes 2, 0, and 7.
-    * 
+
   <img src="img/actions/flip_before_tri.png" width="350"/>
   <img src="img/actions/flip_after_tri.png" width="350"/>
 
@@ -51,6 +51,54 @@ However, in some configurations, collapse action can also lead to problematic co
 * **Adjacency too high :** When nodes A already has an adjacency higher than 10, collapse is not possible. 
 * **Configuration who tends to imply edge reversal**: To detect these situations, we look a the type of darts in the surrounding. When there are some darts with concave surrounding, collapse action can lead to edge reversal.
 
-
 <img src="img/actions/collapse_reversal_before.png" width="400"/>
-<img src="img/actions/collapse_reversal_after.png" width="360"/>
+<img src="img/actions/collapse_reversal_after.png" width="360"/>
+
+---
+
+# Quadrangular Actions
+
+## Flip 
+
+<img src="img/actions/flip_quad.png" width="600"/>
+
+The **flip** action can be decomposed as follows:
+
+* Check whether the dart `d` exists.  
+* Update the **beta1** relations.  
+* Update the links from nodes `n1` and `n2`,  if they were previously linked to the flipped dart.  
+* Update the links from faces `F1` and `F2`, if they were previously linked to the flipped dart.  
+* Update the node links for `d` and `d2`.  
+* Update the face links depending on the orientation (darts which change of face):
+  * If **counterclockwise**: update links for `d1` and `d21`.  
+  * If **clockwise**: update links for `d2111` and `d111`.  
+* Update the scores of the nodes.
+
+## Split 
+
+<img src="img/actions/split_quad.png" width="600"/>
+
+The **split** action can be decomposed as follows:
+
+* Check whether the dart `d` exists.  
+* Create a new node **n10** in the middle of [n1,n2]
+* Update node links of `d` and `d21`, they are now linked to `n10`
+* Create a new face with yellow darts
+* Update the **beta2** relations (`d1112`, `d111`,`d212`, `d21`).
+* Update the scores of the nodes.
+
+## Collapse 
+
+<img src="img/actions/collapse_quad.png" width="600"/>
+
+The **collapse** action can be decomposed as follows:
+
+* Check whether the dart `d` exists.  
+* Save the score of node `n1` for later computation.  
+* If `n3` is not on the boundary, move its coordinates to the midpoint between `n3` and `n1`.  
+* Check if nodes `n2` and `n4` are linked to the dart being deleted. If so, reassign their links to existing darts (node `n3` will be checked later).  
+* Delete the face associated with `d`, along with its related darts.  
+* Reassign all darts previously linked to `n1` so they now point to `n3`.  
+* Link the dart associated with `n3` to an existing node.  
+* Update the **beta2** relations (`d2`/`d12`, `d112`/`d1112`).  
+* Recalculate the scores of the affected nodes.

docs/img/actions/collapse_quad.png

@@ -0,0 +1,38 @@
+# Actions tests
+
+## Triangular actions
+
+### Flip
+
+We want to ensure that **flip** operations are performed correctly.
+Consider the following mesh:
+
+<img src="img/tri_flip_before.png" width="300"/>
+
+If we flip the dart between nodes **n0** and **n3**, we should obtain the following mesh:
+
+<img src="img/tri_flip_after.png" width="300"/>
+
+### Split
+
+We want to ensure that **split** operations are performed correctly.
+Consider the following mesh:
+
+<img src="img/tri_split_before.png" width="300"/>
+
+If we split the dart between nodes **n0** and **n3**, we add a node **n5** on the middle of the edge and two faces are created. 
+We should obtain the following mesh:
+
+<img src="img/tri_split_after.png" width="300"/>
+
+### Collapse
+
+We want to ensure that **split** operations are performed correctly.
+Consider the following mesh:
+
+<img src="img/tri_collapse_before.png" width="300"/>
+
+Here we can't collapse the dart between nodes **n5** and **n2** because n2 is on boundary. However we can collapse the edge between **n4** and **n5**.
+We should obtain the following mesh:
+
+<img src="img/tri_collapse_after.png" width="300"/>