create ipynb rst

docs/.ipynb_checkpoints/community-checkpoint.rst

@@ -0,0 +1,12 @@
+.. _examples:
+
+Examples
+--------
+
+.. toctree::
+   :maxdepth: 1
+
+   Community/Community_1_physics_informed_kan.rst
+   Community/Community_2_protein_sequence_classification.rst
+
+

docs/.ipynb_checkpoints/demos-checkpoint.rst

@@ -1,3 +1,5 @@
+.. _api-demo:
+
 API Demos
 ---------
 
@@ -6,11 +8,13 @@ API Demos
 
    API_demo/API_1_indexing.rst
    API_demo/API_2_plotting.rst
-   API_demo/API_3_grid.rst
-   API_demo/API_4_extract_activations.rst
-   API_demo/API_5_initialization_hyperparameter.rst
+   API_demo/API_3_extract_activations.rst
+   API_demo/API_4_initialization.rst
+   API_demo/API_5_grid.rst
    API_demo/API_6_training_hyperparameter.rst
    API_demo/API_7_pruning.rst
-   API_demo/API_8_checkpoint.rst
+   API_demo/API_8_regularization.rst
    API_demo/API_9_video.rst
-   API_demo/API_10_device.rst
+   API_demo/API_10_device.rst
+   API_demo/API_11_create_dataset.rst
+   API_demo/API_12_checkpoint_save_load_model.rst

docs/.ipynb_checkpoints/examples-checkpoint.rst

@@ -0,0 +1,24 @@
+.. _examples:
+
+Examples
+--------
+
+.. toctree::
+   :maxdepth: 1
+
+   Example/Example_1_function_fitting.rst
+   Example/Example_3_classfication.rst
+   Example/Example_4_classfication.rst
+   Example/Example_5_special_functions.rst
+   Example/Example_6_PDE_interpretation.rst
+   Example/Example_7_PDE_accuracy.rst
+   Example/Example_8_continual_learning.rst
+   Example/Example_9_singularity.rst
+   Example/Example_10_relativity-addition.rst
+   Example/Example_11_encouraing_linear.rst
+   Example/Example_12_unsupervised_learning.rst
+   Example/Example_13_phase_transition.rst
+   Example/Example_14_knot_supervised.rst
+   Example/Example_15_knot_unsupervised.rst
+
+

docs/.ipynb_checkpoints/interp-checkpoint.rst

@@ -0,0 +1,23 @@
+.. _examples:
+
+Examples
+--------
+
+.. toctree::
+   :maxdepth: 1
+
+   Interp/Interp_1_Hello, MultKAN.rst
+   Interp/Interp_2_Advanced MultKAN.rst
+   Interp/Interp_3_KAN_Compiler.rst
+   Interp/Interp_4_feature_attribution.rst
+   Interp/Interp_5_test_symmetry.rst
+   Interp/Interp_6_test_symmetry_NN.rst
+   Interp/Interp_8_adding_auxillary_variables.rst
+   Interp/Interp_9_different_plotting_metrics.rst
+   Interp/Interp_10_hessian.rst
+   Interp/Interp_10A_swap.rst
+   Interp/Interp_10B_swap.rst
+   Interp/Interp_11_sparse_init.rst
+
+
+

docs/.ipynb_checkpoints/physics-checkpoint.rst

@@ -0,0 +1,18 @@
+.. _examples:
+
+Examples
+--------
+
+.. toctree::
+   :maxdepth: 1
+
+   Physics/Physics_1_Lagrangian.rst
+   Physics/Physics_2A_conservation_law.rst
+   Physics/Physics_2B_conservation_law_2D.rst
+   Physics/Physics_3_blackhole.rst
+   Physics/Physics_4A_constitutive_laws_P11.rst
+   Physics/Physics_4B_constitutive_laws_P12_with_prior.rst
+   Physics/Physics_4C_constitutive_laws_P12_without_prior.rst
+
+
+

docs/API_demo/.ipynb_checkpoints/API_6_training_hyperparameter-checkpoint.rst

@@ -0,0 +1,231 @@
+API 6: Training Hyperparamters
+==============================
+
+Regularization helps interpretability by making KANs sparser. This may
+require some hyperparamter tuning. Let’s see how hyperparameters can
+affect training
+
+Load KAN and create_dataset
+
+.. code:: ipython3
+
+    from kan import *
+    import torch
+    
+    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+    print(device)
+    
+    f = lambda x: torch.exp(torch.sin(torch.pi*x[:,[0]]) + x[:,[1]]**2)
+    dataset = create_dataset(f, n_var=2, device=device)
+    dataset['train_input'].shape, dataset['train_label'].shape
+
+
+.. parsed-literal::
+
+    cuda
+
+
+
+
+.. parsed-literal::
+
+    (torch.Size([1000, 2]), torch.Size([1000, 1]))
+
+
+
+Default setup
+
+.. code:: ipython3
+
+    # train the model
+    model = KAN(width=[2,5,1], grid=5, k=3, seed=1, device=device)
+    model.fit(dataset, opt="LBFGS", steps=20, lamb=0.01);
+    model.plot()
+
+
+.. parsed-literal::
+
+    checkpoint directory created: ./model
+    saving model version 0.0
+
+
+.. parsed-literal::
+
+    | train_loss: 3.34e-02 | test_loss: 3.29e-02 | reg: 4.93e+00 | : 100%|█| 20/20 [00:05<00:00,  3.73it
+
+
+.. parsed-literal::
+
+    saving model version 0.1
+
+
+
+.. image:: API_6_training_hyperparameter_files/API_6_training_hyperparameter_4_3.png
+
+
+Parameter 1: :math:`\lambda`, overall penalty strength.
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Previously :math:`\lambda=0.01`, now we try different :math:`\lambda`.
+
+:math:`\lambda=0`
+
+.. code:: ipython3
+
+    # train the model
+    model = KAN(width=[2,5,1], grid=5, k=3, seed=1, device=device)
+    model.fit(dataset, opt="LBFGS", steps=20, lamb=0.00);
+    model.plot()
+
+
+.. parsed-literal::
+
+    checkpoint directory created: ./model
+    saving model version 0.0
+
+
+.. parsed-literal::
+
+    | train_loss: 5.51e-03 | test_loss: 6.14e-03 | reg: 1.52e+01 | : 100%|█| 20/20 [00:03<00:00,  5.84it
+
+
+.. parsed-literal::
+
+    saving model version 0.1
+
+
+
+.. image:: API_6_training_hyperparameter_files/API_6_training_hyperparameter_7_3.png
+
+
+:math:`\lambda=1`
+
+.. code:: ipython3
+
+    # train the model
+    model = KAN(width=[2,5,1], grid=5, k=3, seed=0, device=device)
+    model.fit(dataset, opt="LBFGS", steps=20, lamb=1.0);
+    model.plot()
+
+
+.. parsed-literal::
+
+    checkpoint directory created: ./model
+    saving model version 0.0
+
+
+.. parsed-literal::
+
+    | train_loss: 1.70e+00 | test_loss: 1.73e+00 | reg: 1.08e+01 | : 100%|█| 20/20 [00:04<00:00,  4.59it
+
+
+.. parsed-literal::
+
+    saving model version 0.1
+
+
+
+.. image:: API_6_training_hyperparameter_files/API_6_training_hyperparameter_9_3.png
+
+
+Parameter 2: (relative) penalty strength of entropy :math:`\lambda_{\rm ent}`.
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The absolute magnitude is :math:`\lambda\lambda_{\rm ent}`. Previously
+we set :math:`\lambda=0.1` and :math:`\lambda_{\rm ent}=2.0` (default).
+Below we fix :math:`\lambda=0.1` and vary :math:`\lambda_{\rm ent}`.
+
+:math:`\lambda_{\rm ent}=0.0`
+
+.. code:: ipython3
+
+    # train the model
+    model = KAN(width=[2,5,1], grid=5, k=3, seed=1, device=device)
+    model.fit(dataset, opt="LBFGS", steps=20, lamb=0.01, lamb_entropy=0.0);
+    model.plot()
+
+
+.. parsed-literal::
+
+    checkpoint directory created: ./model
+    saving model version 0.0
+
+
+.. parsed-literal::
+
+    | train_loss: 4.20e-02 | test_loss: 4.50e-02 | reg: 2.57e+00 | : 100%|█| 20/20 [00:04<00:00,  4.68it
+
+
+.. parsed-literal::
+
+    saving model version 0.1
+
+
+
+.. image:: API_6_training_hyperparameter_files/API_6_training_hyperparameter_12_3.png
+
+
+:math:`\lambda_{\rm ent}=10.`
+
+.. code:: ipython3
+
+    # train the model
+    model = KAN(width=[2,5,1], grid=5, k=3, seed=1, device=device)
+    model.fit(dataset, opt="LBFGS", steps=20, lamb=0.01, lamb_entropy=10.0);
+    model.plot()
+
+
+.. parsed-literal::
+
+    checkpoint directory created: ./model
+    saving model version 0.0
+
+
+.. parsed-literal::
+
+    | train_loss: 7.83e-02 | test_loss: 7.74e-02 | reg: 1.54e+01 | : 100%|█| 20/20 [00:05<00:00,  3.77it
+
+
+.. parsed-literal::
+
+    saving model version 0.1
+
+
+
+.. image:: API_6_training_hyperparameter_files/API_6_training_hyperparameter_14_3.png
+
+
+Parameter 3: seed.
+~~~~~~~~~~~~~~~~~~
+
+Previously we use seed = 1. Below we vary seed.
+
+:math:`{\rm seed} = 42`
+
+.. code:: ipython3
+
+    model = KAN(width=[2,5,1], grid=3, k=3, seed=42, device=device)
+    model.fit(dataset, opt="LBFGS", steps=20, lamb=0.01);
+    model.plot()
+
+
+.. parsed-literal::
+
+    checkpoint directory created: ./model
+    saving model version 0.0
+
+
+.. parsed-literal::
+
+    | train_loss: 5.67e-02 | test_loss: 5.72e-02 | reg: 5.81e+00 | : 100%|█| 20/20 [00:04<00:00,  4.81it
+
+
+.. parsed-literal::
+
+    saving model version 0.1
+
+
+
+.. image:: API_6_training_hyperparameter_files/API_6_training_hyperparameter_17_3.png
+
+