Add files via upload

docs/Workshop/KANs Can't Deal with Noise

@@ -0,0 +1 @@
+{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"provenance":[],"authorship_tag":"ABX9TyN4NaBDH4WaPfma4iJvrPsq"},"kernelspec":{"name":"python3","display_name":"Python 3"},"language_info":{"name":"python"}},"cells":[{"cell_type":"markdown","source":["# KANs Example Revisited\n","\n","A\n","\n","Source: https://github.com/KindXiaoming/pykan/blob/master/tutorials/Example_2_deep_formula.ipynb"],"metadata":{"id":"Oa_wQm4BTSKs"}},{"cell_type":"code","execution_count":1,"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"UaC0bshQ7JoI","executionInfo":{"status":"ok","timestamp":1715287638561,"user_tz":240,"elapsed":17886,"user":{"displayName":"Aijun Zhang","userId":"04897292400908543781"}},"outputId":"c1550e97-a31d-4114-de3b-0755e4459707"},"outputs":[{"output_type":"stream","name":"stdout","text":["Collecting pykan\n","  Downloading pykan-0.0.5-py3-none-any.whl (33 kB)\n","Installing collected packages: pykan\n","Successfully installed pykan-0.0.5\n"]}],"source":["pip install pykan"]},{"cell_type":"code","source":["def create_dataset2(f,\n","                   n_var=2,\n","                   ranges = [-1,1],\n","                   train_num=1000,\n","                   test_num=1000,\n","                   normalize_input=True,\n","                   normalize_label=True,\n","                   add_noise=True,  # add noise\n","                   noise_level=0.1,\n","                   device='cpu',\n","                   seed=0):\n","    '''\n","    create dataset\n","\n","    Args:\n","    -----\n","        f : function\n","            the symbolic formula used to create the synthetic dataset\n","        ranges : list or np.array; shape (2,) or (n_var, 2)\n","            the range of input variables. Default: [-1,1].\n","        train_num : int\n","            the number of training samples. Default: 1000.\n","        test_num : int\n","            the number of test samples. Default: 1000.\n","        normalize_input : bool\n","            If True, apply normalization to inputs. Default: False.\n","        normalize_label : bool\n","            If True, apply normalization to labels. Default: False.\n","        device : str\n","            device. Default: 'cpu'.\n","        seed : int\n","            random seed. Default: 0.\n","\n","    Returns:\n","    --------\n","        dataset : dic\n","            Train/test inputs/labels are dataset['train_input'], dataset['train_label'],\n","                        dataset['test_input'], dataset['test_label']\n","\n","    Example\n","    -------\n","    >>> f = lambda x: torch.exp(torch.sin(torch.pi*x[:,[0]]) + x[:,[1]]**2)\n","    >>> dataset = create_dataset(f, n_var=2, train_num=100)\n","    >>> dataset['train_input'].shape\n","    torch.Size([100, 2])\n","    '''\n","\n","    np.random.seed(seed)\n","    torch.manual_seed(seed)\n","\n","    if len(np.array(ranges).shape) == 1:\n","        ranges = np.array(ranges * n_var).reshape(n_var,2)\n","    else:\n","        ranges = np.array(ranges)\n","\n","    train_input = torch.zeros(train_num, n_var)\n","    test_input = torch.zeros(test_num, n_var)\n","    for i in range(n_var):\n","        train_input[:,i] = torch.rand(train_num,)*(ranges[i,1]-ranges[i,0])+ranges[i,0]\n","        test_input[:,i] = torch.rand(test_num,)*(ranges[i,1]-ranges[i,0])+ranges[i,0]\n","\n","\n","    train_label = f(train_input)\n","    test_label = f(test_input)\n","\n","\n","    def normalize(data, mean, std):\n","            return (data-mean)/std\n","\n","    if normalize_input == True:\n","        mean_input = torch.mean(train_input, dim=0, keepdim=True)\n","        std_input = torch.std(train_input, dim=0, keepdim=True)\n","        train_input = normalize(train_input, mean_input, std_input)\n","        test_input = normalize(test_input, mean_input, std_input)\n","\n","    if normalize_label == True:\n","        mean_label = torch.mean(train_label, dim=0, keepdim=True)\n","        std_label = torch.std(train_label, dim=0, keepdim=True)\n","        train_label = normalize(train_label, mean_label, std_label)\n","        test_label = normalize(test_label, mean_label, std_label)\n","\n","    if add_noise:\n","        noise = torch.randn_like(test_label) * noise_level\n","        test_label += noise\n","\n","    dataset = {}\n","    dataset['train_input'] = train_input.to(device)\n","    dataset['test_input'] = test_input.to(device)\n","\n","    dataset['train_label'] = train_label.to(device)\n","    dataset['test_label'] = test_label.to(device)\n","\n","    return dataset\n"],"metadata":{"id":"HqpTRwDOHDf8","executionInfo":{"status":"ok","timestamp":1715290672231,"user_tz":240,"elapsed":183,"user":{"displayName":"Aijun Zhang","userId":"04897292400908543781"}}},"execution_count":29,"outputs":[]},{"cell_type":"code","source":["from kan import KAN, create_dataset\n","import torch\n","import numpy as np\n","\n","\n","# create a KAN: 2D inputs, 1D output, and 5 hidden neurons. cubic spline (k=3), 5 grid intervals (grid=5).\n","model = KAN(width=[4,2,1,1], grid=3, k=3, seed=0)\n","f = lambda x: torch.exp((torch.sin(torch.pi*(x[:,[0]]**2+x[:,[1]]**2))+torch.sin(torch.pi*(x[:,[2]]**2+x[:,[3]]**2)))/2)\n","dataset = create_dataset2(f, n_var=4, train_num=1000, test_num=1000, noise_level=0.1)\n","\n","image_folder = 'video_img'\n","\n","# train the model\n","#model.train(dataset, opt=\"LBFGS\", steps=20, lamb=1e-3, lamb_entropy=2.);\n","model.train(dataset, opt=\"LBFGS\", steps=50, lamb=5e-5, lamb_entropy=2., save_fig=True, beta=10,\n","            in_vars=[r'$x_1$', r'$x_2$', r'$x_3$', r'$x_4$'],\n","            out_vars=[r'${\\rm exp}({\\rm sin}(x_1^2+x_2^2)+{\\rm sin}(x_3^2+x_4^2))$'],\n","            img_folder=image_folder);"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"L6WaqV6c9Hx_","executionInfo":{"status":"ok","timestamp":1715290869500,"user_tz":240,"elapsed":193267,"user":{"displayName":"Aijun Zhang","userId":"04897292400908543781"}},"outputId":"ad364a51-b9fc-4022-9419-fc65834de254"},"execution_count":30,"outputs":[{"output_type":"stream","name":"stderr","text":["train loss: 1.09e-02 | test loss: 1.01e-01 | reg: 1.03e+01 : 100%|██| 50/50 [03:12<00:00,  3.86s/it]\n"]}]},{"cell_type":"code","source":["import os\n","import numpy as np\n","import moviepy.video.io.ImageSequenceClip # moviepy == 1.0.3\n","\n","video_name='video'\n","fps=5\n","\n","fps = fps\n","files = os.listdir(image_folder)\n","train_index = []\n","for file in files:\n","    if file[0].isdigit() and file.endswith('.jpg'):\n","        train_index.append(int(file[:-4]))\n","\n","train_index = np.sort(train_index)\n","\n","image_files = [image_folder+'/'+str(train_index[index])+'.jpg' for index in train_index]\n","\n","clip = moviepy.video.io.ImageSequenceClip.ImageSequenceClip(image_files, fps=fps)\n","clip.write_videofile(video_name+'.mp4')"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"gn7MXIpA9RHG","executionInfo":{"status":"ok","timestamp":1715289864851,"user_tz":240,"elapsed":2839,"user":{"displayName":"Aijun Zhang","userId":"04897292400908543781"}},"outputId":"ab9d6059-be8f-423c-b955-6a829218edb3"},"execution_count":28,"outputs":[{"output_type":"stream","name":"stdout","text":["Moviepy - Building video video3.mp4.\n","Moviepy - Writing video video3.mp4\n","\n"]},{"output_type":"stream","name":"stderr","text":[]},{"output_type":"stream","name":"stdout","text":["Moviepy - Done !\n","Moviepy - video ready video3.mp4\n"]}]}]}