ML lectures

ML-lectures/magic.py

@@ -0,0 +1,45 @@
+import torch
+import numpy as np
+from matplotlib import pyplot as plt
+import matplotlib
+import sys
+
+matplotlib.rcParams['figure.dpi'] = 160
+
+# You need to clone https://github.com/NVlabs/stylegan3
+sys.path.append('./stylegan3/') # because of very strange way to deserialize stylegan
+import importlib.abc # fix import abc error
+
+import dnnlib
+import legacy
+
+PET_MODEL = 'https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/stylegan3-r-afhqv2-512x512.pkl'
+HUMAN_MODEL = 'https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/stylegan3-r-ffhq-1024x1024.pkl'
+
+
+def generate_img(model, seed=None, truncation_psi=1, noise_mode='const'):
+    seed = seed or np.random.randint(0, np.iinfo(np.int32).max)
+    print('Loading networks from "%s"...' % model)
+    device = torch.device('cuda')
+    with dnnlib.util.open_url(model) as f:
+        G = legacy.load_network_pkl(f)['G_ema'].to(device)
+
+    # Labels.
+    label = torch.zeros([1, G.c_dim], device=device)
+
+    # Generate images.
+    print(f'Generating image for seed {seed} ...')
+    z = torch.from_numpy(np.random.RandomState(seed).randn(1, G.z_dim)).to(device)
+
+
+    img = G(z, label, truncation_psi=truncation_psi, noise_mode=noise_mode)
+    img = (img.permute(0, 2, 3, 1) * 127.5 + 128).clamp(0, 255).to(torch.uint8)
+    return img[0].cpu().numpy()
+
+
+def generate_pet(seed=None):
+    return generate_img(PET_MODEL, seed)
+
+
+def generate_human(seed=None):
+    return generate_img(HUMAN_MODEL, seed)

ML-lectures/prelude.py

@@ -0,0 +1,58 @@
+import numpy as np
+import matplotlib
+import scipy.interpolate
+import torch
+
+from numpy.polynomial.polynomial import Polynomial
+from matplotlib import pyplot as plt
+from torch import optim
+from sklearn.model_selection import train_test_split
+from sklearn.linear_model import LogisticRegression, LinearRegression
+from sklearn.preprocessing import PolynomialFeatures
+from sklearn.cluster import KMeans
+from sklearn.ensemble import GradientBoostingClassifier
+from sklearn.tree import DecisionTreeClassifier
+from sklearn.neighbors import KNeighborsClassifier
+from sklearn.tree import plot_tree
+
+
+matplotlib.rcParams.update({'axes.grid' : True})
+# matplotlib.rcParams['figure.figsize'] = [12, 8]
+matplotlib.rcParams['figure.dpi'] = 160
+
+
+def random_points(size: int = 10):
+    x = np.linspace(0, 1, size)
+    y = np.sin(x*10) + x * 3 + np.random.randn(size) * 0.3
+    return np.stack([x, y])
+
+
+def f(x, params):
+    return sum(
+       k * x**i
+       for i, k in enumerate(params)
+    )
+
+
+def lagrange(x, y):
+    poly = scipy.interpolate.lagrange(x, y)
+    # looks too weird https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.lagrange.html
+    return Polynomial(poly.coef[::-1]).coef
+
+
+def cluster(mean, sigma=np.identity(2), count=100):
+   return np.random.randn(count, 2) @ sigma + mean
+
+
+def plot_decision_boundary(model, X, Y, h=0.02, margin=1, cmap='Paired_r'):
+    x_min, x_max = X[:,0].min() - margin, X[:,0].max() + margin
+    y_min, y_max = X[:,1].min() - margin, X[:,1].max() + margin
+    xx, yy = np.meshgrid(
+        np.arange(x_min, x_max, h), np.arange(y_min, y_max, h)
+    )
+    Z = model.predict(np.c_[xx.ravel(), yy.ravel()])
+    Z = Z.reshape(xx.shape)
+
+    plt.contourf(xx, yy, Z, cmap=cmap, alpha=0.55)
+    plt.contour(xx, yy, Z, colors='k', linewidths=0.7)
+    plt.scatter(X[:,0], X[:,1], c=Y, cmap=cmap, edgecolors='k');

ML-lectures/requirements.txt

@@ -0,0 +1,7 @@
+matplotlib==3.5.1
+numpy==1.22.3
+scikit_learn==1.0.2
+scipy==1.8.0
+torch==1.11.0
+tqdm==4.63.2
+tensorflow==2.8.0