# load data
data = scipy.io.loadmat('../datasets/regularization.mat')
X = data['X']
Y = data['y']
# init model
model = miniml.Model()
model.dense(20, 'relu', 'xavier')
model.dense(3, 'relu', 'xavier')
model.dense(1, 'sigmoid', 'xavier')
# init params
rate = 0.3
epochs = 30000
lamb = 0.7
# train model
optimizer = miniml.GradDescent(cost='bce',
epochs=epochs,
init_seed=3,
dropout_seed=1,
store=1000,
verbose=10000)
costs = optimizer.train(model, X, Y, rate, lamb=lamb)
# plot results
miniml.print_accuracy(model, X, Y)
miniml.plot_costs(epochs, costs=costs)
miniml.plot_boundaries(model, X, Y)
# shuffle data
X, Y = miniml.shuffle_data(X, Y, seed=48)
# make the target label virginica = 1 and the rest 0
Y = (Y == 2).astype('int')
# create model
model = miniml.Model()
model.dense(5, 'sigmoid', 'xavier')
model.dense(3, 'sigmoid', 'xavier')
model.dense(1, 'sigmoid', 'xavier')
# init params
rate = 1
epochs = 5000
# train model
# note that original implementation is without averaging across examples in LinearLayer.backward(...)
optimizer = miniml.GradDescent(
cost = 'bce',
epochs = epochs,
init_seed = 48,
verbose = 1000)
costs = optimizer.train(model, X, Y, rate)
# plot results
miniml.print_accuracy(model, X, Y)
miniml.plot_costs(epochs, costs=costs)
miniml.plot_boundaries(model, X, Y)
[-1.0, -1.4]])
y = np.array([0, 0, 1, 0, 2, 1, 1, 1, 1, 0, 0, 2, 2, 2, 1, 0, 1, 2, 2, 2])
# convert to one-hot
Y, cats = miniml.to_categorical(y)
C = len(cats)
# create model
model = miniml.Model()
# model.dense(32, 'relu', 'he')
model.dense(C, 'softmax', 'plain')
# init params
rate = 2
epochs = 40
# train model
optimizer = miniml.GradDescent(cost='ce',
epochs=epochs,
init_seed=48,
store=1,
verbose=10)
costs = optimizer.train(model, X, Y, rate)
# plot results
miniml.print_accuracy(model, X, Y)
miniml.plot_costs(epochs, costs=costs)
miniml.plot_boundaries(model, X, Y)
# Adapted from DeepLearning.AI
# load data
np.random.seed(1)
X, Y = sklearn.datasets.make_circles(n_samples=300, noise=.05)
Y = Y.reshape((len(Y), 1))
# init model
model = miniml.Model()
model.dense(10, 'relu', 'he')
model.dense(5, 'relu', 'he')
model.dense(1, 'sigmoid', 'he')
# init params
rate = 0.01
epochs = 15000
# train model
optimizer = miniml.GradDescent(cost='bce',
epochs=epochs,
init_seed=3,
store=1000)
costs = optimizer.train(model, X, Y, rate=rate)
# plot results
miniml.print_accuracy(model, X, Y)
miniml.plot_costs(epochs, costs=costs)
miniml.plot_boundaries(model, X, Y)
import numpy as np # Adapted from: # https://github.com/RafayAK/NothingButNumPy/blob/master/Understanding_and_Creating_NNs/3_layer_toy_network_XOR.ipynb # init data X = np.array([[0, 0], [0, 1], [1, 0], [1, 1]]) Y = np.array([[0], [1], [1], [0]]) # create model model = miniml.Model() model.dense(5, 'sigmoid', 'xavier') model.dense(3, 'sigmoid', 'xavier') model.dense(1, 'sigmoid', 'xavier') # init params rate = 1 epochs = 20000 # train model # note that original implementation is without averaging across examples in LinearLayer.backward(...) optimizer = miniml.GradDescent(cost='mse', epochs=epochs, init_seed=48) costs = optimizer.train(model, X, Y, rate) # plot results miniml.print_accuracy(model, X, Y) miniml.plot_costs(epochs, costs=costs) miniml.plot_boundaries(model, X, Y)