Task 1: Resampled t Statistics Assignment

Scaffold Head

import numpy as np import tensorflow as tf def cbf(label): T = np.array(range(128)) eta = np.random.normal(0, 1) eps = np.random.normal(0, 1, 128) a = np.random.randint(16, 32) bma = np.random.randint(32, 96) b = bma+a M = np.zeros(128) M[a:b+1] = 1 if label == 0: return (6+eta)*M+eps if label == 1: return (6+eta)*M*(T-a)/(b-a)+eps if label == 2: return (6+eta)*M*(b-T)/(b-a)+eps def get_data(num_samples, znorm=False, onehot=False, seed=42): np.random.seed(seed) series = np.zeros((num_samples, 128)) labels = np.random.randint(0, 3, num_samples) sigma = lambda x : x if znorm: sigma = lambda x : (x-np.mean(x))/np.std(x) for index, label in enumerate(labels): series[index] = sigma(cbf(label)) if onehot: labels_onehot = np.zeros((num_samples, 3)) for index, label in enumerate(labels): labels_onehot[index][label] = 1 labels = labels_onehot return series, labels class AffineRegression(object): def __init__(self, d, c, dtype=np.float32): self.d = d self.c = c self.A = np.zeros((c, d+1)) def fit(self, X, Y): X_one = np.hstack((X, np.ones((X.shape[0],1)))) Y_one = np.zeros((Y.shape[0], self.c)) for index, label in enumerate(Y): Y_one[index, label] = 1 self.A = np.linalg.pinv(X_one).dot(Y_one) recon = X_one.dot(self.A) loss = np.mean(np.square(recon-Y_one)) metric = np.mean(recon.argmax(axis=-1) == Y) # print(loss, metric) def evaluate(self, X, Y): X_one = np.hstack((X, np.ones((X.shape[0],1)))) Y_one = np.zeros((Y.shape[0], self.c)) for index, label in enumerate(Y): Y_one[index, label] = 1 recon = X_one.dot(self.A) loss = np.mean(np.square(recon-Y_one)) metric = np.mean(recon.argmax(axis=-1) == Y) return (loss, metric) class SoftmaxRegression(object): def __init__ (self, d, c, dtype=tf.float32, alpha=1e-3): self.X = tf.placeholder(shape=(None, d), dtype=dtype) self.Y = tf.placeholder(shape=(None, ), dtype=tf.int64) self.A = tf.Variable(np.zeros((d, c)), dtype=dtype) self.a = tf.Variable(np.zeros((c, )), dtype=dtype) self.F = tf.add(tf.matmul(self.X, self.A), self.a) self.L = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=self.F, labels=self.Y)) self.M = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(self.F, axis=1), self.Y),dtype=dtype)) self.S = tf.train.AdamOptimizer(alpha).minimize(self.L) self.sess = tf.Session() self.sess.run(tf.global_variables_initializer()) def fit(self, X, Y, num_iterations=2**10, batch_size=2**4, print_every=2**10+1): for iteration in range(num_iterations): indices = np.random.choice(X.shape[0], batch_size) feed_dict = {self.X: X[indices], self.Y: Y[indices]} self.sess.run(self.S, feed_dict=feed_dict) if iteration % print_every == print_every-1: loss, metric = self.sess.run([self.L, self.M], feed_dict=feed_dict) print(loss, metric) def evaluate(self, X, Y): feed_dict={self.X: X, self.Y: Y} loss, metric = self.sess.run([self.L, self.M], feed_dict=feed_dict) return (loss, metric)

Start time:

Mo 22 Okt 2018 10:51:00

End time:

Mi 21 Nov 2018 12:00:00

General test timeout:

10.0 seconds

Tests

I try to be a good scientist, I think I carefully worked it out!
I try to be a good scientist, I think I carefully worked it out!