kastnerkyle

#!/usr/bin/env python
# Clone or update all a user's gists
# curl -ks https://raw.github.com/gist/5466075/gist-backup.py | USER=fedir python
# USER=fedir python gist-backup.py

import json
import urllib.request, urllib.parse, urllib.error
from subprocess import call
from urllib.request import urlopen
import os

kastnerkyle

import numpy as np
import theano
import theano.tensor as T
import pickle

d = pickle.load(open("info.pkl", mode="rb"))

X = d["X"]
y = d["y"]
fit_function2 = d["fit_function"]

kastnerkyle

# skip hidden files and sed replace
find . -not -path '*/\.*' -type f -exec sed -i -e 's|from pelican|from ../lib/pelican|' {} +

# replace in every file recursive
find . -type f -exec sed -i -e 's/foo/bar/g' {} +

# replace in all files in current dir 
sed -i -- 's/foo/bar/g' *

# tmp_wav.txt is raw text from dropbox downloads list - pull out wav files, making sure to strip leading whitespace

kastnerkyle

import numpy as np

# Broadcast tricks to repeat a matrix
a = np.arange(100 * 10).reshape((100, 10))
# Number of times to clone each entry
clone_count = 2
# axis 0 clone
b = np.ones((1, clone_count, a.shape[1]))
c = (a[:, None, :] * b).reshape((-1, a.shape[-1]))
# axis 1 clone

kastnerkyle

kastnerkyle

#!/usr/bin/env python

'''
Pure Python implementation of some numerical optimizers

Created on Jan 21, 2011

@author Jiahao Chen
'''

kastnerkyle

"""
bitmap utils and much of the ctc code modified from Shawn Tan
"""
# Author: Kyle Kastner
# License: BSD 3-clause
from theano import tensor
from scipy import linalg
import theano
import numpy as np
import matplotlib.pyplot as plt

kastnerkyle

import numpy as np
import matplotlib.pyplot as plt
from scipy import linalg
from sklearn.externals import joblib

mem = joblib.Memory(cachedir='.')


def plot_gp_confidence(gp, show_gp_points=True, X_low=-1, X_high=1,
                       X_count=1000, xlim=None, ylim=None, show=False):

kastnerkyle

class SPPLayer(lasagne.layers.Layer):
    def __init__(self, incoming, **kwargs):
        super(SPPLayer, self).__init__(incoming, **kwargs)

        # divide by 4 gives 16 patches
        self.win1 = (int(np.floor(incoming.output_shape[2]/4.0)), int(np.floor(incoming.output_shape[3]/4.0)))
        self.str1 = (int(np.ceil(incoming.output_shape[2]/4.0)), int(np.ceil(incoming.output_shape[3]/4.0)))

        # divide by 2 gives 4 patches
        self.win2 = (int(np.floor(incoming.output_shape[2]/2.0)), int(np.floor(incoming.output_shape[3]/2.0)))

kastnerkyle

 def gaussian_density_batch(x, mean, stddev, correlation, compute_derivatives=False): 
  """
  Compute the Gaussian density at x for a 2D normal distribution with parameters mean, stddev, correlation.
  
  This works simultaneously on a batch of inputs.  The inputs should have dimensions: 
    x.shape = (n, 1, 2)
    mean.shape = stddev.shape = (n, m, 2)
    correlation.shape = (n, m, 1)
  where n*m is the number of different Gaussian density functions that we want to evaluate, on n input points x.  
  So the same input x is plugged into the density for m Gaussian pdfs.  (This is convenient for evaluating a