JesseLivezey

import numpy as np
import matplotlib.pyplot as plt

pops = np.array([39538223,
29145505,
21538187,
20201249,
13011844,
12812508,
11799448,

JesseLivezey

# All sizes are in fraction of the figure size

figsize = (6, 7.5)
f = plt.figure(figsize=figsize)

# borders
top_edge = .03
bot_edge = .055
l_edge = .15
r_edge = .01

JesseLivezey

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

def numpy_version(prob, cases):
    return np.matmul(prob.transpose(2,0,1), cases.T).T
  
last_dim = 50
dot_dim = 1000
other_dim = 1000

JesseLivezey

import numpy as np
from numpy.fft import fft as nfft, ifft as nifft
from accelerate.mkl.fftpack import fft as mfft, ifft as mifft
from pyfftw.interfaces.numpy_fft import fft as wfft, ifft as wifft
import matplotlib.pyplot as plt
import time

start = 3
end = 5
diff = 4*(end-start)+1

JesseLivezey

#!/usr/bin/env python
import argparse, time, os, subprocess
import numpy as np

import theano
from theano.tensor.nnet.corr import (CorrMM, CorrMM_gradInputs,
                                     CorrMM_gradWeights)
from theano.compat.python2x import OrderedDict

JesseLivezey

#this is the constraint that makes Pyhy[i,i] > Pyhy[i,j]
def con4(inputArray):
    Pyhy_i = inputArray.reshape(n,n)
    # get diagonal vector (Pyhy[i,i])
    Pyhy_diag = np.diag(Pyhy_i)
    # get max of each row
    Pyhy_max = Pyhy.max(axis=1)
    # we want the difference to be positive or zero
    diff = Pyhy_diag - Pyhy_max
    # we want all of the differences to be positive, so we can just return the most negative one

JesseLivezey

def corrcoef(X, Y):
    """
    Return Pearson correlation coefficients between 1 and n variables.
    
    Parameters
    ----------
    X : (1 x dim) matrix, single voxel
    Y : (n x dim) matrix, other voxels
    """
    X_no_mean = X - X.mean(keepdims=True)

JesseLivezey

def log_multivariate_normal_density(X, means, covars, min_covar=1.e-7):
    """Log probability for full covariance matrices. """
    if hasattr(linalg, 'solve_triangular'):
        # only in scipy since 0.9
        solve_triangular = linalg.solve_triangular
    else:
        # slower, but works
        solve_triangular = linalg.solve
    n_samples, n_dim = X.shape
    nmix = len(means)