fMRI example

In [1]:

    

import numpy as np
from scipy.io import loadmat
from os.path import join
import glob
import pandas as pd
from sklearn.model_selection import train_test_split


    

In [2]:

    

def center_normalize(x):
    return (x - np.mean(x)) / np.std(x)


    

In [3]:

    

#data dir test 
data_dir = '../data/greco_mri'
behav_dir = '../data/greco_behav'

#rois
rois = ['left_CA1','right_CA1','left_DG','right_DG'];

#subjec list
subjects = ['S1_A','S2_B','S3_A','S4_A','S5_A','S6_A','S7_A','S8_B',
            'S9_A','S10_B','S11_B','S12_B','S13_B','S14_B','S15_B',
            'S16_A','S21_B','S22_B','S23_B','S24_A'];

voxel_cutoff = 130;

subjects = np.array(subjects);


# does tf want a list or tuple?
data_y = [];
data_x = [];
data_y_TEST = [];
data_x_TEST = [];

sub_rand_idx = np.random.permutation(20);
subjects_train = subjects[sub_rand_idx[0:-1]]
subjects_test  = [subjects[sub_rand_idx[-1]]];

num_voxels = np.empty(subjects_train.shape)
num_trials = np.empty(subjects_train.shape)


for i,subj in enumerate(subjects_train):
    fname = join(data_dir,subj + '_right_DG.csv')
    betas= pd.read_csv(fname,header=None)
    behav = pd.read_table(join(behav_dir,subj,subj + '.txt'))
#     idx = np.random.permutation(100);
    idx = np.array(range(0,voxel_cutoff))

    for j in range(0,betas.shape[0]):
 
        data_x.append(center_normalize(betas.iloc[j,idx].values))
        data_y.append(behav['currCity'][j])
    num_trials[i] = betas.shape[0];
    num_voxels[i] = betas.shape[1];
 
for i,subj in enumerate(subjects_test):
    fname = join(data_dir,subj + '_right_DG.csv')
    betas= pd.read_csv(fname,header=None)
    behav = pd.read_table(join(behav_dir,subj,subj + '.txt'))
    idx = np.array(range(0,voxel_cutoff))

    for j in range(0,betas.shape[0]):

        data_x_TEST.append(center_normalize(betas.iloc[j,idx].values))
        data_y_TEST.append(behav['currCity'][j])


    

In [4]:

    

print('minimum number of voxels: ' + str(np.min(num_voxels)))
print('number of trials per subject: ' + str(np.mean(num_trials)))
print('number of subjects: ' + str(len(subjects_train)))


    

    




minimum number of voxels: 142.0
number of trials per subject: 99.9473684211
number of subjects: 19

In [5]:

    

# one hot encoding
ynp=np.array(data_y)
data_y_onehot=[0]*len(ynp)
for i,j in enumerate(ynp):
    data_y_onehot[i]=[0]*ynp.max()
    data_y_onehot[i][j-1]=1
    
    
# one hot encoding
ynp=np.array(data_y_TEST)
data_y_TEST_onehot=[0]*len(ynp)
for i,j in enumerate(ynp):
    data_y_TEST_onehot[i]=[0]*ynp.max()
    data_y_TEST_onehot[i][j-1]=1


    

In [6]:

    

# X_train, X_valid, Y_train, Y_valid = train_test_split(data_x, data_y_onehot, 
#                                                     test_size=0.2, 
#                                                     stratify=data_y_onehot)

X_train = data_x;
Y_train = data_y_onehot;

X_valid = data_x_TEST;
Y_valid = data_y_TEST_onehot;


    

In [7]:

    

import tensorflow as tf 

sess = tf.InteractiveSession()

x = tf.placeholder(tf.float32, shape=[None, voxel_cutoff])
y_ = tf.placeholder(tf.float32, shape=[None, 3])


W = tf.Variable(tf.zeros([voxel_cutoff,3]))
b = tf.Variable(tf.zeros([3]))

sess.run(tf.global_variables_initializer())

y = tf.matmul(x,W) + b

cross_entropy = tf.reduce_mean(
    tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y))


    

In [8]:

    

%%time
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)


for i in range(1000):
    train_step.run(feed_dict={x: X_train, y_: Y_train})


    

    




CPU times: user 5.31 s, sys: 251 ms, total: 5.56 s
Wall time: 4.14 s

In [9]:

    

#tf.argmax gives an index of the highest entry in a tensor along some axis
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))

#we can take this list of booleans and calculate the fraction correct
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

#print the accuracy
print(accuracy.eval(feed_dict={x: X_valid, y_: Y_valid}))


    

    




0.32

In [48]:

    

def init_weights(shape):
    """ Weight initialization """
    weights = tf.random_normal(shape, stddev=0.1)
    return tf.Variable(weights)

def forwardprop(X, w_1, w_2):
    """
    Forward-propagation.
    IMPORTANT: yhat is not softmax since TensorFlow's softmax_cross_entropy_with_logits() does that internally.
    """
    h    = tf.nn.sigmoid(tf.matmul(X, w_1))  # The \sigma function
    yhat = tf.matmul(h, w_2)  # The \varphi function
    return yhat


# Layer's sizes
x_size = voxel_cutoff   
h_size = 256               
y_size = 3  

# Symbols
X = tf.placeholder("float", shape=[None, x_size])
y = tf.placeholder("float", shape=[None, y_size])

# Weight initializations
w_1 = init_weights((x_size, h_size))
w_2 = init_weights((h_size, y_size))

# Forward propagation
yhat    = forwardprop(X, w_1, w_2)
predict = tf.argmax(yhat, axis=1)

# Backward propagation
cost    = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=yhat))
updates = tf.train.GradientDescentOptimizer(0.01).minimize(cost)

# Run SGD
sess = tf.Session()
init = tf.global_variables_initializer()
sess.run(init)

for epoch in range(100):
    # Train with each example
    for i in range(len(X_train)):
        sess.run(updates, feed_dict={X: X_train[i: i + 1], y: Y_train[i: i + 1]})

    train_accuracy = np.mean(np.argmax(Y_train, axis=1) ==
                             sess.run(predict, feed_dict={X: X_train, y: Y_train}))
    test_accuracy  = np.mean(np.argmax(Y_valid, axis=1) ==
                             sess.run(predict, feed_dict={X: X_valid, y: Y_valid}))

    print("Epoch = %d, train accuracy = %.2f%%, test accuracy = %.2f%%"
          % (epoch + 1, 100. * train_accuracy, 100. * test_accuracy))


    

    




Epoch = 1, train accuracy = 37.60%, test accuracy = 35.00%
Epoch = 2, train accuracy = 39.44%, test accuracy = 33.00%
Epoch = 3, train accuracy = 39.86%, test accuracy = 33.00%
Epoch = 4, train accuracy = 40.55%, test accuracy = 30.00%
Epoch = 5, train accuracy = 41.34%, test accuracy = 31.00%
Epoch = 6, train accuracy = 42.02%, test accuracy = 30.00%
Epoch = 7, train accuracy = 42.92%, test accuracy = 30.00%
Epoch = 8, train accuracy = 43.44%, test accuracy = 32.00%
Epoch = 9, train accuracy = 44.39%, test accuracy = 30.00%
Epoch = 10, train accuracy = 45.44%, test accuracy = 29.00%
Epoch = 11, train accuracy = 46.55%, test accuracy = 30.00%
Epoch = 12, train accuracy = 48.29%, test accuracy = 30.00%
Epoch = 13, train accuracy = 49.97%, test accuracy = 29.00%
Epoch = 14, train accuracy = 51.45%, test accuracy = 28.00%
Epoch = 15, train accuracy = 54.13%, test accuracy = 30.00%
Epoch = 16, train accuracy = 56.40%, test accuracy = 30.00%
Epoch = 17, train accuracy = 59.35%, test accuracy = 30.00%
Epoch = 18, train accuracy = 63.24%, test accuracy = 32.00%
Epoch = 19, train accuracy = 67.35%, test accuracy = 31.00%
Epoch = 20, train accuracy = 71.09%, test accuracy = 31.00%
Epoch = 21, train accuracy = 76.57%, test accuracy = 32.00%
Epoch = 22, train accuracy = 80.25%, test accuracy = 33.00%
Epoch = 23, train accuracy = 85.52%, test accuracy = 36.00%
Epoch = 24, train accuracy = 88.63%, test accuracy = 38.00%
Epoch = 25, train accuracy = 92.00%, test accuracy = 38.00%
Epoch = 26, train accuracy = 94.05%, test accuracy = 36.00%
Epoch = 27, train accuracy = 95.26%, test accuracy = 37.00%
Epoch = 28, train accuracy = 96.74%, test accuracy = 37.00%
Epoch = 29, train accuracy = 97.63%, test accuracy = 36.00%
Epoch = 30, train accuracy = 98.42%, test accuracy = 33.00%
Epoch = 31, train accuracy = 99.05%, test accuracy = 33.00%
Epoch = 32, train accuracy = 99.63%, test accuracy = 31.00%
Epoch = 33, train accuracy = 99.79%, test accuracy = 31.00%
Epoch = 34, train accuracy = 99.84%, test accuracy = 32.00%
Epoch = 35, train accuracy = 99.84%, test accuracy = 32.00%
Epoch = 36, train accuracy = 99.95%, test accuracy = 30.00%
Epoch = 37, train accuracy = 99.95%, test accuracy = 30.00%
Epoch = 38, train accuracy = 99.95%, test accuracy = 29.00%
Epoch = 39, train accuracy = 100.00%, test accuracy = 29.00%
Epoch = 40, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 41, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 42, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 43, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 44, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 45, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 46, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 47, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 48, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 49, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 50, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 51, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 52, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 53, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 54, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 55, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 56, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 57, train accuracy = 100.00%, test accuracy = 29.00%
Epoch = 58, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 59, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 60, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 61, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 62, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 63, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 64, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 65, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 66, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 67, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 68, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 69, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 70, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 71, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 72, train accuracy = 100.00%, test accuracy = 28.00%
Epoch = 73, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 74, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 75, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 76, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 77, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 78, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 79, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 80, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 81, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 82, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 83, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 84, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 85, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 86, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 87, train accuracy = 100.00%, test accuracy = 27.00%
Epoch = 88, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 89, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 90, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 91, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 92, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 93, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 94, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 95, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 96, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 97, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 98, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 99, train accuracy = 100.00%, test accuracy = 26.00%
Epoch = 100, train accuracy = 100.00%, test accuracy = 26.00%

In [ ]: