/
fmriStarter.py
118 lines (104 loc) · 4.4 KB
/
fmriStarter.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
'''
Different functions for the final project of CSE546
for predicting words from fMRI images
Authors: Stella Stylianidou, Tomasz Sakrejda
'''
import scipy.io as io
import LassoClass as lassoSolver
from pylab import *
from scipy import sparse
import numpy as np
from fmridataloader import *
from sklearn.decomposition import PCA
import test_suite
def withPCA (dimensions):
'''
It finds the principal components of fmri_train and keeps the
number of components given in dimensions. It then runs lasso on every
semantic feature for a list of lambdas from 80 to 120 and keeps the the
w with the least RMSE on the validation data. It returns the accuracy on
the test data, the best weights and the pca fit. It also saves w on a file.
:param dimensions: number of dimensions for the principal components
:return: accuracy, bestw, the the pca fit
'''
pca = PCA(n_components=dimensions)
pca.fit(fmri_train)
xtrainpcaed= pca.transform(fmri_train)
xtrainPCA = sparse.csc_matrix (xtrainpcaed)
xtest = pca.transform (fmri_test)
num_features = ytrain.shape[1]
d = xtrainPCA.shape[1]
ntotdata = xtrainPCA.shape[0]
ntrain = 250 # number of data to be trained on, rest are used as cross validation
bestw = np.zeros([num_features,d])
accuracy = np.zeros(d)
lasso = lassoSolver.LassoClass()
lambda_list = list(range(80,120)) # list of lambdas to use
for i in range(num_features):
print ('looking at feature ', i)
bestw[i,:] = lasso.descendingLambda(ytrain[0:ntrain,i].reshape(ntrain,1), xtrainPCA[0:ntrain,:], ytrain[ntrain:,i].reshape(ntotdata-ntrain,1), xtrainPCA[ntrain:,:], lambda_list).reshape(d)
wfile = "allwallfeatures_pca300_lambda80_120.mtx" # name of w file to save as
io.mmwrite(wfile, bestw)
test_suite.main(wfile,wordid_train,wordid_test,wordfeature_std,xtest)
return [accuracy,bestw, pca]
def findlambda_crossvalidation():
'''
Cross validation using 60 samples for each validation set. Take median lambda.
:return: l_cross_val
'''
num_features = ytrain.shape[1]
d = xtrain.shape[1]
ntotdata = xtrain.shape[0]
n_crossvalid = 60
bestw = np.zeros([num_features,d])
lasso = lassoSolver.LassoClass()
l_cross_val = []
for i in range(ntotdata/n_crossvalid):
validation = range(i * n_crossvalid, (i+1)*n_crossvalid) #generates the numbers corresponding to the validation set
train = list(set(range(ntotdata)) - set(validation)) # generates nums corresponding to test set
ytraintrain = ytrain[train,1].reshape(len(train),1)
xtraintrain = xtrain[train,:]
yvalid = ytrain[validation,1].reshape(len(validation),1)
xvalid = xtrain[validation,:]
l_best = lasso.descendingLambdaFromMax(ytraintrain, xtraintrain, yvalid, xvalid)
l_cross_val.append(l_best)
print (l_cross_val)
return l_cross_val
def findlambda_50validation():
'''
Finds the best lambda for the first semantic feature,
using the last 50 datapoints for cross validation.
:return: l_best round : the lambda with the least rmse on the cross validation set.
'''
d = xtrain.shape[1]
ntrain = 250
ntotdata = xtrain.shape[0]
lasso = lassoSolver.LassoClass()
l_best = lasso.descendingLambdaFromMax(ytrain[0:ntrain,1].reshape(ntrain,1), xtrain[0:ntrain,:], ytrain[ntrain:,1].reshape(ntotdata-ntrain,1), xtrain[ntrain:,:])
l_best_round = int(round(l_best))
print (l_best_round)
return l_best_round
def run():
'''
It finds the best lambda for the first semantic feature and then
around that value (+/-5) finds the best lambda for the rest
of the semantic features
:return: bestw
'''
num_features = ytrain.shape[1]
d = xtrain.shape[1]
ntrain = 250
ntotdata = xtrain.shape[0]
bestw = np.zeros([num_features,d])
lasso = lassoSolver.LassoClass()
l_best_round = findlambda_50validation()
lambda_list = list(range(l_best_round + 5, l_best_round - 5))
for i in range(num_features):
print ('looking at feature ', i)
bestw[i,:] = lasso.descendingLambda(ytrain[0:ntrain,i].reshape(ntrain,1), xtrain[0:ntrain,:], ytrain[ntrain:,i].reshape(ntotdata-ntrain,1), xtrain[ntrain:,:], lambda_list).reshape(d)
io.mmwrite("allwallfeatures_nov17.mtx", bestw)
return bestw
def main():
print('hello from fmri starter type run () or findlambda_crossvalidation()')
if __name__ == '__main__':
main()