I am using the [b,stats] = robustfit(X,y) to do a regression analysis of my data. Here is what I am doing:
[b,stats_visuomotor_visual] = robustfit(metaRatios(:,3), metaRatios(:,4))
As an output, I get:
b =
0.5221
0.2510
stats_visuomotor_visual =
struct with fields:
ols_s: 0.6643
robust_s: 0.4185
mad_s: 0.4337
s: 0.4577
resid: [27×1 double]
rstud: [27×1 double]
se: [2×1 double]
covb: [2×2 double]
coeffcorr: [2×2 double]
t: [2×1 double]
p: [2×1 double]
w: [27×1 double]
Qy: [2×1 double]
R: [2×2 double]
dfe: 25
h: [27×1 double]
Rtol: 3.2366e-14
In p, there are two values:
0.0124044825989566
0.190293700095903
Why are there two values and what does each of them mean? I expected there to be one p-value.
Does anyone of you have expertise in this? I already looked through the documentation and forums. Some people were asking this question but they did not get any reply as far as I can tell.
Thanks in advance for your help!
Best,
Caroline
Related
I have a problem converting my results to a csv file.
My cell is like this:
[1403x36 double] [1290x36 double] [1813x36 double] [1363x36 double] [1286x36 double]
[1355x36 double] [1194x36 double] [1130x36 double] [1277x36 double] [1494x36 double]
[1447x36 double] [1455x36 double] [1817x36 double] [1434x36 double] [1536x36 double]
I want my CSV file to have (rows x 36).
I tried already cell2csv and i did a loop of fprint also, but neither of them worked.
Thank you in advance.
As with the other users who have commented, it's not clear to me what the desired structure or ordering of the contents of your CSV should be, but here is an example workflow that assumes you would be comfortable first reshaping your cell array into an Q x 1 size (in column-major order), and then concatenating the contents in each cell into an Mx36 matrix.
myCell = {rand(10,36),rand(5,36); rand(4,36),rand(7,36)};
myCell = myCell(:);
myMatrix = cell2mat(myCell);
csvwrite('filename.csv',myMatrix);
Good day,
I have a nested structure in this format.
Data: [1x1 struct]
a: [1x1 struct]
a1: [10x1 double]
a2: [10x1 double]
a3: [10x1 double]
b: [1x1 struct]
b1: [10x1 double]
b2: [10x1 double]
c: [1x1 struct]
c1: [10x1 double]
c2: [10x1 double]
c3: [10x1 double]
c4: [10x1 double]
Each of sub-fields of a, b & c are duration percentage of an event in buckets. The number of elements in each bucket are number of data-sets. I'd like sum the bucket values of each data set and discard the entire data-set if they don't add up to 100%. How may I access each element of the buckets for a, b & c fields of Data using for loop in a simply format.
EDIT: I figured out how to call the sub-fields & it's elements, sum the percentages & now, if the data-sets not adding up to 100 need to fully removed from each sub-field.
field = fieldnames(Data);
for group = 1:length(field)
for subfield = fieldnames(Data.(field{group}))
fieldSize = structfun(#(field) length(field),Data.(field{group}));
nb_datasets = fieldSize(1,1);
for jj = 1:nb_datasets
for ii = 1:length(subfield)
a_dataset_pcts(jj,ii) = Data.(field{group}).(subfield{ii})(jj,1);
end
a_pct_total(jj,:) = sum(a_dataset_pcts(jj,:));
end
end
end
Matlab like matrices Bensa.... You should drop your approach now.
Fixed size datasets
If all a, b, c arrays are 10x1 sized, you should switch immediately to a cell of arrays, being the index the durations, and the arrays columns the datasets for each duration:
a: [1x1 cell]
a{1}: [10xna double]
a{2}: [10xnb double]
a{3}: [10xnc double]
All statistics go easy. You have two for loops, one for the durations, other for the dataset. Matlab vectorizes the functions inside each dataset (most of the time). In here, i am assuming max(.) is your aggregation array function for obtaining statistics:
n=length(a);
for i=1:n
ni=size(a{i},2);
for j=1:ni
max(a{i})
mean(a{i}')'
if j==j0 && i==i0
a{i}(:,j)=[]; % Cleanup j0th event in i0th dataset
end
end
end
Do you need to compare a.a1 & b.b1 & c.c1? If that is true, then you forcedly should create the auxiliar array x, and then get your stats:
for i=1:n
x(:,i)=a{i};
end
max(x')'
Variable size datasets
If the a, b, c arrays are different each time, then you upgrade the last data and switch to a cell of cells:
a{1}: [1xna cell]
a{1,1}: [la(1)x1 double]
...
a{1,na}: [la(na)x1 double]
a{2}: [1xnb cell]
...
a{3}: [1xnc cell]
...
In here the max(.')' approach is useless, because every dataset could be of different length. If that is the case, you still can reference the data as usually through columns:
n=length(a);
for i=1:n
ni(i)=size(a{i},2);
for j=1:ni(i)
li=length(a{i,j}) % Length of each dataset
max(a{i,j})
% mean(a{i}')' % This do not work anymore
if j==j0 && i==i0
a{i,j}=[]; % Cleanup j0th event in i0th dataset
end
end
end
Do this make sense?. Feel free to comment below...
MATLAB :
vso' is graph based visual saliency map (gbvs) of original image, and 'vsd' is graph based visual saliency map (gbvs) of blurred version of the same image.
vso=gbvs('../images/I03.bmp');
vsd=gbvs('../images/i03_17_5.bmp');
Trying to find out similarity map between these both image as -
Svs=(2*vso*vsd+cvs)/(vso^2+vsd^2+cvs);
where cvs is constant, cvs =1.27
Error is : Undefined operator '*' for input arguments of type 'struct'
vso=gbvs('../images/I03.bmp')
vso =
master_map: [24x32 double]
master_map_resized: [384x512 double]
top_level_feat_maps: {[24x32 double] [24x32 double] [24x32 double]}
map_types: {'dklcolor' 'intensity' 'orientation'}
intermed_maps: [1x1 struct]
rawfeatmaps: [1x1 struct]
paramsUsed: [1x1 struct]
vsd=gbvs('../images/i03_17_5.bmp')
vsd =
master_map: [24x32 double]
master_map_resized: [384x512 double]
top_level_feat_maps: {[24x32 double] [24x32 double] [24x32 double]}
map_types: {'dklcolor' 'intensity' 'orientation'}
intermed_maps: [1x1 struct]
rawfeatmaps: [1x1 struct]
paramsUsed: [1x1 struct]
These are variable struct type...
I am working on a Semi Markov model. In it my task is to find the interval transition probability which is a recursive procedure.
The code that I used is below. For this to run the initial condition is F(0)=eye(3,3) matrix.
I am unable to call this initial value in a loop. Is the code that I have written is right? I need a proper suggestion.
Other data used is
C =
Columns 1 through 4
[3x3 double] [3x3 double] [3x3 double] [3x3 double]
Column 5
[3x3 double]
Y =
Columns 1 through 4
[3x3 double] [3x3 double] [3x3 double] [3x3 double]
Column 5
[3x3 double]
The code:
F=mat2cell(zeros(3,15),[3],[3,3,3,3,3])
for j=1:5
for m=1:j
if (j-m)==0
F{:,j}=eye(3,3)
end
F{:,j}=Y{:,j}+sum(C{:,m}*F{:,(j-m)})
end
end
While you overwrite F{:,j} when j-m==0, you still try to access F{:,(j-m)} later on. Doesn't it say "Cell contents indices must be greater than 0" or "Subscript indices must either be real positive integers or logicals."? Matlab arrays and cells are indexed from 1 upwards.
You might need something like
if (j-m)==0
F{:,j}=eye(3,3);
else
F{:,j}=Y{:,j}+sum(C{:,m}*F{:,(j-m)});
end
Don't forget your semicolons or you'll get loads of unnecessary output.
And do you really need cells? What you wrote could easily be implemented using multidimensional arrays, in which case your colons are actually needed. In your specific example, I think they are redundant, it would suffice to write, for instance, F{j-m}.
I'm trying to convert a .mat file into csv, preserving the vector/variable names.
This is one example of what I'm dealing with:
mymat =
model_id: [2217x1 double]
own_dummies: [2217x26 double]
id: [2217x1 double]
product: [2217x1 double]
const: [2217x1 double]
mpd: [2217x1 double]
air: [2217x1 double]
mpg: [2217x1 double]
trend: [2217x1 double]
space: [2217x1 double]
hpwt: [2217x1 double]
cdindex: [20x1 double]
cdid: [2217x1 double]
outshr: [2217x1 double]
firmid: [2217x1 double]
share: [2217x1 double]
price: [2217x1 double]
I've tried using csvwrite('test.csv', mymat) but it gives me an error:
??? Undefined function or method 'real' for input arguments of type 'struct'.
Error in ==> dlmwrite at 192
str = sprintf('%.*g%+.*gi',precn,real(m(i,j)),precn,imag(m(i,j)));\
Error in ==> csvwrite at 32
dlmwrite(filename, m, ',', r, c);
I guess the problem is that I'm feeding csvwrite with a struct instead of a matrix.
I can convert the struct variable to variable to a matrix, but then I would loose the variable names.
Surely there is a better way?
how would you actually map the struct to a csv? you cannot do a straight forward map of fieldnames as entries on first line, because the matrices have incompatible dimensions. So this example you'd have to 1. introduce several columns for own_dummies and 2. expand the cdindex.
So you could write a wrapper to make the matrices themselves compatible and to write the columns yourself into the file. something along the (untested,conceptual) lines of
function saveData(filename, data, type)
% at first bring struct data to sensible format
if strcmp(type)='my_mat_type'
data.own_dummies2 = data.own_dummies[2,:]; % split into seperate columns
%...
data.own_dummies26 = data.own_dummies[26,:];
data.own_dummies = data.own_dummies[1,:];
data.cdindex = [data.cdindex -ones(1,2217-length(data.cdindex)]; % pad any missing values as -1
end;
FD = fopen(filename, 'w');
%todo did it open?
fields = fieldnames(data);
nfields = length(fields);
% create column name values
columns = strcat(strcat(fields,',')); % creates string = col1,col2,...coln,
columns = columns(1:length(columns)-1); %remove trailing comma
CRNL = char([10 13]); % or so
% print columns and newline
fprintf(FD,strcat(columns,CRNL));
dataout = cell2mat(struct2cell(data));
% use your method to write the data to the file
fclose(FD);
naturally, you could also just split the individual fields into separate files where the fieldname is contained within the filename, which might be simpler, generally speaking.