However your input_file has been constructed, it has changed from character vector before to string object now. In particular in R2017a, importdata() imports strings as string objects instead of as character vectors.
However, in each case that I have been able to think of in which you might have had your datatype silently change from a character vector to a string object, your previous data type would have been cell array of character vector. For example,
T = array2table({'foo';'bar'}, 'variablenames',{'hello'});
T2 = array2table({"foo";"bar"}, 'variablenames',{'hello'});
if the tables had been generated with readtable() then the data type of T2.hello(2) might have changed to string object, but before T1.hello(2) would be cell array containing character vector, not character vector itself. So if you got a string into input_file now, then unless you changed code yourself, the previous datatype would have had to have been cell array of character vector, and that would not be accepted by textread() as its first argument. So I am suspicious that nothing else has been changed.
In any case, the work around, which will work in both releases, is to use
[a,b,c,d,e,g,h,i] = textread(char(input_file), '%s%s%d%d%f%f%f%f%*f%*f%*f%*f','delimiter',',');
This will work whether input_file turns out to be a character vector, or a cell array entry containing a character vector, or a string object.
