Open illdopejake opened 5 years ago
Hi Jake,
Thank you so much for the feedback, I really appreciate it.
I realized that today as I was working on the project. I am a tad bit of a perfectionist, and if I can't word something as clear as I'd like to, I tend to think about it for a long time without even realizing it. If I want to do this the way I want it done, I don't think I'll be able to cover all the material I wanted to do. So yes, I was actually thinking about cutting out SVMs and CNNs so that I can finish kNN and Logistic Regression by Friday, and hopefully implement SVMs (and hopefully CNNs, although I think it is a stretch) by the official due date of the project (September 6th, if I remember correctly). In any case, I will continue working on this project until I can include CNNs as well.
preprocessing: Yes, I 100% agree that preprocessing is essential to fMRI analysis. However, the reason I wanted to cut it out was because I was worried about time. I do hear your point though, so I will try my best to add a short bit about preprocessing.
I now understand what %matplotlib inline
does. I was going to search that up but just haven't got around to it, so thanks!
I agree, it is a lot of text, but I feel like there are a lot of concepts that have to be described through detailed explanations. The vision I I had initially was that the Brain_Decoding would contain the code, and the ML crash course would contain the theory and details behind it. So, I thought that examples in the ML crash course itself would not be necessary. In order to avoid the monotony of the heavy text, I tried to add some pictures. Once I finish the essentials, I will try to add code snippets, but let's talk about it then I get there.
Yes @emdupre, please do let us know what you think. Also, @glatard suggested that I submit this to NeuroLibre, which was an idea that completely slipped my mind, but is definitely something I'd be interested in doing! @pbellec @nstikov
This might be out of scope for now, but I think your project would like excellent in a Jupyter Book. What do you think @emdupre ?
I think it would be great as a Jupyter Book once the code and text are integrated ! For now, it seems as though the code is isolated from the narrative text, so I'm not sure that it yet makes sense. But if you're intending to bring them together, later, then it'd be nice to keep it as clean and readable as you currently have it -- this is where Jupyter Book really :star2: shines :star2:
Thanks a lot for the input Elizabeth! Given the time constraints, I will continue to keep the files separate. However, once I have time, I will consider putting it altogether and putting it in a Jupyter Book. We can talk about it once we get there.
Hi Norman,
I went through the material you have so far and it's really great. You're a very good writer and the material is very clear and mostly accurate. I wanted to make a note of a few small things.
One general comment is that it looks like you won't be able to get to quite everything. You will have to decide if you want to
a) just keep doing as much as you can and finish with an unfinished project or
b) really focus on having a complete project and cut things that you won't finish altogether. For example, if you finished the fmri notebook, cut out all but maybe two ML algorithms, you could definitely complete that by Friday.
Here are a few small things:
preprocessing -- This is actually a crucial part of doing fmri analysis. I understand why you omitted it here -- you have already preprocessed data -- but its such an integral aspect of dealing with fmri. There are many confounds and considerations that are dealt with to varying degrees of success during the preprocessing. While it may be out of scope to discuss in detail, it might be worthwhile to just touch on the most common steps of fmri processing (e.g. slice timing correction, motion correction, confound regression, registration, smoothing, etc.). Motion and confounds in particular can make a big difference in the signal, see here for example
At some point you ask what
%matplotlib inline
does. This is an example of ipython magic. In ipython or a jupyter notebook running an ipython kernel, after you create a plot with matplotlib, you must do plt.show() to visualize the object. When you do%matplotlib inline
,it automatically plots when you run a cell, without having to specify plt.show(), saving you some time and code.It struck me when going through your ML crashcourse that there wasn't any actual code. The content is fantastic, but it would be even cooler if you could actually demonstrate some of what you say with a few tiny code snippets built into the notebook. Otherwise, it's a lot of text.
This might be out of scope for now, but I think your project would like excellent in a Jupyter Book. What do you think @emdupre ?