-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathml-doc-notes.txt
127 lines (103 loc) · 6.16 KB
/
ml-doc-notes.txt
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
119
120
121
122
123
124
125
126
127
add "Previous:" link at bottom of primary pages
x link to home page on primary pages
x most links need to be contextualized a bit more (see notes below)
x marginal notes??
x target _blank external links
x ...BB styling
where in the sections below can we link to our code examples? maybe in the "General Purpose Algorithms" section?
---------------------------
general overview sections
---------------------------
INTRO PAGE
x p1: "reference" instead of "hypertext tutorial/reference manual"
- p1: link to commercial projects? maybe BB ML projects (reorganize???)
x p2: no longer "digging deeper" so maybe we get rid of this paragraph
x list: What is Machine Learning (start here) << instead of Next:
x p3: should we remove site index at this point?
WHAT IS MACHINE LEARNING
x "modern machine learning" (see...section)
x "You have a lot of data" (see...section)
x "bleeding-edge ML research" context? marginal note?
GENERAL PURPOSE ALGORITHMS
x CHOSE TO UNLINK "model" (see...section)
x "dig into" (see...section)
x Q: "machine learning algorithms excel most at various forms of pattern recognition"... is there something else they excel at that is diff form this?
x "continuous" broken-link
- Q: is translation/transcription a classification or regression problem? same w/ anomaly detection?
x it's cool that Synthesis and Sampling link to examples, can we do that for others? this section could be a great way to really paint the picture for what's possible,
- Q: de-noising: does that include "zoom and enhance"?
DATA IS KEY
x replace key-emoji with "key" (cause the guys)
x "ML Pipeline" (see...section)
x "pre-processing" broken-link
x "model-capacity" (marginal note?)
x "empirically tested" (marginal note?)
x "features" (too big for marginal note? new section?)
x "performance of that model" (see ...section)
x "overfitting" (marginal note?)
x CHOSE TO UNLINK "supervised" (see...section)
x "data split Wikipedia page]()" << markdown code showing in text
- maybe add section on generating ur own data ( style transfer, pix2pix )
MACHINE LEARNING MODELS
x CHOSE TO UNLINK "training" broken-link
x models are: ...The algorithm << disambiguate from way u use algorithm in 1st section. maybe combine with first buttlet: The "trained/learned" algorithm, or maybe keep but change to "algorithms" instead of "The..."
x "Function approximators" (see...section)
x "model architecture" (marginal note?)
x (removed) NEED TO FIND A NEW PLACE FOR IT what's the reasoning behind last paragraph?... a bit confusing, maybe put in general purpose algorithms section
- should the rest of this be explained earlier in docs, along side it's relationship to "architectures"?
MACHINE LEARNING PIPELINE
- change graph to be shades of BB color (#e40477)
- FIGURE OUT HOW TO HANDLE THE SECTION THAT GETS ADDED TO DATA AS KEY HERE. FIFTH BULLET POINT, NEW PARAGRAPH? Get Data: maybe also add here "create your own" like pix2pix mention before
x "normalization" broken-link (noticed there was a secondary page, new section?), or should this info be in the ipython example?
x "split itd" spelling edit
x "train your model", this is a good link, but maybe this is the wrong place for it (link to it elsewhere), maybe this should be a marginal note that references the examples/guides as good examples of training in context
x "model architecture" (marginal note?)
x "validation loss)" where's this link supposed to go?
x "evaluate its performance" (see...section)
x "many iterations of training and evaluating" (marginal note?)
x "different experiments" (marginal note?)
x "using your trained model" (marginal note?)
x should we mention tensorflow.js in "deploy your model"
x "These models" (marginal note?)
TYPES OF LEARNING
x "K-Means" broken-link
x "can be though of as a form" spelling edit
TYPES OF TASKS
- consider making this a part of the re-working of General Purpose Algorithms section
x "we mentioned before" (see...section)
x "a a probability distribution" grammar edit
x "probability distribution" (too big for marginal note? new section?)
PERFORMANCE MEASURES
x "model to know has well it is doing itself" spelling edit
x "sythesized portrait." spelling edit
x could we see MSE in JS as well?
x "loss functions for common tasks", links to keras, do we use this lib in examples? should we link to tensorflow?
LINEAR REGRESSION
x CHOSE TO UNLINK "regression problem" (see...section)
x LEAVE AS IS "an optimization algorithm", also keras, same note as before
x "local minima" broken-link
x "test data" (see...section)
x "design matrix" (features, mentioned before, new section?)
x (removed) ...what's the reasoning for this section?
- example, is that's what's going on here: http://www.r2d3.us/visual-intro-to-machine-learning-part-1/
NEURAL NETWORKS AND DEEP LEARNING
- "model capacity" links to page rather than marginal note here?
x "overfitting" (marginal note?)
x "regularization" (too big for marginal note? new section?)
x should the example be part of "code examples" section?
x "baseline model" broken-link
x "Branger_Briz machine learning projects" broken-link
---------------------------
guides && code examples
---------------------------
contextualize "guides" into index page
move linear regression && NN into guies section
maybe also those couple of orphan pages
can we make these more like "small projects", demos that could get press on their own + demos that provide a bit more context for the "kinds of" situations u'd use these techniques?
for example:
- s the "Clustering and Dimensionality Reduction" related to ur word calculator? could be that
- Text generation in browser, instead of Shakespeare what about Trump Tweet generator?
- Real-time pose estimation, what about a "dabbing" alarm, or student raising hand for question, or something more artful >> controlling music through dance?
( side note: wonder if there's a way to use LEAP data to trin a NN that speaks words from ASL )
we don't have teachable machines mentioned anywhere?
we dont' have our pix2pix demo anywhere?