-
Notifications
You must be signed in to change notification settings - Fork 0
/
TODO
95 lines (65 loc) · 3.36 KB
/
TODO
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
> "Age" is still an issue. I don't remember seeing a reponse to my earlier
> email on that one. Please let me know what you think of CCost (cumulative
> cost). We can't use 'age'; that's backwards.
>
I like CCost. At times i've referred to the total age of the search (or
simulation), which I think is NOT backwards. There is a duality between age of
the simulation and cost to reach a particular point. I think it is worth
mentioning this.
> Also, the idea of "score" is introduced twice. Perhaps these introductions
> can be merged, and the negated log probability can be motivated more briefly,
> where Score is defined?
>
Sounds good.
> Finally, in a 9pt font the paper must be formatted in block style.
> The other style borders on unreadable. At preset this will cost us
> anotehr two column inches.
>
*sigh*
> A diligent person would find a way to reduce the space around
> \smallverbatiminput{...} and \begin{smallverbatim}. There's more
> space there than we need...
You have the most LaTeX expertise; any suggestions?
--------------
"Age" is still an issue. I don't remember seeing a reponse to my earlier
email on that one. Please let me know what you think of CCost (cumulative
cost). We can't use 'age'; that's backwards.
Also, the idea of "score" is introduced twice. Perhaps these introductions
can be merged, and the negated log probability can be motivated more briefly,
where Score is defined?
Finally, in a 9pt font the paper must be formatted in block style.
The other style borders on unreadable. At preset this will cost us
anotehr two column inches.
A diligent person would find a way to reduce the space around
\smallverbatiminput{...} and \begin{smallverbatim}. There's more
space there than we need...
-------------------
> 1. It wasn't obvious how to use quickcheck, especially writing
> generators and shrinkers, so we put it off to deal with later.
Exactly. A lot of the tutorials I stumbled across personally focused
heavily on using QuickCheck and defining properties for types that
already implemented Arbitrary. There was usually only a smattering
about implementing Arbitrary for custom types. And at that point in
time (for me), working in the Gen monad was baffling.
> - If you can't think of quickcheck properties for your functions,
> keep trying---you will.
This actually wasn't so bad. There were countless times when Noah and
I would stumble across an invariant in the initial design phase and
say, "Oh, that's definitely a QuickCheck property." We just blithely
skipped over the generator/shrinking part.
I do agree that it is an important piece of information to add to the
paper, though.
----------------------------------------------------------------
want to look at Age outside of mrfy in order to do SCIENCE!!!
- not just implementing search; doing research on search algorithsm
- therefore we want age to escape
- accept function: every node has an infinite sequence of children,
then you pick the first good one (might need splittable random
numbers and unsafe performio)
- using gen0, you centralize the consumption of randomness
eliminate these stupid (Bool, Bool) argument to Viterbi
possibility: pass out *all* explored nodes with no age labels, tag
accepted nodes
*or* pass out only accepted nodes, tag each with age
either way, age has to be used internally, because it's used to
decide acceptance