/
summarize.jl
195 lines (158 loc) · 5.98 KB
/
summarize.jl
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
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
struct ChainDataFrame{NT<:NamedTuple}
name::String
nt::NT
nrows::Int
ncols::Int
function ChainDataFrame(name::String, nt::NamedTuple)
lengths = length(first(nt))
all(x -> length(x) == lengths, nt) || error("Lengths must be equal.")
return new{typeof(nt)}(name, nt, lengths, length(nt))
end
end
ChainDataFrame(nt::NamedTuple) = ChainDataFrame("", nt)
Base.size(c::ChainDataFrame) = (c.nrows, c.ncols)
Base.names(c::ChainDataFrame) = collect(keys(c.nt))
# Display
function Base.show(io::IO, df::ChainDataFrame)
print(io, df.name, " (", df.nrows, " x ", df.ncols, ")")
end
function Base.show(io::IO, ::MIME"text/plain", df::ChainDataFrame)
digits = get(io, :digits, 4)
formatter = PrettyTables.ft_printf("%.$(digits)f")
println(io, df.name)
# Support for PrettyTables 0.9 (`borderless`) and 0.10 (`tf_borderless`)
PrettyTables.pretty_table(
io, df.nt;
formatters = formatter,
tf = isdefined(PrettyTables, :borderless) ? PrettyTables.borderless : PrettyTables.tf_borderless,
)
end
Base.isequal(c1::ChainDataFrame, c2::ChainDataFrame) = isequal(c1, c2)
# Index functions
function Base.getindex(c::ChainDataFrame, s::Union{Colon, Integer, UnitRange}, g::Union{Colon, Integer, UnitRange})
convert(Array, getindex(c, c.nt[:parameters][s], collect(keys(c.nt))[g]))
end
Base.getindex(c::ChainDataFrame, s::Vector{Symbol}, ::Colon) = getindex(c, s)
function Base.getindex(c::ChainDataFrame, s::Union{Symbol, Vector{Symbol}})
getindex(c, s, collect(keys(c.nt)))
end
function Base.getindex(c::ChainDataFrame, s::Union{Colon, Integer, UnitRange}, ks)
getindex(c, c.nt[:parameters][s], ks)
end
# dispatches involing `String` and `AbstractVector{String}`
Base.getindex(c::ChainDataFrame, s::String, ks) = getindex(c, Symbol(s), ks)
function Base.getindex(c::ChainDataFrame, s::AbstractVector{String}, ks)
return getindex(c, Symbol.(s), ks)
end
# dispatch for `Symbol`
Base.getindex(c::ChainDataFrame, s::Symbol, ks) = getindex(c, [s], ks)
function Base.getindex(c::ChainDataFrame, s::AbstractVector{Symbol}, ks::Symbol)
return getindex(c, s, [ks])
end
function Base.getindex(
c::ChainDataFrame,
s::AbstractVector{Symbol},
ks::AbstractVector{Symbol}
)
ind = indexin(s, c.nt[:parameters])
not_found = map(x -> x === nothing, ind)
any(not_found) && error("Cannot find parameters $(s[not_found]) in chain")
# If there are multiple columns, return a new CDF.
if length(ks) > 1
if !(:parameters in ks)
ks = vcat(:parameters, ks)
end
nt = NamedTuple{tuple(ks...)}(tuple([c.nt[k][ind] for k in ks]...))
return ChainDataFrame(c.name, nt)
else
# Otherwise, return a vector if there's multiple parameters
# or just a scalar if there's one parameter.
if length(s) == 1
return c.nt[ks[1]][ind][1]
else
return c.nt[ks[1]][ind]
end
end
end
function Base.lastindex(c::ChainDataFrame, i::Integer)
if i == 1
return c.nrows
elseif i ==2
return c.ncols
else
error("No such dimension")
end
end
function Base.convert(::Type{Array}, c::ChainDataFrame)
T = promote_eltype_namedtuple_tail(c.nt)
return convert(Array{T}, c)
end
function Base.convert(::Type{Array{T}}, c::ChainDataFrame) where {T}
arr = Array{T, 2}(undef, c.nrows, c.ncols - 1)
for (i, k) in enumerate(Iterators.drop(keys(c.nt), 1))
arr[:, i] = c.nt[k]
end
return arr
end
function Base.convert(::Type{Array}, cs::Vector{ChainDataFrame{NamedTuple{K,V}}}) where {K,V}
T = promote_eltype_tuple_type(Base.tuple_type_tail(V))
return convert(Array{T}, cs)
end
function Base.convert(::Type{Array{T}}, cs::Vector{<:ChainDataFrame}) where {T}
return mapreduce((x, y) -> cat(x, y; dims = Val(3)), cs) do c
reshape(convert(Array{T}, c), Val(3))
end
end
"""
summarize(chains, funs...[; sections, func_names = [], name = "", append_chains = true])
Summarize `chains` in a `ChainsDataFrame`.
# Examples
* `summarize(chns)` : Complete chain summary
* `summarize(chns[[:parm1, :parm2]])` : Chain summary of selected parameters
* `summarize(chns; sections=[:parameters])` : Chain summary of :parameters section
* `summarize(chns; sections=[:parameters, :internals])` : Chain summary for multiple sections
"""
function summarize(
chains::Chains, funs...;
sections = _default_sections(chains),
func_names::AbstractVector{Symbol} = Symbol[],
append_chains::Bool = true,
name::String = "",
additional_df = nothing
)
# If we weren't given any functions, fall back to summary stats.
if isempty(funs)
return summarystats(chains; sections, append_chains, name)
end
# Generate a chain to work on.
chn = Chains(chains, _clean_sections(chains, sections))
# Obtain names of parameters.
names_of_params = names(chn)
# If no function names were given, make a new list.
fnames = isempty(func_names) ? collect(nameof.(funs)) : func_names
# Obtain the additional named tuple.
additional_nt = additional_df === nothing ? NamedTuple() : additional_df.nt
if append_chains
# Evaluate the functions.
data = to_matrix(chn)
fvals = [[f(data[:, i]) for i in axes(data, 2)] for f in funs]
# Build the ChainDataFrame.
nt = merge((; parameters = names_of_params, zip(fnames, fvals)...), additional_nt)
df = ChainDataFrame(name, nt)
return df
else
# Evaluate the functions.
data = to_vector_of_matrices(chn)
vector_of_fvals = [[[f(x[:, i]) for i in axes(x, 2)] for f in funs] for x in data]
# Build the ChainDataFrames.
vector_of_nt = [
merge((; parameters = names_of_params, zip(fnames, fvals)...), additional_nt)
for fvals in vector_of_fvals
]
vector_of_df = [
ChainDataFrame(name * " (Chain $i)", nt)
for (i, nt) in enumerate(vector_of_nt)
]
return vector_of_df
end
end