Skip to content

Commit

Permalink
progress
Browse files Browse the repository at this point in the history
  • Loading branch information
Bryan A. Knowles committed Apr 23, 2020
1 parent 3f7fa31 commit 7f88644
Show file tree
Hide file tree
Showing 2 changed files with 91 additions and 51 deletions.
141 changes: 90 additions & 51 deletions src/ENAModel.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,10 +2,11 @@ struct ENAModel
data::DataFrame # the raw data
codes::Array{Symbol,1}
conversations::Array{Symbol,1}
units::Array{Symbol,1}
unitVar::Symbol
windowSize::Int
groupVar::Union{Nothing,Symbol}
treatmentGroup::Any # if groupVar == treatmentGroup, then 1, else 0
treatmentGroup::Any # if groupVar == treatmentGroup, then 1
controlGroup::Any # if groupVar == treatmentGroup, then 0; if neither, leave it out
confounds::Array{Symbol,1}
metadata::Array{Symbol,1}
# plots::Dict{String,Scene} # dropping this in favor of doing it with display/plot methods, for now maybe
Expand All @@ -16,110 +17,148 @@ struct ENAModel
codeModel::DataFrame # all the code-level data we compute
end

function ENAModel(data::DataFrame, codes::Array{Symbol,1}, conversations::Array{Symbol,1}, units::Array{Symbol,1},
function ENAModel(data::DataFrame, codes::Array{Symbol,1}, conversations::Array{Symbol,1}, unitVar::Symbol,
metadata::Array{Symbol,1}=Symbol[], windowSize::Int=4, confounds::Union{Array{Symbol,1},Nothing}=nothing,
groupVar::Union{Symbol,Nothing}=nothing, treatmentGroup::Any=nothing, rotateBy::Function=svd_rotation)
groupVar::Union{Symbol,Nothing}=nothing, treatmentGroup::Any=nothing, controlGroup::Any=nothing,
rotateBy::Function=svd_rotation!)

# Preparing model structures
## Relationships between codes
relationships = Dict(Symbol(string(code1, "_", code2)) => (i, j)
for (i, code1) in enumerate(codes)
for (j, code2) in enumerate(codes)
if i < j)

unitModel = data[[], [units..., conversations..., metadata...]]
if !isnothing(groupVar)
unitModel = hcat(unitModel, data[[], groupVar])
end

if !isnothing(confounds)
unitModel = hcat(unitModel, data[[], [confounds...]])
## Unit model
unitModel = DataFrame()
if !isnothing(groupVar) && !isnothing(confounds)
unitModel = by(data, unitVar,
[m=>first for m in metadata]...,
groupVar=>first,
[c=>first for c in confounds]...)
elseif !isnothing(groupVar)
unitModel = by(data, unitVar,
[m=>first for m in metadata]...,
groupVar=>first)
elseif !isnothing(confounds)
unitModel = by(data, unitVar,
[m=>first for m in metadata]...,
[c=>first for c in confounds]...)
else
unitModel = by(data, unitVar,
[m=>first for m in metadata]...)
end

unitModel = hcat(unitModel, data[[], [codes...]])
unitModel = hcat(unitModel, DataFrame(Dict(r => Real[]
unitModel = hcat(unitModel, DataFrame(Dict(r => Real[0 for i in 1:nrow(unitModel)]
for r in keys(relationships))))

unitModel = hcat(unitModel, DataFrame(dim_x=Real[], dim_y=Real[], # real sum of my counts times relationship weights
fit_x=Real[], fit_y=Real[])) # fitted sum of the above
unitModel = hcat(unitModel, DataFrame(dim_x=Real[0 for i in 1:nrow(unitModel)], # real sum of my counts times relationship weights
dim_y=Real[0 for i in 1:nrow(unitModel)],
fit_x=Real[0 for i in 1:nrow(unitModel)], # fitted sum of the above
fit_y=Real[0 for i in 1:nrow(unitModel)]))

## Network model
networkModel = DataFrame(relationship=keys(relationships),
thickness=Real[0 for r in relationships], # how thick to make the line
weight_x=Real[0 for r in relationships], # the weight I contribute to dim_x's
weight_y=Real[0 for r in relationships]) # the weight I contribute to dim_y's

## Code model
codeModel = DataFrame(code=codes,
thickness=Real[0 for c in codes], # how thick to make the dot
fit_x=Real[0 for c in codes], # where to plot this code on the fitted plot's x-axis
fit_y=Real[0 for c in codes]) # where to plot this code on the fitted plot's y-axis

# Accumulation step
for convo in groupby(data, [conversations..., units...])
counts = [[0 for j in codes] for i in codes]
howrecents = [Inf for i in codes]

## Raw counts
for line in eachrow(convo)
for (i, code) in enumerate(codes)
if line[code] > 0
howrecents[i] = 0
else
howrecents[i] += 1
end
end
## Raw counts for all the units
counts = Dict(unit => [[0 for j in codes] for i in codes]
for unit in unitModel[!, unitVar])

prev_convo = data[1, conversations]
howrecents = [Inf for c in codes]
for line in eachrow(data)
if prev_convo != line[conversations]
prev_convo = line[conversations]
howrecents .= Inf
end

for r in keys(relationships)
i, j = relationships[r]
if howrecents[i] == 0 && howrecents[j] <= windowSize
counts[i][j] += 1
elseif howrecents[j] == 0 && howrecents[i] <= windowSize
counts[i][j] += 1
end
for (i, code) in enumerate(codes)
if line[code] > 0
howrecents[i] = 0
else
howrecents[i] += 1
end
end

## Put raw counts into the data frame
row = convo[1, [units..., conversations..., metadata..., codes...]]
unit = line[unitVar]
for r in keys(relationships)
i, j = relationships[r]
row = hcat(row, DataFrame(Dict(r => Real[counts[i][j]])))
if howrecents[i] == 0 && howrecents[j] <= windowSize
counts[unit][i][j] += 1
elseif howrecents[j] == 0 && howrecents[i] <= windowSize
counts[unit][i][j] += 1
end
end
end

## Sphere normalize the raw counts
s = std(row[keys(relationships)])
row[keys(relationships)] = map(row[keys(relationships)]) do raw
return raw/s
## Normalize and overwrite the model's placeholders
for unitRow in eachrow(unitModel)
unit = unitRow[unitVar]
vector = [counts[unit][i][j] for (i,j) in values(relationships)]
s = std(vector)
vector /= s
for (k, r) in enumerate(keys(relationships))
unitRow[r] = vector[k]
end
end

## Add the row to the unitModel data frame
row = hcat(row, DataFrame(dim_x=Real[0], dim_y=Real[0], fit_x=Real[0], fit_y=Real[0]))
push!(unitModel, row)
## Simplify the unitModel down to just those in the treatment/control when groupVar is present
if !isnothing(groupVar)
filter!(unitModel) do unitRow
if unitRow[groupVar] in [treatmentGroup, controlGroup]
return true
else
return false
end
end
end

# Rotation step
config = Dict{Symbol,Any}()
if !isnothing(groupVar)
config[:groupVar] = groupVar
config[:treatmentGroup] = treatmentGroup
config[:controlGroup] = controlGroup
end

if !isnothing(confounds)
config[:confounds] = confounds
end

networkModel = rotateBy(unitModel, networkModel, config)
rotateBy(networkModel, unitModel, config)

# Layout step
# TODO compute dim_x and dim_y of the unit-level model
# TODO fit the x and y positions of the unit-level model and code-level model
# TODO compute the dot sizes for the code-level model
end

function svd_rotation(unitModel, networkModel, config)
function svd_rotation!(networkModel, unitModel, config)
# TODO compute the thickness and weights of the network-level model
# TODO use LASSO when confounds present
if haskey(config, :confounds)
# TODO use AC-PCA when confounds present
else
# TODO use PCA otherwise
end
end

function means_rotation(unitModel, networkModel, config)
function means_rotation!(networkModel, unitModel, config)
# TODO compute the thickness and weights of the network-level model
# TODO use moderated MR1 when confounds present
# HERE
if haskey(config, :confounds) && haskey(config, :groupVar)
# TODO use moderated MR1 when confounds present
elseif haskey(config, :groupVar)
# TODO use default MR1 otherwise (this can probably be generalized into the above)
else
error("means_rotation requires a groupVar")
end
end
1 change: 1 addition & 0 deletions src/EpistemicNetworkAnalysis.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@ using GLM
# TODO using Lasso
using Makie

# TODO sample dataset
include("./ENAModel.jl")

end # module

0 comments on commit 7f88644

Please sign in to comment.