versioninfo()

# http://yomichi.hateblo.jp/entry/2015/09/27/151257
# https://gist.github.com/yomichi/def5921f6b81eb5f7b44

#=
Yuichi Motoyama 2015
This is distributed under the Boost Software License Version 1.0
http://www.boost.org/LICENSE_1_0.txt
=#

@enum DepwarnFlag DepwarnOff=0 DepwarnOn=1 DepwarnError=2

doc"""
- `switch_depwarn!(flag :: Bool)`
- `switch_depwarn!(flag :: DepwarnFlag)`
Enable/Disable deprecation warning. 
- `DepwarnOff` or `false` : switch off deprecation warning
- `DepwarnOn` or `true` : switch on deprecation warning
- `DepwarnError` : turn deprecation warning into error
"""
switch_depwarn!(flag :: Bool) = switch_depwarn!(flag ? DepwarnOn : DepwarnOff)
function switch_depwarn!(flag :: DepwarnFlag)
    old_opt = Base.JLOptions()
    params = map(fieldnames(Base.JLOptions)) do name
        name == :depwarn ? Int(flag) : getfield(old_opt, name)
    end
    new_opt = Base.JLOptions(params...)
    unsafe_store!(cglobal(:jl_options, Base.JLOptions), new_opt)
    flag
end

# one-liner
# unsafe_store!(cglobal(:jl_options, Base.JLOptions), Base.JLOptions(map(fieldnames(Base.JLOptions)) do name; name==:depwarn?0:getfield(Base.JLOptions(), name) end...))

switch_depwarn!(false)

using PyPlot
using Distributions

# xlog(x,y) = if x == 0 then 0 else x*log(x/y)
#
# Warning:
#
#   Don't use x*log(x/y) instead of xlog(x,y).
#   Because log(0) = -Inf and 0*Inf = NaN.
#
function xlog(x::Float64, y::Float64)
    if x == zero(x)
        return zero(x)
    elseif y == zero(y)
        return Inf
    else
        return x*log(x/y)
    end
end
xlog(x,y) = xlog(Float64(x),Float64(y))

x, y = 0, 1
@show x, y
@show x*log(x/y)
@show xlog(x,y);

prodprob(p::Float64, q::Float64) = [p*q, p*(1-q), (1-p)*q, (1-p)*(1-q)]

function expected(a::AbstractArray{T,1}) where T
    n = sum(a)
    return [(a[1]+a[2])*(a[1]+a[3])/n, (a[1]+a[2])*(a[2]+a[4])/n, 
            (a[3]+a[4])*(a[1]+a[3])/n, (a[3]+a[4])*(a[2]+a[4])/n]
end

function chisqtest(a::AbstractArray{Int64,1})
    mu = expected(a)
    chisq = sum((a .- mu).^2 ./mu)
    pval = ccdf(Chisq(1), chisq)
    return pval
end

function gtest(a::AbstractArray{Int64,1})
    mu = expected(a)
    g = 2*sum(xlog.(a,mu)) # Don't use a.*log(a./mu)
    pval = ccdf(Chisq(1), g)
    return pval
end

function pvaluehg(d::Hypergeometric, k::Int64)
    c = params(d)
    amax = min(c[1],c[3])
    p0 = pdf(d, k)
    p1 = 0.0
    pval = 0.0
    for j in 0:amax
        p1 = pdf(d, j)
        pval += ifelse(p1 ≤ p0, p1, 0.0)
    end
    return min(pval, 1.0)
end

function fishertest(a::AbstractArray{Int64,1})
    d = Hypergeometric(a[1]+a[2], a[3]+a[4], a[1]+a[3])
    return pvaluehg(d, a[1])
end

# a が補正パラメーター
# a を小さくするほどP値はより小さくなる.
# a = 1.0 で通常のFisher検定の場合に戻る.
# a = 0.5 で mid-p 版になる.
# 
function pvaluehg_corr(d::Hypergeometric, k::Int64; a=0.5)
    c = params(d)
    amax = min(c[1],c[3])
    p0 = pdf(d, k)
    p1 = 0.0
    pval = 0.0
    for j in 0:amax
        p1 = pdf(d, j)
        pval += ifelse(p1 == p0, a*p1, ifelse(p1 < p0, p1, 0.0))
    end
    return min(pval, 1.0)
end

function fishertest_corr(a::AbstractArray{Int64,1})
    d = Hypergeometric(a[1]+a[2], a[3]+a[4], a[1]+a[3])
    return pvaluehg_corr(d, a[1])
end

ecdf(pval::AbstractArray{Float64,1}, x::Float64) = count(p -> p ≤ x, pval)/length(pval)
ecdf(pval, x) = ecdf(pval, Float64(x))

function randPoisson(n::Int64, p::AbstractArray{Float64}, N::Int64)
    return [
        rand(Poisson(n*p[1]),N)';
        rand(Poisson(n*p[2]),N)';
        rand(Poisson(n*p[3]),N)';
        rand(Poisson(n*p[4]),N)';
    ]
end

function randMultinomial(n::Int64, p::AbstractArray{Float64}, N::Int64)
    return rand(Multinomial(n,p), N)
end
    
function randBinomial(n::Int64, p::AbstractArray{Float64}, N::Int64)
    m = Int64(round(n*(p[1]+p[2])))
    q1 = p[1]/(p[1]+p[2])
    q2 = p[3]/(p[3]+p[4])
    return [rand(Multinomial(m, [q1, 1.0-q1]), N); rand(Multinomial(n-m, [q2, 1.0-q2]), N)]
end

function randHypergeometric(n::Int64, p::AbstractArray{Float64}, N::Int64)
    m1 = Int64(round(n*(p[1]+p[2])))
    m2 = Int64(round(n*(p[3]+p[4])))
    n1 = Int64(round(n*(p[1]+p[3])))
    a = rand(Hypergeometric(m1, m2, n1), N)'
    b = m1 .- a
    c = n1 .- a
    d = m2 .- c
    return [a; b; c; d]
end

function pvaluesby(sampler::T, n::Int64;
        N = 10^4, alpha = 0.05) where T<:Function
    px = collect(0.05:0.05:0.50)
    py = px
    np = length(px)
    prob_chisq    = Array{Float64,2}(np,np)
    prob_g        = Array{Float64,2}(np,np)
    prob_fisher   = Array{Float64,2}(np,np)
    prob_fisher_c = Array{Float64,2}(np,np)
    a = Array{Float64,2}(4,N)
    
    for i in 1:np
        for j in 1:np
            if i > j && sampler != randBinomial
                prob_chisq[i,j]    = prob_chisq[j,i]
                prob_g[i,j]        = prob_g[j,i]
                prob_fisher[i,j]   = prob_fisher[j,i]
                prob_fisher_c[i,j] = prob_fisher_c[j,i]
            else
                a = sampler(n, prodprob(px[i],py[j]), N)
                prob_chisq[i,j]    = ecdf([chisqtest(a[:,i])       for i in 1:size(a,2)], alpha)
                prob_g[i,j]        = ecdf([gtest(a[:,i])           for i in 1:size(a,2)], alpha)
                prob_fisher[i,j]   = ecdf([fishertest(a[:,i])      for i in 1:size(a,2)], alpha)
                prob_fisher_c[i,j] = ecdf([fishertest_corr(a[:,i]) for i in 1:size(a,2)], alpha)
            end
        end
    end
    return alpha, px, prob_chisq, prob_g, prob_fisher, prob_fisher_c
end

function plotcomparisontest(sampler, n; N = 10^4, alpha = 0.05)
    alpha, px, prob_chisq, prob_g, prob_fisher, prob_fisher_c = pvaluesby(sampler, n; N = N, alpha = alpha)
    py = px
    np = length(px)
    ps = [0.0;px]
    vmin = 0.0
    vmax = 2*alpha
    cmap = "RdBu_r"

    figure(figsize=(8,6.4))

    ax1 = subplot2grid((16,20), (0,0), rowspan=7, colspan=8)
    pcolormesh(ps, ps, prob_chisq, cmap=cmap, vmin=vmin, vmax=vmax)
    colorbar(label="probability of p-value \$\\leqq\$ $alpha")
    title("\$\\chi^2\$-test")
    xlabel("marginal probability", fontsize=8)
    ylabel("marginal probability", fontsize=8)

    ax2 = subplot2grid((16,20), (0,12), rowspan=7, colspan=8)
    pcolormesh(ps, ps, prob_g, cmap=cmap, vmin=vmin, vmax=vmax)
    colorbar(label="probability of p-value \$\\leqq\$ $alpha")
    title("G-test")
    xlabel("marginal probability", fontsize=8)
    ylabel("marginal probability", fontsize=8)

    ax3 = subplot2grid((16,20), (9,0), rowspan=7, colspan=8)
    pcolormesh(ps, ps, prob_fisher, cmap=cmap, vmin=vmin, vmax=vmax)
    colorbar(label="probability of p-value \$\\leqq\$ $alpha")
    title("Fisher's exact test")
    xlabel("marginal probability", fontsize=8)
    ylabel("marginal probability", fontsize=8)

    ax3 = subplot2grid((16,20), (9,12), rowspan=7, colspan=8)
    pcolormesh(ps, ps, prob_fisher_c, cmap=cmap, vmin=vmin, vmax=vmax)
    colorbar(label="probability of p-value \$\\leqq\$ $alpha")
    title("mid-p Fisher's exact test")
    xlabel("marginal probability", fontsize=8)
    ylabel("marginal probability", fontsize=8)

    suptitle("sampler = $(typeof(sampler)),  n = $n,  \$\\alpha\$ = $alpha")
end

@time plotcomparisontest(randPoisson, 25, N=10000, alpha=0.05)

@time plotcomparisontest(randPoisson, 50, N=10000, alpha=0.05)

@time plotcomparisontest(randPoisson, 100, N=10000, alpha=0.05)

@time plotcomparisontest(randPoisson, 200, N=10000, alpha=0.05)

@time plotcomparisontest(randMultinomial, 25, N=10000, alpha=0.05)

@time plotcomparisontest(randMultinomial, 50, N=10000, alpha=0.05)

@time plotcomparisontest(randMultinomial, 100, N=10000, alpha=0.05)

@time plotcomparisontest(randMultinomial, 200, N=10000, alpha=0.05)

@time plotcomparisontest(randBinomial, 25, N=10000, alpha=0.05)

@time plotcomparisontest(randBinomial, 50, N=10000, alpha=0.05)

@time plotcomparisontest(randBinomial, 100, N=10000, alpha=0.05)

@time plotcomparisontest(randBinomial, 200, N=10000, alpha=0.05)

@time plotcomparisontest(randHypergeometric, 25, N=10000, alpha=0.05)

@time plotcomparisontest(randHypergeometric, 50, N=10000, alpha=0.05)

@time plotcomparisontest(randHypergeometric, 100, N=10000, alpha=0.05)

@time plotcomparisontest(randHypergeometric, 200, N=10000, alpha=0.05)

function randprob(n, p, q)
    P = prodprob(p,q)
    R = similar(P)
    for i in 1:10000
        R = P .* (1.0 .+ 0.5*rand(4))
        R = R/sum(R)
        if 0.4 < chisqtest(Int64.(round.(n*R))) < 0.6
            break
        end
    end
    return R
end
    
function randpvaluesby(sampler::T, n::Int64;
        N = 10^4, alpha = 0.05) where T<:Function
    px = collect(0.05:0.05:0.50)
    py = px
    np = length(px)
    prob_chisq    = Array{Float64,2}(np,np)
    prob_g        = Array{Float64,2}(np,np)
    prob_fisher   = Array{Float64,2}(np,np)
    prob_fisher_c = Array{Float64,2}(np,np)
    a = Array{Float64,2}(4,N)
    
    for i in 1:np
        for j in 1:np
            a = sampler(n, randprob(n, px[i], py[j]), N)
            prob_chisq[i,j]    = ecdf([chisqtest(a[:,i])  for i in 1:size(a,2)], alpha)
            prob_g[i,j]        = ecdf([gtest(a[:,i])      for i in 1:size(a,2)], alpha)
            prob_fisher[i,j]   = ecdf([fishertest(a[:,i]) for i in 1:size(a,2)], alpha)
            prob_fisher_c[i,j] = ecdf([fishertest(a[:,i]) for i in 1:size(a,2)], alpha)
        end
    end
    return alpha, px, prob_chisq, prob_g, prob_fisher, prob_fisher_c
end

function plotrandcomparisontest(sampler, n; N = 10^4, alpha = 0.05)
    alpha, px, prob_chisq, prob_g, prob_fisher, prob_fisher_c = randpvaluesby(sampler, n; N = N, alpha = alpha)
    py = px
    np = length(px)
    ps = [0.0;px]
    vmin = 0.0
    vmax = 2*alpha
    cmap = "RdBu_r"

    figure(figsize=(8,6.4))

    ax1 = subplot2grid((16,20), (0,0), rowspan=7, colspan=8)
    pcolormesh(ps, ps, prob_chisq, cmap=cmap, vmin=vmin, vmax=vmax)
    colorbar(label="probability of p-value \$\\leqq\$ $alpha")
    title("\$\\chi^2\$-test")
    xlabel("marginal probability", fontsize=8)
    ylabel("marginal probability", fontsize=8)

    ax2 = subplot2grid((16,20), (0,12), rowspan=7, colspan=8)
    pcolormesh(ps, ps, prob_g, cmap=cmap, vmin=vmin, vmax=vmax)
    colorbar(label="probability of p-value \$\\leqq\$ $alpha")
    title("G-test")
    xlabel("marginal probability", fontsize=8)
    ylabel("marginal probability", fontsize=8)

    ax3 = subplot2grid((16,20), (9,0), rowspan=7, colspan=8)
    pcolormesh(ps, ps, prob_fisher, cmap=cmap, vmin=vmin, vmax=vmax)
    colorbar(label="probability of p-value \$\\leqq\$ $alpha")
    title("Fisher's exact test")
    xlabel("marginal probability", fontsize=8)
    ylabel("marginal probability", fontsize=8)

    ax4 = subplot2grid((16,20), (9,12), rowspan=7, colspan=8)
    pcolormesh(ps, ps, prob_fisher_c, cmap=cmap, vmin=vmin, vmax=vmax)
    colorbar(label="probability of p-value \$\\leqq\$ $alpha")
    title("mid-p Fisher's exact test")
    xlabel("marginal probability", fontsize=8)
    ylabel("marginal probability", fontsize=8)

    suptitle("Dependent case: sampler = $(typeof(sampler)),  n = $n,  \$\\alpha\$ = $alpha")
end

n = 100
@show p = prodprob(0.2,0.3)
@show r = randprob(n,0.2,0.3)
chisqtest(Int.(round.(n*r)))

n = 100
@time reshape([chisqtest(Int64.(round.(n*randprob(n,p,q)))) for p in 0.05:0.05:0.5 for q in 0.05:0.05:0.5],10,10)

@time plotcomparisontest(randMultinomial, 25, N=10000, alpha=0.05)
@time plotrandcomparisontest(randMultinomial, 25, N=10000, alpha=0.05)

@time plotcomparisontest(randMultinomial, 50, N=10000, alpha=0.05)
@time plotrandcomparisontest(randMultinomial, 50, N=10000, alpha=0.05)

@time plotcomparisontest(randMultinomial, 100, N=10000, alpha=0.05)
@time plotrandcomparisontest(randMultinomial, 100, N=10000, alpha=0.05)

@time plotcomparisontest(randMultinomial, 200, N=10000, alpha=0.05)
@time plotrandcomparisontest(randMultinomial, 200, N=10000, alpha=0.05)

@time plotcomparisontest(randBinomial, 25, N=10000, alpha=0.05)
@time plotrandcomparisontest(randBinomial, 25, N=10000, alpha=0.05)

@time plotcomparisontest(randBinomial, 50, N=10000, alpha=0.05)
@time plotrandcomparisontest(randBinomial, 50, N=10000, alpha=0.05)

@time plotcomparisontest(randBinomial, 100, N=10000, alpha=0.05)
@time plotrandcomparisontest(randBinomial, 100, N=10000, alpha=0.05)

@time plotcomparisontest(randBinomial, 200, N=10000, alpha=0.05)
@time plotrandcomparisontest(randBinomial, 200, N=10000, alpha=0.05)