Merge pull request #113 from legend-exp/singlefits

Bugfix in `singlefits.jl`

ext/LegendSpecFitsRecipesBaseExt.jl

@@ -71,7 +71,7 @@ end
             xlims := (ustrip(Measurements.value(report.μ - 5*report.σ)), ustrip(Measurements.value(report.μ + 5*report.σ)))
             yscale --> :identity
             yticks := ([-3, 0, 3])
-            collect(report.h.edges[1])[1:end-1] .+ diff(collect(report.h.edges[1]))[1]/2 , [ifelse(abs(r) < 1e-6, 0.0, r) for r in report.gof.residuals_norm]
+            report.gof.bin_centers, [ifelse(abs(r) < 1e-6, 0.0, r) for r in report.gof.residuals_norm]
         end
     end
 end

src/gof.jl

@@ -168,6 +168,7 @@ Calcualte bin-wise p-value based on poisson distribution for each bin.
  * `residuals` difference: model - data (histogram bin count)
  * `residuals_norm` normalized residuals: model - data / sqrt(model)
  * `p_value_binwise` p-value for each bin based on poisson distribution
+ * `bin_centers` centers of the bins for which the `residuals` were determined
 """
 function get_residuals(f_fit::Base.Callable, h::Histogram{<:Real,1}, v_ml::Union{NamedTuple, AbstractVector})
     # prepare data
@@ -187,6 +188,6 @@ function get_residuals(f_fit::Base.Callable, h::Histogram{<:Real,1}, v_ml::Union
     cdf_value_low = cdf.(dist, model_counts[model_counts .> 0] .- abs.(residuals))
     cdf_value_up = 1 .- cdf.(dist, model_counts[model_counts .> 0] .+ abs.(residuals))
     p_value_binwise = cdf_value_low .+ cdf_value_up # significance of residuals -> ~proabability that residual (for a given bin) is as large as observed or larger
-    return residuals, residuals_norm, p_value_binwise, bin_centers
+    return residuals, residuals_norm, p_value_binwise, bin_centers[model_counts .> 0]
 end
 

src/singlefit.jl

@@ -11,20 +11,20 @@ Returns `report` and `result`` with:
     * `n`: number of counts in the peak
 """
 function fit_single_trunc_gauss(x::Vector{<:Unitful.RealOrRealQuantity}, cuts::NamedTuple{(:low, :high, :max), Tuple{<:T, <:T, <:T}}=(low = zero(first(x))*NaN, high = zero(first(x))*NaN, max = zero(first(x))*NaN); uncertainty::Bool=true) where T<:Unitful.RealOrRealQuantity
-    @assert unit(cuts.low) == unit(cuts.high) == unit(cuts.max) == unit(x[1]) "Units of min_x, max_x and x must be the same"
-    x_unit = unit(x[1])
-    x, cut_low, cut_high, cut_max = ustrip.(x), ustrip(cuts.low), ustrip(cuts.high), ustrip(cuts.max)
+    @assert unit(cuts.low) == unit(cuts.high) == unit(cuts.max) == unit(first(x)) "Units of min_x, max_x and x must be the same"
+    x_unit = unit(first(x))
+    x, cut_low, cut_high, cut_max = ustrip.(x_unit, x), ustrip(x_unit, cuts.low), ustrip(x_unit, cuts.high), ustrip(x_unit, cuts.max)
     cut_low, cut_high = ifelse(isnan(cut_low), minimum(x), cut_low), ifelse(isnan(cut_high), maximum(x), cut_high)
 
-    bin_width = get_friedman_diaconis_bin_width(x[(x .> cut_low) .&& (x .< cut_high)])
-    x_min, x_max = minimum(x), maximum(x)
+    bin_width = get_friedman_diaconis_bin_width(x[cut_low .<= x .<= cut_high])
+    x_min, x_max = extrema(x)
     x_nocut = copy(x)
     h_nocut = fit(Histogram, x, x_min:bin_width:x_max)
     ps = estimate_single_peak_stats_simple(h_nocut)
     @debug "Peak stats: $ps"
 
     # cut peak out of data
-    x = x[(x .> cut_low) .&& (x .< cut_high)]
+    x = x[cut_low .<= x .<= cut_high]
     h = fit(Histogram, x, cut_low:bin_width:cut_high)
     n = length(x)
 
@@ -104,9 +104,11 @@ function fit_single_trunc_gauss(x::Vector{<:Unitful.RealOrRealQuantity}, cuts::N
     end
 
     # create histogram of nocut data for normalization 20 sigma around peak
-    h_nocut = fit(Histogram, x_nocut, v_ml.μ - 20*v_ml.σ:bin_width:v_ml.μ + 20*v_ml.σ)
+    # ensuring that the bin edges are identical to those of h (on which the fit was performed)
+    h_nocut = fit(Histogram, x_nocut, (cut_low + floor(((v_ml.μ-20*v_ml.σ) - cut_low) / bin_width) * bin_width):bin_width:(cut_low + ceil(((v_ml.μ+20*v_ml.σ) - cut_low) / bin_width) * bin_width))
+
     # normalize nocut histogram to PDF of cut histogram
-    h_pdf = Histogram(h_nocut.edges[1], h_nocut.weights ./ sum(abs.(h.weights)) ./ step(h.edges[1]))
+    h_pdf = Histogram(first(h_nocut.edges), h_nocut.weights ./ sum(abs.(h.weights)) ./ step(first(h.edges)))
 
     report = (
         f_fit = t -> Base.Fix2(f_fit, v_ml)(t),
@@ -133,15 +135,15 @@ function fit_half_centered_trunc_gauss(x::Vector{<:Unitful.RealOrRealQuantity},
     x, cut_low, cut_high, cut_max, μ = ustrip.(x), ustrip(cuts.low), ustrip(cuts.high), ustrip(cuts.max), ustrip(μ)
 
     # get peak stats
-    bin_width = get_friedman_diaconis_bin_width(x[(x .> cut_low) .&& (x .< cut_high)])
+    bin_width = get_friedman_diaconis_bin_width(x[cut_low .<= x .<= cut_high])
     x_min, x_max = minimum(x), maximum(x)
     x_nocut = copy(x)
     h_nocut = fit(Histogram, x, x_min:bin_width:x_max)
     ps = estimate_single_peak_stats_simple(h_nocut)
     @debug "Peak stats: $ps"
 
     # cut peak out of data
-    x = ifelse(left, x[(x .> cut_low) .&& (x .< cut_high) .&& x .< μ], x[(x .> cut_low) .&& (x .< cut_high) .&& x .> μ])
+    x = x[cut_low .<= x .<= cut_high .&& ifelse(left, x .<= μ, x .>= μ)]
     h = fit(Histogram, x, ifelse(left, cut_low, μ):bin_width:ifelse(left, μ, cut_high))
     n = length(x)
 
@@ -221,7 +223,9 @@ function fit_half_centered_trunc_gauss(x::Vector{<:Unitful.RealOrRealQuantity},
     end
 
     # create histogram of nocut data for normalization 20 sigma around peak
-    h_nocut = fit(Histogram, x_nocut, v_ml.μ - 20*v_ml.σ:bin_width:v_ml.μ + 20*v_ml.σ)
+    # ensuring that the bin edges are identical to those of h (on which the fit was performed)
+    h_nocut = fit(Histogram, x_nocut, (ifelse(left, cut_low, μ) + floor(((v_ml.μ-20*v_ml.σ) - ifelse(left, cut_low, μ)) / bin_width) * bin_width):bin_width:(ifelse(left, cut_low, μ) + ceil(((v_ml.μ+20*v_ml.σ) - ifelse(left, cut_low, μ)) / bin_width) * bin_width))
+
     # normalize nocut histogram to PDF of cut histogram
     h_pdf = Histogram(h_nocut.edges[1], h_nocut.weights ./ sum(abs.(h.weights)) ./ step(h.edges[1]))
 
@@ -252,15 +256,15 @@ function fit_half_trunc_gauss(x::Vector{<:Unitful.RealOrRealQuantity}, cuts::Nam
     x, cut_low, cut_high, cut_max = ustrip.(x), ustrip(cuts.low), ustrip(cuts.high), ustrip(cuts.max)
 
     # get peak stats
-    bin_width = get_friedman_diaconis_bin_width(x[(x .> cut_low) .&& (x .< cut_high)])
+    bin_width = get_friedman_diaconis_bin_width(x[cut_low .<= x .<= cut_high])
     x_min, x_max = minimum(x), maximum(x)
     x_nocut = copy(x)
     h_nocut = fit(Histogram, x, x_min:bin_width:x_max)
     ps = estimate_single_peak_stats_simple(h_nocut)
     @debug "Peak stats: $ps"
 
     # cut peak out of data
-    x = x[(x .> ifelse(left, cut_low, cut_max)) .&& (x .< ifelse(left, cut_max, cut_high))]
+    x = x[ifelse(left, cut_low, cut_max) .<= x .<= ifelse(left, cut_max, cut_high)]
     h = fit(Histogram, x, ifelse(left, cut_low, cut_max):bin_width:ifelse(left, cut_max, cut_high))
     n = length(x)
 
@@ -340,7 +344,8 @@ function fit_half_trunc_gauss(x::Vector{<:Unitful.RealOrRealQuantity}, cuts::Nam
     end
 
     # create histogram of nocut data for normalization 20 sigma around peak
-    h_nocut = fit(Histogram, x_nocut, v_ml.μ - 20*v_ml.σ:bin_width:v_ml.μ + 20*v_ml.σ)
+    h_nocut = fit(Histogram, x_nocut, (ifelse(left, cut_low, cut_max) + floor(((v_ml.μ-20*v_ml.σ) - ifelse(left, cut_low, cut_max)) / bin_width) * bin_width):bin_width:(ifelse(left, cut_low, cut_max) + ceil(((v_ml.μ+20*v_ml.σ) - ifelse(left, cut_low, cut_max)) / bin_width) * bin_width))
+
     # normalize nocut histogram to PDF of cut histogram
     h_pdf = Histogram(h_nocut.edges[1], h_nocut.weights ./ sum(abs.(h.weights)) ./ step(h.edges[1]))