From a3f80b3c0205b15e11fe0c99955f5c021cc1e2c8 Mon Sep 17 00:00:00 2001
From: Tenzin Chan <tenzin_chan@mymail.sutd.edu.sg>
Date: Sun, 28 Apr 2024 02:44:58 -0700
Subject: [PATCH] Added some code for analyzing converged neural code

---
 scripts/analyze.jl | 84 ++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 84 insertions(+)
 create mode 100644 scripts/analyze.jl

diff --git a/scripts/analyze.jl b/scripts/analyze.jl
new file mode 100644
index 0000000..7dcc745
--- /dev/null
+++ b/scripts/analyze.jl
@@ -0,0 +1,84 @@
+import DrWatson
+DrWatson.@quickactivate
+import MEFK, UnicodePlots
+
+
+function restack(line::Vector, length::Int, width::Int)
+    reshape(line, (length, width))
+end
+
+
+function restack(output::Matrix, winsz::Int)
+    outlength, outwidth = size(output)
+    num_neuron = outwidth / winsz |> Int
+    inlength = outlength + winsz - 1
+    res = zeros(Int8, inlength, num_neuron, winsz)
+    for i in 1:outlength
+        rs = restack(output[i, :], winsz, num_neuron)
+        for j in 1:winsz
+            res[i+j-1, :, j] = rs[j, :]  # place signal along diagonal
+        end
+    end
+    res[winsz:end-winsz+1, :, :]  # truncate first winsz-1 and last winsz-1 since they won't have winsz outputs
+end
+
+
+function restack_splits(save_dir::String, fname::String)
+    fpath = joinpath(save_dir, fname)
+    data = DrWatson.wload(fpath)
+    params = DrWatson.parse_savename(fname)[2]
+    winsz = params["winsz"]
+    trial_outputs = Dict()
+    for i in keys(data)
+        trial = data[i]
+        trial_outputs[i] = restack(trial["output"], winsz)
+    end
+    trial_outputs
+end
+
+
+function frac_most_pop(trial_output)
+    winsz = size(trial_output)[end]
+    num_ts = size(trial_output)[1]
+    popular_patt = []
+    counts = Int[]
+    for i in 1:num_ts
+        ts_out = @view trial_output[i, :, :]
+        ts_patterns = unique(ts_out, dims=2)
+        ts_patterns = Dict(ts_patterns[:, j]=>0 for j in 1:size(ts_patterns)[2])
+        for j in 1:winsz
+            for (patt, _) in ts_patterns
+                if patt == @view ts_out[:, j]
+                    ts_patterns[patt] += 1
+                    break
+                end
+            end
+        end
+        cnt, pop_patt = findmax(ts_patterns)
+        push!(popular_patt, pop_patt)
+        push!(counts, cnt)
+    end
+    popular_patt = reduce(hcat, popular_patt)' |> Array
+    probs = counts ./ winsz
+    popular_patt, probs
+end
+
+
+if abspath(PROGRAM_FILE) == @__FILE__
+    save_dir = DrWatson.datadir("exp_pro", "matrix", "split", "complete")#"../data/exp_pro/matrix/split/complete/"
+    for f in readdir(save_dir)[1:10]
+        params = DrWatson.parse_savename(f)[2]
+        if params["binsz"] != 1000
+            continue
+        end
+        println(f)
+        trial_outputs = restack_splits(save_dir, f)
+        # TODO analyze output here, not enough memory to store all results
+        patt, probs = frac_most_pop(trial_outputs["1"])
+        unique_ts_patt = unique(patt, dims=1) 
+        println(size(unique_ts_patt))
+        plot = UnicodePlots.histogram(probs, nbins=10)
+        println(plot)
+    end
+end
+