From 5526f7ac73d02aff5c078b553e4e66a2d61e9308 Mon Sep 17 00:00:00 2001 From: Florian Jung Date: Sun, 30 Aug 2015 22:39:22 +0200 Subject: more analyzing stuff --- stats.py | 35 ++++++++++++++++++++++++++++------- 1 file changed, 28 insertions(+), 7 deletions(-) (limited to 'stats.py') diff --git a/stats.py b/stats.py index 42ee0ea..d65c6d3 100644 --- a/stats.py +++ b/stats.py @@ -26,6 +26,8 @@ def quantile(values, q): return 0 def find_smallest_q_confidence_area(values, q): + """Calculates the (mid, delta) with the smallest delta, such that at least q * len(values) + lie within the interval [mid-delta, mid+delta].""" try: mid = min(values, key = lambda value : quantile(list(map(lambda x : abs(x-value), values)), q)) deviation = quantile(list(map(lambda x : abs(x-mid), values)),q) @@ -34,6 +36,14 @@ def find_smallest_q_confidence_area(values, q): except: return 0,0 +def get_delta_confidence(values, mid, delta): + #"""Calculates which fraction of the values lie within [mid-delta, mid+delta]""" + #try: + return len(list(filter(lambda v : (mid-delta <= v and v <= mid+delta), values))) / len(values) + #except: + # raise + # return 0 + def avg(values): if not isinstance(values, dict): return sum(values)/len(values) @@ -359,11 +369,17 @@ class Stats: for ncells in sorted(self.data.size_vs_visible_window.keys()): if len(self.data.size_vs_visible_window[ncells]) > 0: print("\nwith "+str(ncells)+" cells, depending on sum(size)") - self.analyze_visible_window_helper(self.data.size_vs_visible_window[ncells], verbose) + try: + self.analyze_visible_window_helper(self.data.size_vs_visible_window[ncells], verbose) + except ZeroDivisionError: + print("\toops.") for ncells in sorted(self.data.mass_vs_visible_window.keys()): if len(self.data.mass_vs_visible_window[ncells]) > 0: print("\nwith "+str(ncells)+" cells, depending on sum(mass)") - self.analyze_visible_window_helper(self.data.mass_vs_visible_window[ncells], verbose) + try: + self.analyze_visible_window_helper(self.data.mass_vs_visible_window[ncells], verbose) + except ZeroDivisionError: + print("\toops.") def analyze_deviations(self, celltype): ds = self.data.eject_deviations[celltype] @@ -412,8 +428,13 @@ class Stats: #quant = quantile(list(map(lambda v : abs(v-midpoints[maxidx]), ds)), q/100) #print("\t"+str(q)+"% of values lie have a distance of at most "+str(quant)+" from the maximum") - print("\t75%% of the distances lie in the interval %.2f plusminus %.2f" % find_smallest_q_confidence_area(ds, 0.75)) - print("\t75%% of the flight lengths lie in the interval %.2f plusminus %.2f" % find_smallest_q_confidence_area(ns, 0.75)) + mid, delta = find_smallest_q_confidence_area(ds, 0.75) + print("\t75%% of the distances lie in the interval %.2f plusminus %.2f" % (mid,delta)) + print("\t%2d%% of the distances lie in the interval %.2f plusminus %.2f" % (100*get_delta_confidence(ds, mid, delta*1.2), mid, delta*1.2) ) + print("\tmax = %.2f" % (max(ds))) + mid, delta = find_smallest_q_confidence_area(ns, 0.75) + print("\t75%% of the flight lengths lie in the interval %.2f plusminus %.2f" % (mid,delta)) + print("\t%2d%% of the flight lengths lie in the interval %.2f plusminus %.2f" % (100*get_delta_confidence(ns,mid,delta*1.2),mid,delta*1.2)) print("") def analyze_virus_sizes(self): @@ -424,7 +445,7 @@ class Stats: def analyze_remerge(self): relevant = list(filter(lambda r : r.is_parent_child, self.data.remerging.values())) durations = list(map(lambda r : r.end_time - r.begin_time, relevant)) - print(fit_gaussian(durations)) + print("75%% of the remerge durations lie at %.2f plusminus %.2f frames" % find_smallest_q_confidence_area(durations,0.75)) waittimes = list(map(lambda r : r.begin_time - max(r.birth1, r.birth2), relevant)) - print(fit_gaussian(waittimes)) - + print("75%% of the remerges were started after %.2f plusminus %.2f frames" % find_smallest_q_confidence_area(waittimes,0.75)) + -- cgit v1.2.3