239 lines
8.9 KiB
Python
239 lines
8.9 KiB
Python
import os
|
|
from collections import Counter
|
|
from time import sleep
|
|
import logging
|
|
|
|
import matplotlib.pyplot as plt
|
|
from matplotlib import rcParams
|
|
import numpy as np
|
|
import pandas as pd
|
|
|
|
logger = logging.getLogger(__name__)
|
|
logging.basicConfig(level=logging.INFO) #change for debug prints
|
|
|
|
INJECTION_FILE = os.path.join(".","out","2025-07-31T17-25-54.txt")
|
|
DETECTION_FILE = os.path.join(".","out","plotOut.txt")
|
|
|
|
def linesplit(line):
|
|
"""
|
|
Splits line of file into useful components.
|
|
|
|
Returns (dm, pulse width, originating filename).
|
|
"""
|
|
|
|
dm, line = line.split("pc/cc")
|
|
dm = int(dm)
|
|
pw, fname = line.split(" s ")
|
|
pw = float(pw)
|
|
fname = fname.strip().removesuffix("_injected")
|
|
return (dm, pw, fname)
|
|
|
|
def updateDic(lines, dic):
|
|
"""
|
|
Processes a list of lines and adds them to given dictionary in-place.
|
|
"""
|
|
|
|
for line in lines:
|
|
dm, pw, fname = linesplit(line)
|
|
dic["file"].append(fname)
|
|
dic["dm"].append(dm)
|
|
dic["pulseWidth"].append(pw)
|
|
|
|
injectedDic = {
|
|
"file" : [],
|
|
"dm" : [],
|
|
"pulseWidth" : []
|
|
}
|
|
|
|
detectedDic = {
|
|
"file" : [],
|
|
"dm" : [],
|
|
"pulseWidth" : []
|
|
}
|
|
|
|
#load injected data into dataframe
|
|
with open(INJECTION_FILE, "r") as file:
|
|
lines = file.readlines()
|
|
updateDic(lines, injectedDic)
|
|
|
|
injections = pd.DataFrame(data=injectedDic) #this is our main object for injection data
|
|
|
|
#load detection data into dataframe
|
|
with open(DETECTION_FILE, "r") as file:
|
|
lines = file.readlines()
|
|
updateDic(lines, detectedDic)
|
|
|
|
detections = pd.DataFrame(data=detectedDic) #this is our main object for detection data
|
|
|
|
#define summary printing for multiple steps
|
|
def summary(stage):
|
|
logger.info(f"Summary Stage {stage}")
|
|
logger.info(f"Number of files injected: {len(Counter(injections['file']))}")
|
|
logger.info(f"Number of files with detections: {len(Counter(detections['file']))}")
|
|
logger.info("=========================")
|
|
logger.info(f"Number of pulses injected: {len(injections['dm'])}")
|
|
logger.info(f"Number of pulses detected: {len(detections['dm'])}")
|
|
logger.info("=========================")
|
|
logger.info(f"File ratio: {round(len(Counter(detections['file']))/len(Counter(injections['file'])), 3)}")
|
|
logger.info("=========================")
|
|
|
|
#print initial summary
|
|
summary(1)
|
|
|
|
#let's track how many detections get removed by filtering
|
|
preFilterCount = len(detections['dm'])
|
|
|
|
#many files contained pulsar bursts, so we filter those out via DM
|
|
minDM = (10**2.5) * 0.95 #as per signal generation, plus a bit of wiggle room
|
|
logger.info(f"Filtering out pulsars (DM below {int(minDM)}...)")
|
|
detections = detections[detections['dm'] > minDM]
|
|
detections = detections.reset_index(drop=True)
|
|
summary(2)
|
|
|
|
#five files have SO MANY FALSE POSITIVES so get rid of them here?
|
|
remFiles = ["data_2025-04-30_07-53-07", "data_2025-05-01_07-47-34", "data_2025-04-24_07-36-04", "data_2025-04-29_07-50-16", "data_2025-04-30_08-18-17"]
|
|
logger.info("Filtering out the five problem files...")
|
|
remMask = [True] * len(detections['dm'])
|
|
for detection in detections.itertuples():
|
|
if detection.file in remFiles:
|
|
remMask[detection.Index] = False
|
|
detections = detections[remMask]
|
|
detections = detections.reset_index(drop=True)
|
|
summary(3)
|
|
|
|
postFilterCount = len(detections['dm'])
|
|
logger.info(f"Removed {preFilterCount-postFilterCount} detections by filtering the following:")
|
|
for file in remFiles:
|
|
logger.info(file+"_injected.fil")
|
|
|
|
#Let's do detection matching! Yaaaay!
|
|
#What detections line up to which injections? This will determine which ones got missed entirely.
|
|
#Define some kind of epsilon for DM and pulse width; if detection is within epsilon in DM we can match it.
|
|
dmEps = 5
|
|
#and define an auxiliary array of 0s for injections. List of detection counts!
|
|
matchCount = np.zeros(len(injections['dm']), dtype=int)
|
|
#also keep track of false positives:
|
|
falsePositiveMask = [False] * len(detections['dm'])
|
|
#Use queries to find matches
|
|
for detection in detections.itertuples():
|
|
qstring = (
|
|
f"(file == '{detection.file}') & "
|
|
f"((dm - @dmEps) < {detection.dm}) & "
|
|
f"((dm + @dmEps) > {detection.dm})"
|
|
)
|
|
matches = injections.query(qstring)
|
|
if len(matches) > 0:
|
|
logger.debug(f"Detection: DM {detection.dm} and PW {detection.pulseWidth}")
|
|
logger.debug(matches)
|
|
if len(matches) == 1:
|
|
i = matches.index[0]
|
|
matchCount[i] += 1
|
|
logger.debug("======")
|
|
elif len(matches) > 1:
|
|
raise ValueError("MULTIPLE MATCHES OHNO")
|
|
else: #no matching injection...
|
|
falsePositiveMask[detection.Index] = True
|
|
logger.debug(f"NO MATCH FOR: DM {detection.dm} and PW {detection.pulseWidth}")
|
|
logger.debug("Injections in file:")
|
|
logger.debug(injections.query(f"(file == '{detection.file}')"))
|
|
matchMaskInj = matchCount > 0
|
|
matchMaskDet = np.logical_not(falsePositiveMask)
|
|
missedMask = matchCount == 0
|
|
|
|
#So where are we?
|
|
#We have multiple datasets.
|
|
#1. Dataframe of all injected pulses. [injections]
|
|
#2. Dataframe of detections with pulsars filtered out. [detections]
|
|
#3. List of number of times each injection was detected [matchCount]
|
|
#4. A mask for only detected injections [matchMaskInj]
|
|
#5. A mask for only true positives [matchMaskDet]
|
|
#6. A mask for only missed injections [missedMask]
|
|
#7. A mask for false positives [falsePositiveMask]
|
|
|
|
logger.info(f"Successful detection ratio: {Counter(matchMaskInj)[True]/(len(injections['dm']))}")
|
|
|
|
#Let's try to figure out if certain files are responsible for the weird amount of false
|
|
#positives at around 10^2.7 pc/cc
|
|
#IT WAS 5 FILES. Filtered out above.
|
|
|
|
fpFileCounts = Counter(detections[falsePositiveMask]['file'])
|
|
sortedFPCounts = [(k, v) for k,v in sorted(fpFileCounts.items(), key=lambda value: -value[1])]
|
|
logger.info("False positive counts:")
|
|
for f, c in sortedFPCounts:
|
|
logger.info(f"{f}: {c}")
|
|
|
|
|
|
#Let's set a matplotlib default to make figures a bit bigger cos i like them
|
|
rcParams["figure.figsize"] = [7,5]
|
|
plt.rcParams['figure.constrained_layout.use'] = True
|
|
|
|
# #Injected pulses
|
|
# plt.figure(figsize=(7,10))
|
|
# allAx = plt.subplot(311)
|
|
# _, bins, _ = allAx.hist(np.log10(injections['dm']), bins=15)
|
|
# allAx.set_title("All injected pulses")
|
|
# detAx = plt.subplot(312, sharex=allAx, sharey=allAx)
|
|
# detAx.hist(np.log10(injections['dm'][matchMaskInj]), bins=bins)
|
|
# detAx.set_title("Detected pulses")
|
|
# misAx = plt.subplot(313, sharex=allAx, sharey=allAx)
|
|
# misAx.hist(np.log10(injections['dm'][missedMask]), bins=bins)
|
|
# misAx.set_title("Missed pulses")
|
|
# plt.ylabel("Count")
|
|
# plt.xlabel(r"DM (log pc cm$^3$)")
|
|
# plt.draw()
|
|
|
|
# #Detected pulses unstacked
|
|
# plt.figure(figsize=(7,10))
|
|
# allAx = plt.subplot(311)
|
|
# _, bins, _ = allAx.hist(np.log10(detections['dm']), bins=15)
|
|
# allAx.set_title("All detections")
|
|
# detAx = plt.subplot(312, sharex=allAx, sharey=allAx)
|
|
# detAx.hist(np.log10(detections['dm'][matchMaskDet]), bins=bins)
|
|
# detAx.set_title("Detected pulses")
|
|
# misAx = plt.subplot(313, sharex=allAx, sharey=allAx)
|
|
# misAx.hist(np.log10(detections['dm'][falsePositiveMask]), bins=bins)
|
|
# misAx.set_title("False positives")
|
|
# plt.ylabel("Count")
|
|
# plt.xlabel(r"DM (log pc cm$^3$)")
|
|
# plt.draw()
|
|
|
|
#Stacked histogram of injections
|
|
plt.figure(figsize=(7,10))
|
|
ax = plt.subplot(17,1,(1,8))
|
|
ax.hist([np.log10(injections['dm'][matchMaskInj]), np.log10(injections['dm'][missedMask])],
|
|
stacked=True, label=['Detected', 'Missed'], bins=15)
|
|
ax.yaxis.get_major_locator().set_params(integer=True)
|
|
ax.legend(loc='upper center')
|
|
ax.set_xlabel(r"DM (log pc cm$^{-3}$)")
|
|
ax.tick_params(labelbottom=True)
|
|
ax2 = plt.subplot(17,1,(9,16))
|
|
ax2.hist([np.log10(injections['pulseWidth'][matchMaskInj]), np.log10(injections['pulseWidth'][missedMask])],
|
|
stacked=True, label=['Detected', 'Missed'], bins=15)
|
|
ax2.yaxis.get_major_locator().set_params(integer=True)
|
|
ax2.set_xlabel(r"Pulse width (log s)")
|
|
plt.ylabel("Count")
|
|
plt.title("Injected pulses")
|
|
wordAx = plt.subplot(17,1,17)
|
|
wordAx.text(.3,.5,f"Overall detection rate: {round(Counter(matchMaskInj)[True]/(len(injections['dm']))*100,1)}%", size=14)
|
|
wordAx.set_axis_off()
|
|
plt.savefig(os.path.join("out","injections.png"))
|
|
|
|
#Stacked histogram of detections
|
|
plt.figure(figsize=(7,10))
|
|
ax = plt.subplot(17,1,(1,8))
|
|
ax.hist([np.log10(detections['dm'][matchMaskDet]), np.log10(detections['dm'][falsePositiveMask])],
|
|
stacked=True, label=['True detection', 'False positive'], bins=15)
|
|
ax.yaxis.get_major_locator().set_params(integer=True)
|
|
ax.set_xlabel(r"DM (log pc cm$^{-3}$)")
|
|
ax.legend(loc='upper center')
|
|
ax2 = plt.subplot(17,1,(9,16))
|
|
ax2.hist([np.log10(detections['pulseWidth'][matchMaskDet]), np.log10(detections['pulseWidth'][falsePositiveMask])],
|
|
stacked=True, label=['True detection', 'False positive'], bins=15)
|
|
ax2.yaxis.get_major_locator().set_params(integer=True)
|
|
ax2.set_xlabel("Pulse width (log s)")
|
|
plt.ylabel("Count")
|
|
plt.title("Detections")
|
|
wordAx = plt.subplot(17,1,17)
|
|
wordAx.text(.3,.5,f"False positive rate: {round(Counter(falsePositiveMask)[True]/(len(detections['dm']))*100,1)}%", size=14)
|
|
wordAx.set_axis_off()
|
|
plt.savefig(os.path.join("out","detections.png")) |