# plot.py
# Mark Rowan, School of Computer Science, University of Birmingham, UK
# April 2012
# Use NEURON / neurosim's grvec hoc routines to read saved spike data
# and then to allow manipulation and plotting in a Python+pyplot environment
# Run interactively, where <filepath> is a directory containing spks, .spks, aux and vars files:
# python plot.py <filepath>
# or in batch:
# python plot.py <filepath> [options] [plots]
# Options and plot types are listed at the bottom of this file, and can be
# viewed by running interactively (see above). A good default is 'all [noinhib]'.
# Each option and plot type should be separated by a colon, e.g:
# python plot.py <filepath> detail 4: activity 1-15: info noinhib
############## DEFINE ROUTINES ###############
# loadspks(filepath)
# takes name of file to be read and calls grvec read procedure to load file
def loadspks(filepath):
filename = filepath + "/spks"
if os.path.exists(filename):
h.grv_.read_vfile(filename) # open file using grvec
print "There are %d segments in this file" % numsegs()
else:
print "ERROR: No spks file found at path %s!" % filename
sys.exit()
# numsegs()
# returns number of data segments in the file
def numsegs():
return int(h.grv_.llist.count())
# loadvars(filepath)
# Reads variables saved in the vars file into memory
def loadvars(filepath):
filename = filepath + "/vars"
if os.path.exists(filename):
# Read the file into 'auxvars'
import imp
f = open(filename)
auxvars = imp.load_source('auxvars', '', f) # Read parameters (e.g. numcells) from aux file
f.close()
else:
print "ERROR: No vars file found at path %s!\n" % filename
sys.exit()
return auxvars
# getauxdata(filepath, cellID)
# Returns a list of all aux data items for a given cell ID, read via random access from the given aux file
# Note: does not sanity-check that requested cell ID < numcells!
# Note: all cells must be present and listed in order at each save step, with one 't = ...' line separator
# e.g. cell 0 must always be on line 1, numcells + 1 + 1, numcells * 2 + 1 + 1
def getauxdata(filepath, cell):
global detail # Declare that we want to use the global 'detail' value
filename = filepath + "/aux"
datalist = [] # Create empty list
if os.path.exists(filename):
lineno = 2 + cell # Start at line 2 (after "t = ..." line)
curline = "-1"
while curline != "":
# linecache.getline returns "" on read errors, e.g. eof
curline = linecache.getline(filename, lineno)
lineno += (auxvars.numcells + 1) * detail # +1 accounts for the "t = ..." lines between segments
splitline = curline.split(",")
# Check that this is correct cell before adding to list; exit otherwise!
if len(splitline) > 1:
if int(splitline[CELLID]) == cell:
datalist.append(splitline)
else:
print "ERROR: Expected cell %d, found cell %s at line %d!" % (cell, splitline[CELLID], lineno)
sys.exit()
else:
print "ERROR: No aux file found at path %s!\n" % filename
sys.exit()
return datalist;
# getIcellIDs(filepath)
# Returns a list of all cell IDs for I-cells (so they can be explicitly plotted / ignored)
def getIcellIDs(filepath):
cell_list = []
for i in range(auxvars.numcells):
celldata = getauxdata(filepath, i) # Read data
celltype = int(celldata[0][TYPE])
if (h.strm(h.CTYP.o(celltype).s,"^I")):
# Cell is I type
cell_list.append(int(celldata[0][CELLID]))
return cell_list
# getEcellIDs(filepath)
# Returns a list of all cell IDs for E-cells (so they can be explicitly plotted / ignored)
def getEcellIDs(filepath):
cell_list = []
for i in range(auxvars.numcells):
celldata = getauxdata(filepath, i) # Read data
celltype = int(celldata[0][TYPE])
if (h.strm(h.CTYP.o(celltype).s,"^I")) == False:
# Cell is E type
cell_list.append(int(celldata[0][CELLID]))
return cell_list
# read(index)
# takes index of file segment to be read (1 to numsegs)
# returns vec and tvec spiking data in a 2-D array
# data[0] is the spike time vector
# data[1] is the vector of cell IDs which spike at times in data[0]
def read(index):
print "Reading grvec data"
data = np.zeros( (1, 1) ) # Preassign zero-length data variable
# Check we're not requesting an index beyond the extent of the file
if index > numsegs():
print "Index must be <= %d" % (numsegs)
else:
h.grv_.rv_readvec(index, h.tvec, h.vec) # Read segment 'index' into vec/tvec
# Convert h.tvec and h.vec into Python / numpy arrays.
data = np.zeros( (2, h.vec.size()) ) # Preallocate for speed
for i in range(int(h.vec.size())):
data[0,i] = h.tvec.x[i]
data[1,i] = h.vec.x[i]
#print data.shape
return data
# Depending on the length of the time series of scale and activity graphs,
# return an x-axis delimited in seconds, hours, or days.
# Returns two values: a list with the x axis points, and a string for the units
def makexaxis(start, numpoints, interval, detail):
# Default to s
units = 0.001
unitstring = "seconds"
# If > 0.8e7 ms (8000s), use 'hours'
if start + numpoints * interval * detail > 0.8e7:
units = 0.001 / 60 / 60
unitstring = "hours"
# If > 0.8e8 ms (80 000s, 22 hrs), use 'days'
if start + numpoints * interval * detail > 0.8e8:
units = 0.001 / 60 / 60 / 24
unitstring = "days"
return np.arange(float(start*units), float((start + numpoints * interval * detail)*units), float((interval * detail)*units)), unitstring
############# DEFINE DIFFERENT PLOT TYPES #################
# raster
# basic spike raster (cell index vs time)
def raster(filepath, rasterfrom, rasterto):
global titleson # Declare that we want to use the global 'titleson' variable
rasterfrom = float(rasterfrom)
rasterto = float(rasterto)
if rasterto == rasterfrom:
# By default (when rasterto = 0), draw a separate plot for each data segment.
# If rasterfrom != rasterto, draw only a single plot for the data between rasterfrom and rasterto
# Create save directory if not already present
if (not os.path.isdir("%s/rasters" % filepath)):
os.mkdir("%s/rasters" % filepath)
for i in range(numsegs()):
print i
data = read(i) # Read data for segment i
pyplot.plot(data[0], data[1], '.', markersize=1, color='black')
pyplot.xlabel("Time (ms)")
pyplot.ylabel("Cell ID")
if titleson:
pyplot.suptitle(filepath)
pyplot.savefig("%s/rasters/%d.png" % (filepath, i), dpi=300, format="png") # Save at 300 DPI as 'filepath/raster/segment.png'
pyplot.hold(False) # Don't add next segment's data to current figure
pyplot.clf() # Clear figure for next plot
else:
segmentstart = int(rasterfrom / auxvars.buffertime)
segmentend = int(rasterto / auxvars.buffertime)
print "%d - %d" % (segmentstart, segmentend)
plotx = []
ploty = []
for i in range (segmentstart, segmentend + 1):
print "Reading segment %d" % i
data = read(i) # Read data for segment i
for j in range(len(data[0])):
if data[0,j] >= rasterfrom and data[0,j] <= rasterto:
# If data time is within requested bounds, add to plotdata
plotx.append(data[0,j])
ploty.append(data[1,j])
print "Plotting..."
pyplot.plot(plotx, ploty, '.', markersize=1, color='black')
pyplot.xlabel("Time (ms)")
pyplot.ylabel("Cell ID")
if titleson:
pyplot.suptitle(filepath)
pyplot.savefig("%s/raster.png" % filepath, dpi=300, format="png") # Save at 300 DPI
pyplot.clf() # Clear figure for next plot
# info
# Information contribution of each neuron in the network, per data segment
# Using mlabwrap to call Crumiller et al.'s MATLAB Fourier information calculation code
def info(filepath, cells, noinhib):
# Spike data is saved in alternating 'unique' and 'repeat' stimulation blocks
# Crumiller's code requires a N-by-M matrix (N = number of trials, M = number of cells),
# in which each element is a vector containing the spike times of cell M in trial N.
# This procedure takes the first n saved data segments (up to the auxvars.deletionstart time)
# and packs all the unique/repeat trials' spike times into separate CSV files
stimlength = auxvars.infotriallength # Length of 'repeat' or 'unique' stimulation block (ms)
buffertime = auxvars.buffertime # Length of whole data segment (ms)
#numtrials = int((auxvars.GetInfoDur / stimlength) / 2) # Number of 'unique' and 'repeat' trials
if cells == "":
# Make a list of cells whose data we want to write out
# (Either we requested all cells, or all-but-I cells)
if noinhib:
cells = getEcellIDs(filepath)
else:
cells = range(auxvars.numcells)
if numsegs() * buffertime < auxvars.infotriallength:
print "Not enough spike segments to obtain pre-deletion information (needs at least %d s)" % (auxvars.deletionstart/1000)
else:
uniquesfile = open("%s/info-uniques.csv" % filepath, 'w')
repeatsfile = open("%s/info-repeats.csv" % filepath, 'w')
for i in range(int(auxvars.GetInfoDur / buffertime)):
print "Scanning segment %d to add to CSV file" % i
# For each data segment in the file
data = read(i) # Read NEURON data for segment i
for index in range(len(data[0])):
spiketime = data[0, index] # What time was this cell spike?
cellID = data[1, index] # Which cell just fired?
if cellID in cells:
trialnum = int(spiketime / (stimlength * 2)) # Obtain trial number.
# Force int, as in Python 2 and 3 '/' operator means one or other of integer or float div...
# Find out whether we are in 1st stimlength ms (repeats), or last stimlength ms (uniques)
uniquestrial = ((spiketime % float(stimlength * 2)) / float(stimlength * 2)) > 0.5
# Tests (stimlength = 4000):
# spiketime = 0 => trialnum = 0, uniquestrial = false
# spiketime = 3999 => trialnum = 0, uniquestrial = false
# spiketime = 4000 => trialnum = 0, uniquestrial = true
# spiketime = 7999 => trialnum = 0, uniquestrial = true
# spiketime = 8000 => trialnum = 1, uniquestrial = false
# spiketime = 11999 => trialnum = 1, uniquestrial = false
# spiketime = 12000 => trialnum = 1, uniquestrial = true
# spiketime = 15999 => trialnum = 1, uniquestrial = true
# spiketime = 16000 => trialnum = 2, uniquestrial = false
# Remove offset so that each unique/repeat block has spikes beginning at 0ms
# Also div by 1000 to convert ms to s
spiketime = (spiketime % stimlength) / 1000
# Dump line to file
if uniquestrial:
uniquesfile.write("%d,%d,%f\n" % (cellID,trialnum,spiketime))
else:
repeatsfile.write("%d,%d,%f\n" % (cellID,trialnum,spiketime))
uniquesfile.close()
repeatsfile.close()
# Import mlabwrap
#from mlabwrap import mlab
# scale
# Scalefactors over time, read from the aux file
# 'cells' is a list of cells which should be plotted (or leave blank to plot all)
def scale(filepath, cells, noinhib):
print "Plotting scale factors..."
drawlegend = False
if len(cells) < 8 and len(cells) > 1:
drawlegend = True # Only draw a legend when examining 2-7 specific cells
plotavg = False
if cells == "":
plotavg = True
# Make a list of cells whose data we want to write out
# (Either we requested all cells, or all-but-I cells)
if noinhib:
cells = getEcellIDs(filepath)
else:
cells = range(auxvars.numcells)
celldata = getauxdata(filepath, 0) # Get data for an arbitrary cell, to obtain number of entries for pre-allocating array
numentries = len(celldata)
yaxis = np.zeros([len(cells), numentries]) # Collect cell data for plotting average values
#xaxis = range(auxvars.t_start, (auxvars.t_start + numentries * auxvars.recording_interval * detail), auxvars.recording_interval * detail) # Create x-axis
xaxis, xaxisunits = makexaxis(auxvars.t_start, numentries, auxvars.recording_interval, detail)
for cellnum in range(len(cells)):
# For each cell in the requested list
celldata = getauxdata(filepath, int(cells[cellnum])) # Get a list of all entries for this particular cell
# If the cell is not dead, add the data point to the y-axis
for entrynum in range(len(celldata)):
if int(celldata[entrynum][DEAD]):
yaxis[cellnum, entrynum] = -1
else:
yaxis[cellnum, entrynum] = float(celldata[entrynum][SCALE])
if plotavg:
yavg = np.zeros(numentries)
stdy = np.zeros(numentries)
for i in range(numentries):
yavg[i] = np.mean(yaxis[ yaxis[:,i]!=-1, i ] ) # Get avg scale for each non-dead time point
stdy[i] = np.std(yaxis[ yaxis[:,i]!=-1, i ] ) # Get std dev for each non-dead time point
pyplot.errorbar(xaxis, yavg, yerr=stdy, ecolor='grey', linestyle='-', marker='.', markersize=1.0)
else:
pyplot.plot(xaxis, yaxis) # Plot each cell individually
# Set axes limits to appropriate values
pyplot.ylim(ymin=0)
pyplot.xlim(xmax=xaxis[len(xaxis)-1], xmin=0) # Cut graph off at end of data, rather than leaving a margin until the next 'significant point'
# Draw labels
pyplot.xlabel("Time (%s)" % xaxisunits)
pyplot.ylabel("Scale factor")
if titleson:
pyplot.suptitle(filepath)
# Draw legend if only a small number of cells
if drawlegend:
pyplot.legend(cells, loc='upper right') # Draw legend
# Save at 300 DPI as 'filepath/scale.pdf'
pyplot.savefig("%s/scale.pdf" % filepath, dpi=300, format="pdf")
# Save processed plot data to a file for later analysis
np.savez("%s/scale" % filepath, x=xaxis, y=yavg, err=stdy)
pyplot.clf() # Clear figure for next plot
print "Done"
# activity
# Activity values over time, read from the aux file
# 'cells' is a list of cells which should be plotted (or leave blank to plot all)
def activity(filepath, cells, noinhib):
global titleson # Declare that we want to use the global 'titleson' variable
print "Plotting activity values..."
drawlegend = False
if len(cells) < 8 and len(cells) > 1:
drawlegend = True # Only draw a legend when examining 2-7 specific cells
plotavg = False
if cells == "":
plotavg = True
# Make a list of cells whose data we want to write out
# (Either we requested all cells, or all-but-I cells)
if noinhib:
cells = getEcellIDs(filepath)
else:
cells = range(auxvars.numcells)
celldata = getauxdata(filepath, 0) # Get data for an arbitrary cell, to obtain number of entries for pre-allocating array
numentries = len(celldata)
yaxis = np.zeros([len(cells), numentries]) # Collect cell data for plotting average values
#xaxis = range(auxvars.t_start, (auxvars.t_start + numentries * auxvars.recording_interval * detail), auxvars.recording_interval * detail) # Create x-axis
xaxis, xaxisunits = makexaxis(auxvars.t_start, numentries, auxvars.recording_interval, detail)
for cellnum in range(len(cells)):
# For each cell in the requested list
celldata = getauxdata(filepath, int(cells[cellnum])) # Get a list of all entries for this particular cell
# If the cell is not dead, add the data point to the y-axis
for entrynum in range(len(celldata)):
if int(celldata[entrynum][DEAD]):
yaxis[cellnum, entrynum] = -1
else:
yaxis[cellnum, entrynum] = float(celldata[entrynum][ACTIVITY]) * 1000
if plotavg:
yavg = np.zeros(numentries)
stdy = np.zeros(numentries)
for i in range(numentries):
yavg[i] = np.mean(yaxis[ yaxis[:,i]!=-1, i ] ) # Get avg activity for each non-dead time point
stdy[i] = np.std(yaxis[ yaxis[:,i]!=-1, i ] ) # Get std dev for each non-dead time point
pyplot.errorbar(xaxis, yavg, yerr=stdy, ecolor='grey', linestyle='-', marker='.', markersize=1.0)
else:
pyplot.plot(xaxis, yaxis) # Plot each cell individually
# Set axes limits to appropriate values
pyplot.ylim(ymin=0)
pyplot.xlim(xmax=xaxis[len(xaxis)-1], xmin=0) # Cut graph off at end of data, rather than leaving a margin until the next 'significant point'
# Draw labels
pyplot.xlabel("Time (%s)" % xaxisunits)
pyplot.ylabel("Activity (Hz)")
if titleson:
pyplot.suptitle(filepath)
# Draw legend if only a small number of cells
if drawlegend:
pyplot.legend(cells, loc='upper right') # Draw legend
# Save at 300 DPI as 'filepath/activity.pdf'
pyplot.savefig("%s/activity.pdf" % filepath, dpi=300, format="pdf")
# Save processed plot data to a file for later analysis
np.savez("%s/activity" % filepath, x=xaxis, y=yavg, err=stdy)
pyplot.clf() # Clear figure for next plot
print "Done"
# deletionscale
# Plots scaling factors of cells at time-of-deletion, vs total time
def deletionscale(filepath):
global titleson # Declare that we want to use the global 'titleson' variable
cells = range(auxvars.numcells)
celldata = getauxdata(filepath, 0) # Get data for an arbitrary cell, to obtain number of entries for pre-allocating array
numentries = len(celldata)
yaxis = []
xaxis = []
for cellnum in range(len(cells)):
# For each cell in the requested list
celldata = getauxdata(filepath, int(cells[cellnum])) # Get a list of all entries for this particular cell
# Find out at which point this cell died (if at all), and its scalefactor at this time
for entrynum in range(len(celldata)):
if int(celldata[entrynum][DEAD]) or entrynum >= len(celldata)-1:
# If cell has just died, or we've reached the end of the data...
cellscale = float(celldata[entrynum][SCALE]) # Get scalefactor
yaxis.append(cellscale) # Add scalefactor to y-axis
xaxis.append(auxvars.t_start + (entrynum * auxvars.recording_interval)) # Add time to x-axis
break # Stop processing
pyplot.plot(xaxis, yaxis, '.') # Plot each cell individually
# Set axes limits to appropriate values
pyplot.ylim(ymin=0)
pyplot.xlim(xmax=xaxis(len(xaxis)-1), xmin=0) # Cut graph off at end of data, rather than leaving a margin until the next 'significant point'
# Draw labels
pyplot.xlabel("Time (ms)")
pyplot.ylabel("Scale factor at time of deletion")
if titleson:
pyplot.suptitle(filepath)
# Save at 300 DPI as 'filepath/deletionscale.pdf'
pyplot.savefig("%s/deletionscale.pdf" % filepath, dpi=300, format="pdf")
# Save processed plot data to a file for later analysis
np.savez("%s/deletionscale" % filepath, x=xaxis, y=yaxis)
pyplot.clf() # Clear figure for next plot
print "Done"
# power
# Plots power spectra of the network separately for E and I cells, obtained from the neural spike trains
def power(filepath):
binsize = 5.0 # ms for bin size (5ms = 200Hz sampling rate)
Icells = getIcellIDs(filepath) # Make list of I cells
# Load each segment of saved spike data
for i in range(numsegs()):
# For each data segment in the file
data = read(i) # Read NEURON data for segment i
# Create empty numpy array for number of spikes-per-bin
firstspk = data[0,0]
lastspk = data[0,len(data[1])-1]
timecovered = lastspk-firstspk
numbins = int(timecovered / binsize)
spikesperbinI = np.zeros( numbins )
spikesperbinE = np.zeros( numbins )
# Make MUA time series vector by counting all population spikes during each 'binsize' ms
print "Binning file segment %d into %d bins (window size = %f ms)" % (i, numbins, binsize)
for j in range(len(data[1])):
spiketime = data[0,j]
bin = int(float((spiketime % (numbins * binsize)) / binsize)) # Find bin number into which this spike should go
cellID = data[1,j]
if cellID in Icells:
# Check if this is an I or E cell
spikesperbinI[bin] = spikesperbinI[bin] + 1
else:
spikesperbinE[bin] = spikesperbinE[bin] + 1
# Find and subtract mean from the MUA vector
meanmuaI = float(np.sum(spikesperbinI) / len(spikesperbinI))
spikesperbinI = spikesperbinI - meanmuaI
meanmuaE = float(np.sum(spikesperbinE) / len(spikesperbinE))
spikesperbinE = spikesperbinE - meanmuaE
# Run MTSpec's spectral analysis on the MUA time series
# MTSpec documentation: http://krischer.github.com/mtspec/mtspec.multitaper.mtspec.html#mtspec.multitaper.mtspec
delta = 1.0 / float(1000.0/binsize) # 1 / sample rate (Hz)
time_bandwidth = 4
nfft = None
number_of_tapers = None
quadratic = False
adaptive = True
verbose = False
optional_output = False
statistics = False
rshape = False
fcrit = False
print "Running mtspec for I cells"
#spec, freq = multitaper.mtspec(spikesperbin, delta, time_bandwidth, nfft, quadratic, adaptive, verbose, optional_output, statistics, rshape, fcrit)
specI, freqI = mtspec(spikesperbinI, delta, time_bandwidth)
print "Running mtspec for E cells"
specE, freqE = mtspec(spikesperbinE, delta, time_bandwidth)
# Save to file for further examination
#numpy.savetxt("%s/power-%d-muaE.txt" % (filepath, i), spikesperbinE)
#numpy.savetxt("%s/power-%d-specE.txt" % (filepath, i), specE)
#numpy.savetxt("%s/power-%d-freqE.txt" % (filepath, i), freqE)
print "Generating plots\n"
# Create save directory if not already present
if (not os.path.isdir("%s/power" % filepath)):
os.mkdir("%s/power" % filepath)
# I power
# Draw labels
pyplot.xlabel("Frequency (Hz)")
pyplot.ylabel("Normalised power")
if titleson:
pyplot.suptitle(filepath)
pyplot.plot(freqI,specI)
# Set axes limits to appropriate values
pyplot.ylim(ymax=0.08, ymin=0)
pyplot.xlim(xmax=100, xmin=0)
pyplot.savefig("%s/power/I-%d.pdf" % (filepath, i), dpi=300, format="pdf")
pyplot.clf()
# E power
# Draw labels
pyplot.xlabel("Frequency (Hz)")
pyplot.ylabel("Normalised power")
if titleson:
pyplot.suptitle(filepath)
pyplot.plot(freqE,specE)
# Set axes limits to appropriate values
pyplot.ylim(ymax=0.08, ymin=0)
pyplot.xlim(xmax=100, xmin=0)
pyplot.savefig("%s/power/E-%d.pdf" % (filepath, i), dpi=300, format="pdf")
pyplot.clf()
# Save processed plot data to a file for later analysis
np.savez("%s/power/I-%d.npz" % (filepath, i), x=freqI, y=specI)
np.savez("%s/power/E-%d.npz" % (filepath, i), x=freqE, y=specE)
print "Done"
################# COMMAND LINE PROCESSING #################
def processargs(args):
# Checks list of arguments, and returns a list of cells from either a comma-separated list or
# a hyphenated range (e.g. 0,1,3,7,9 or 0-10)
if len(args.split()) > 3:
print "\n\nERROR: Too many arguments! Permitted: 'noinhib', 'cell1,cell2,..,celln', or '0-100'"
if len(args.split()) == 1 or (len(args.split()) == 2 and 'noinhib' in args):
# No arguments (or just 'noinhib') supplied
return "" # Return an empty list
else:
# A cell / cell range / time range argument was supplied...
args = args.split()
args.pop(0) # Remove command name from the front of the list of args
if "-" in args[0]:
# We're dealing with a request for a range of cells
args = args[0].split("-") # Split dash-separated range
cellslist = []
for i in range(int(args[0]), int(args[1])+1):
cellslist.append(i)
return cellslist
else:
args = args[0].split(",") # Split comma-separated list
return args
def makegraph(graphtype):
if "noinhib" in graphtype:
noinhib = True
else:
noinhib = False
command = graphtype.split()[0]
if command == "quit":
sys.exit()
elif command == "raster":
if len(graphtype.split()) == 2:
rasterrange = graphtype.split()[1]
rasterrange = rasterrange.split('-')
rasterfrom = rasterrange[0]
rasterto = rasterrange[1]
else:
rasterfrom = 0
rasterto = 0
raster(filepath, rasterfrom, rasterto)
elif command == "info":
info(filepath, processargs(graphtype), noinhib)
elif command == "scale":
scale(filepath, processargs(graphtype), noinhib)
elif command == "activity":
activity(filepath, processargs(graphtype), noinhib)
elif command == "power":
power(filepath)
elif command == "deletionscale":
deletionscale(filepath)
elif command == "all":
scale(filepath, "", noinhib)
activity(filepath, "", noinhib)
deletionscale(filepath)
raster(filepath, 0, 0)
info(filepath, "", noinhib)
power(filepath)
elif command == "titles":
global titleson
titleson = not titleson
print "\n\n\nTitles on: %s" % titleson
elif command == "detail":
global detail
detail = int(graphtype.split()[1])
print "\nSetting detail level to %d" % detail
else:
print "\nUnknown graph type or option '%s'" % command
############### MAIN METHOD ################
# Imports
import sys
import readline
import numpy as np
import string
import os
import linecache
import matplotlib
matplotlib.use('agg') # Prevent pyplot from trying to open a graphical front-end on headless clients
from matplotlib import pyplot
from mtspec import * # Multitaper spectral analysis (MTSpec library from http://pypi.python.org/pypi/mtspec)
# Check filename was supplied
if len(sys.argv) < 2:
print "Usage:\npython plot.py <data dir path> [option arg:] [plot_type args:] [plot_type args:] [...:]"
print "For a list of plot types and arguments, run interactively (i.e. python plot.py <data dir>)"
sys.exit()
if len(sys.argv) >= 3:
plot_as_arg = True
args = sys.argv[2:] # Get all plot type args
graphtypelist = ""
for item in args:
graphtypelist += item + " " # Concatenate the args string back together (including a space between each item)
else:
plot_as_arg = False
# Handle NEURON imports
print "Loading NEURON/grvec routines... \n\n"
from neuron import h
# Load grvec / intfzip simulation
h('xopen("setup.hoc")')
h('xopen("nrnoc.hoc")')
h('load_file("init.hoc")')
h('load_file("grvec.hoc")')
h('load_file("labels.hoc")')
# Load file
global auxvars # Vars from the vars file are held here, to be accessible by all methods
# Set index numbers for the aux file data lines (format: CELL,TYPE,ACTIVITY,TARGET,SCALE,DEAD)
global CELLID
CELLID = 0
global TYPE
TYPE = 1
global ACTIVITY
ACTIVITY = 2
global TARGET
TARGET = 3
global SCALE
SCALE = 4
global DEAD
DEAD = 5
global titleson
titleson = False
global detail
detail = 1 # Take every nth point for plotting (set == 1 to plot every point, or > 1 to reduce points)
filepath = sys.argv[1]
print "\nLoading data from %s" % filepath
loadspks(filepath)
auxvars = loadvars(filepath)
print "Done."
if plot_as_arg:
for graphtype in graphtypelist.split(":"):
# Make all the requested graph types, then quit
makegraph(graphtype)
sys.exit()
else:
while 1:
# Get user's graph preference interactively
print "\n\n\n"
print filepath
print "titles toggle graph titles (use 'off' for printing; default off)"
print "detail set level of detail for plots (every 'n'th point; default=1)"
print "\n"
print "all [noinhib] draw all plots [excluding inhibitory cells]"
print "raster [starttime-endtime] basic spike raster [between given ms times; default=ALL]"
print "info [cells | noinhib] information contribution [comma-separated list of individual cells or range of cells e.g. 0-20 | without inhibitory cells]"
print "scale [cell1s | noinhib] scale factor values [comma-separated list of individual cells or range of cells e.g. 0-20 | without inhibitory cells]"
print "activity [cells | noinhib] activity values [comma-separated list of individual cells or range of cells e.g. 0-20 | without inhibitory cells]"
print "deletionscale scalefactors of cells at time-of-deletion, plotted against total time"
print "power oscillatory power spectra"
print "quit exit program"
graphtype = raw_input("\ne.g. \"raster 100000-150000\", \"scale 5,23,190\", \"info 0-49\", or \"activity no-inhib\"\n>")
makegraph(graphtype)