def main(hdf, stat, name):
""" Given a <path> and the <hdf> name, plot and save all the models in
the <hdf>, prefixing each with <basename>.
"""
stat = str(stat)
# Create a csv file to tabulate into
f = open('{0}_hist_{1}.csv'.format(name, stat), 'w')
csvw = csv.writer(f)
# Make a header for the table
header = [stat, "count", "cond", "boldmeta", "model"]
csvw.writerow(header)
# Make a list of the models
# to tabulate and get going...
models = get_model_names(hdf)
for mod in models:
meta = get_model_meta(hdf, mod)
hist_list = create_hist(hdf, mod, stat)
for hist in hist_list:
cond = hist.name
boldmeta = "_".join([str(b) for b in meta["bold"]])
[csvw.writerow([k, v, cond, boldmeta, mod]) for
k, v in hist.h.items()]
f.close()
def main(hdf, name):
# Create a csv file to tabulate into
f = open('{0}.csv'.format(name), 'w')
csvw = csv.writer(f)
# Make a header for the table
header = ["x", "count", "cond"]
csvw.writerow(header)
conds = set(["acc", "value", "rpe", "p", "rand", "box"])
# Find a dmcol from models,
# pick the first model with
# that dm cond. Conds across
# models are the same.
locations = {}
models = get_model_names(hdf)
for model in models:
# Get model meta data and compare it too conds
meta = get_model_meta(hdf, model)
for ii, dmname in enumerate(meta["dm"]):
if dmname in conds:
locations[dmname] = (model, ii)
conds.remove(dmname)
## conds loses elements!
# If conds is empty, stop
if not conds:
print("All conds found. {0}".format(model))
break
# Get each dm's data and pick a col using pos;
# pos matches cond.
dmdata = {}
for cond, loci in locations.items():
print("Getting {0}".format(cond))
model, pos = loci
dm = np.array(read_hdf(hdf, '/' + model + '/dm'))
dmdata[cond] = dm[:,:,pos].flatten()
# Create a histogram then write it out.
for cond, data in dmdata.items():
print("Histogramming {0}".format(cond))
# Instantiate a RHist instance and
# use it to make a histogram
hist = RHist(name=cond, decimals=2)
[hist.add(x) for x in data]
# Tell that textfile a tale.
[csvw.writerow([k, v, cond]) for k, v in hist.h.items()]
f.close()
def random_timecourses(hdf, model, N, nsim, name=None):
""" Plot <N> randomly selected BOLD and design matrix timecourses
from <nsim> options for <model> from <hdf>.
If <name> is not None, the results are saved as a pdf. """
# Open a handle the hdf file
f = h5py.File(hdf, 'r')
# Create (or use) a handle to a multi-page pdf
if name != None:
if isinstance(name, PdfPages):
pdf = name ## Use
else:
pdf = PdfPages('{0}.pdf'.format(name)) ## Create
# Randomly select sims between 0-nsim
# And mae seperate plots for each
# adding them to the same pdf.
selected = np.random.randint(0, nsim, N)
for sel in selected:
# Get metadata
meta = get_model_meta(hdf, model)
# Create a figure window
fig = plt.figure()
ax = fig.add_subplot(111)
# Get the data
dm = f[os.path.join("/", str(sel), model, "dm")].value
bold = f[os.path.join("/", str(sel), model, "bold")].value
# PLot it
ax.plot(dm)
ax.plot(bold, color="grey")
# Pretty up the plot
boldleg = [
"".join([
"BOLD: ",
] + meta["bold"].tolist()),
]
plt.legend(meta["dm"] + boldleg)
plt.title('Timecourse {0} from {1}.'.format(sel, model))
## TODO add labels?
if name != None:
# Add to pdf...
plt.savefig(pdf, format="pdf")
if name != None:
pdf.close()
def main(hdf, name):
f = open('{0}.csv'.format(name), 'w')
csvw = csv.writer(f)
# A header for the table
head = ["tau", "p_tau",
"r", "p_r",
"rho", "p_rho",
"cond",
"boldmeta",
"model"]
csvw.writerow(head)
models = get_model_names(hdf)
for model in models:
# Get model meta data
meta = get_model_meta(hdf, model)
boldmeta = "_".join([str(b) for b in meta["bold"]])
## meta['bold'] can be a list
## but we need a string....
# Get all the design matrices and
# Get all the bold signals as a 1d array
dm = np.array(read_hdf(hdf, '/' + model + '/dm'))
bold = np.array(read_hdf(hdf, '/' + model + '/bold')).flatten()
# Loop over the dm cols in the dm:
# axis 1 is a sim index/count,
# axis 2 is the dm row
# axis 3 is the dm cols
for jj in range(dm.shape[2]):
# Get all this cols in a 1d array
# matching what was done to
# bold above
x1 = dm[:, :, jj].flatten()
x1name = meta["dm"][jj]
# And calc the corr between
# the two 1d arrays
tau, p_tau = kendalltau(bold, x1)
r, p_r = pearsonr(bold, x1)
rho, p_rho = spearmanr(bold, x1)
# then write it all out.
csvw.writerow([
tau, p_tau,
r, p_r,
rho, p_rho,
x1name,
boldmeta,
model])
f.close()
def hist_t(hdf, model, name=None):
"""
Plot histograms of the t values in <hdf> for each condition in
<model>.
If <name> is not None the plot is saved as <name>.pdf.
"""
meta = get_model_meta(hdf, model)
hist_list = []
for dm_col in meta['dm']:
# Make an instance RHist for the list.
hist = RHist(name=dm_col, decimals=1)
hist_list.append(hist)
# read_hdf_inc returns a generator so....
tdata = read_hdf_inc(hdf, '/' + model + '/t')
for ts in tdata:
# get the tvals for each instance of model
# and add them to the hist_list,
[hist_list[ii].add(ts[ii]) for ii in range(len(ts) - 1)]
## The last t in ts is the constant, which we
## do not want to plot.
# Create a fig, loop over the hist_list
# plotting each on fig.axes = 0.
fig = plt.figure()
fig.add_subplot(111)
colors = itertools.cycle(
['DarkGray', 'DarkBlue', 'DarkGreen', 'MediumSeaGreen'])
## Using html colors...
[h.plot(fig=fig, color=colors.next(), norm=True) for h in hist_list]
# Prettify the plot
ax = fig.axes[0]
ax.set_xlabel('t-values')
ax.set_ylabel('P(t)')
# Add vetical lines representing significance tresholds
ax.axvline(x=1.7822, label='p < 0.05', color='red', linewidth=4)
ax.axvline(x=2.6810, label='p < 0.01', color='red', linewidth=3)
ax.axvline(x=3.0545, label='p < 0.005', color='red', linewidth=2)
ax.axvline(x=4.3178, label='p < 0.0005', color='red', linewidth=1)
## tval lines assume N=12 subjects
plt.xlim(-10, 15)
plt.legend()
plt.title('{0} -- BOLD: {1}'.format(model, meta['bold']))
if name != None:
plt.savefig(name, format="pdf")
def hist_t(hdf, model, name=None):
"""
Plot histograms of the t values in <hdf> for each condition in
<model>.
If <name> is not None the plot is saved as <name>.pdf.
"""
meta = get_model_meta(hdf, model)
hist_list = []
for dm_col in meta["dm"]:
# Make an instance RHist for the list.
hist = RHist(name=dm_col, decimals=1)
hist_list.append(hist)
# read_hdf_inc returns a generator so....
tdata = read_hdf_inc(hdf, "/" + model + "/t")
for ts in tdata:
# get the tvals for each instance of model
# and add them to the hist_list,
[hist_list[ii].add(ts[ii]) for ii in range(len(ts) - 1)]
## The last t in ts is the constant, which we
## do not want to plot.
# Create a fig, loop over the hist_list
# plotting each on fig.axes = 0.
fig = plt.figure()
fig.add_subplot(111)
colors = itertools.cycle(["DarkGray", "DarkBlue", "DarkGreen", "MediumSeaGreen"])
## Using html colors...
[h.plot(fig=fig, color=colors.next(), norm=True) for h in hist_list]
# Prettify the plot
ax = fig.axes[0]
ax.set_xlabel("t-values")
ax.set_ylabel("P(t)")
# Add vetical lines representing significance tresholds
ax.axvline(x=1.7822, label="p < 0.05", color="red", linewidth=4)
ax.axvline(x=2.6810, label="p < 0.01", color="red", linewidth=3)
ax.axvline(x=3.0545, label="p < 0.005", color="red", linewidth=2)
ax.axvline(x=4.3178, label="p < 0.0005", color="red", linewidth=1)
## tval lines assume N=12 subjects
plt.xlim(-10, 15)
plt.legend()
plt.title("{0} -- BOLD: {1}".format(model, meta["bold"]))
if name != None:
plt.savefig(name, format="pdf")
def main(posargs):
# Check then handle the positional arguements
# brough in from the command line
if len(posargs) < 3:
raise ValueError("At least three arguments are required.\n")
name = posargs.pop()
stat = posargs[0]
tocompare = posargs[1:]
# Create a csv file to tabulate into
f = open('{0}_{1}.csv'.format(name, stat), 'w')
csvw = csv.writer(f)
csvw.writerow(["mean", "sd", "D", "dataset", "model", "boldmeta",
"dmmeta", "cond"])
models = get_model_names(tocompare[0]) ## Assume models are identical for
## each hdf to compare
for model in models:
for ii, comp in enumerate(tocompare):
meta = get_model_meta(comp, model)
boldmeta = "_".join([str(b) for b in meta["bold"]])
dmmeta = "_".join([str(d) for d in meta["dm"]])
# Generate the data to add to the table.
hist_list = create_hist_list(comp, model, stat)
means = [hist.mean() for hist in hist_list]
stdevs = [hist.stdev() for hist in hist_list]
names = [hist.name for hist in hist_list]
# Calculate effect sizes
ns = [hist.n() for hist in hist_list]
k = len(hist_list) + 1 ## Number of predcitors
## +1 for the dummy
cohen_ds = [(2.0 * mean) / np.sqrt(n - k - 1.0) for
mean, n in zip(means, ns)]
## d = 2*t / sqrt(DF)
## DF = n - k - 1
## n = sample number
## k = predictor number
# And add it.
databasename = os.path.splitext(os.path.basename(comp))[0]
for mean, sd, d, name in zip(means, stdevs, cohen_ds, names):
row = [mean, sd, d, databasename, model, boldmeta, dmmeta, name]
csvw.writerow(row)
f.close()
def random_timecourses(hdf, model, N, nsim, name=None):
""" Plot <N> randomly selected BOLD and design matrix timecourses
from <nsim> options for <model> from <hdf>.
If <name> is not None, the results are saved as a pdf. """
# Open a handle the hdf file
f = h5py.File(hdf, "r")
# Create (or use) a handle to a multi-page pdf
if name != None:
if isinstance(name, PdfPages):
pdf = name ## Use
else:
pdf = PdfPages("{0}.pdf".format(name)) ## Create
# Randomly select sims between 0-nsim
# And mae seperate plots for each
# adding them to the same pdf.
selected = np.random.randint(0, nsim, N)
for sel in selected:
# Get metadata
meta = get_model_meta(hdf, model)
# Create a figure window
fig = plt.figure()
ax = fig.add_subplot(111)
# Get the data
dm = f[os.path.join("/", str(sel), model, "dm")].value
bold = f[os.path.join("/", str(sel), model, "bold")].value
# PLot it
ax.plot(dm)
ax.plot(bold, color="grey")
# Pretty up the plot
boldleg = ["".join(["BOLD: "] + meta["bold"].tolist())]
plt.legend(meta["dm"] + boldleg)
plt.title("Timecourse {0} from {1}.".format(sel, model))
## TODO add labels?
if name != None:
# Add to pdf...
plt.savefig(pdf, format="pdf")
if name != None:
pdf.close()
def create_hist_list(hdf, model, stat):
""" Create a list of Rhist (histogram) objects for <model> and
<stat> in the given <hdf>.
If <stat> has only one entry (as is the case for 'aic') the list will have
only one entry. If however <stat> has n entries per model (like't')
the list will have n-1 entries. As n matches the number of columns in
the design matrix, the rightmost will always correspond to the dummy
predictor and is therefore discarded. """
hist_list = [] ## A list of RHist objects.
meta = get_model_meta(hdf, model) ## metadata for naming
# A handle on the hdf data
hdfdata = read_hdf(hdf, '/' + model + '/' + stat)
# Loop over the nodes, adding the data
# for each to a RHist.
for node in hdfdata:
# Some data will be list-like
# so try to iterate, if that fails
# assume the data is a single number
try:
for ii in range(len(node) - 1):
# Init entries in hist_list as needed
try:
hist_list[ii].add(node[ii])
except IndexError:
hist_list.append(RHist(name=meta['dm'][ii], decimals=2))
hist_list[ii].add(node[ii])
except TypeError:
# Assume a number so hist_list has only one
# entry (i.e. 0).
#
# Init entries in hist_list as needed
try:
hist_list[0].add(node)
except IndexError:
hist_list.append(RHist(name=stat, decimals=2))
hist_list[0].add(node)
return hist_list
def create_hist_list(hdf, model, stat):
""" Create a list of Rhist (histogram) objects for <model> and
<stat> in the given <hdf>.
If <stat> has only one entry (as is the case for 'aic') the list will have
only one entry. If however <stat> has n entries per model (like't')
the list will have n-1 entries. As n matches the number of columns in
the design matrix, the rightmost will always correspond to the dummy
predictor and is therefore discarded. """
hist_list = [] ## A list of RHist objects.
meta = get_model_meta(hdf, model) ## metadata for naming
# A handle on the hdf data
hdfdata = read_hdf(hdf, '/' + model + '/' + stat)
# Loop over the nodes, adding the data
# for each to a RHist.
for node in hdfdata:
# Some data will be list-like
# so try to iterate, if that fails
# assume the data is a single number
try:
for ii in range(len(node)-1):
# Init entries in hist_list as needed
try:
hist_list[ii].add(node[ii])
except IndexError:
hist_list.append(RHist(name=meta['dm'][ii], decimals=2))
hist_list[ii].add(node[ii])
except TypeError:
# Assume a number so hist_list has only one
# entry (i.e. 0).
#
# Init entries in hist_list as needed
try:
hist_list[0].add(node)
except IndexError:
hist_list.append(RHist(name=stat, decimals=2))
hist_list[0].add(node)
return hist_list
def main(posargs):
# Check then handle the positional arguements
# brough in from the command line
if len(posargs) != 4:
raise ValueError("Four arguments are required.\n")
hdf = posargs[0]
stat = posargs[1]
criterion = float(posargs[2])
name = posargs[3]
# Create csv writer named name
# And give it a header
f = open("{0}_{1}{2}.csv".format(name, stat, criterion), "w")
csvw = csv.writer(f)
csvw.writerow(["area", "model", "boldmeta", "dmmeta", "cond"])
models = get_model_names(hdf)
for model in models:
hist_list = create_hist_list(hdf, model, stat)
# Loop over the hist_list adding
# the results from each hist.above(criterion)
# to a bar plot.
areas = [hist.above(criterion) for hist in hist_list]
names = [hist.name for hist in hist_list]
# Pretty things up then save
# this barplot to the pdf
# and move onto the next model
meta = get_model_meta(hdf, model)
boldmeta = "_".join([str(b) for b in meta["bold"]])
dmmeta = "_".join([str(d) for d in meta["dm"]])
for area, name in zip(areas, names):
row = [area, model, boldmeta, dmmeta, name]
csvw.writerow(row)
f.close()
def create_hist(hdf, model, stat):
"""
Create histograms of stat values in <hdf> for each condition in
<model>.
"""
meta = get_model_meta(hdf, model)
hist_list = []
for dm_col in meta['dm']:
# Make an instance RHist for the list.
hist = RHist(name=dm_col, decimals=1)
hist_list.append(hist)
# read_hdf_inc returns a generator so....
data = read_hdf_inc(hdf,'/'+ model + '/' + stat)
for row in data:
# get the row values for each instance of model
# and add them to the hist_list,
[hist_list[ii].add(row[ii]) for ii in range(len(row)-1)]
## The last t in ts is the constant, which we
## do not want to plot.
return hist_list
def main(hdf, name):
# Create a csv file to tabulate into
f = open('{0}.csv'.format(name), 'w')
csvw = csv.writer(f)
# A header for the table
head = ["tau", "p_tau",
"r", "p_r",
"rho", "p_rho",
"cond0",
"cond1"]
csvw.writerow(head)
conds = set(["acc", "value", "rpe", "p", "rand", "box"])
pairs = permutations(conds, 2)
# Find a dmcol from models,
# pick the first model with
# that dm cond. Conds across
# models are the same.
locations = {}
models = get_model_names(hdf)
for model in models:
# Get model meta data and compare it too conds
meta = get_model_meta(hdf, model)
for ii, dmname in enumerate(meta["dm"]):
if dmname in conds:
locations[dmname] = (model, ii)
conds.remove(dmname)
## conds loses elements!
# If conds is empty, stop
if not conds:
print("All conds found by {0}".format(model))
break
# Get the data
dmdata = {}
for cond, loci in locations.items():
print("Getting {0}".format(cond))
model, pos = loci
dm = np.array(read_hdf(hdf, '/' + model + '/dm'))
## Get all the dm's data
dmdata[cond] = dm[:,:,pos].flatten()
## pick a col using pos, as this is
## the data matching name
# Calculate all pair-wise correlations
for pair in pairs:
print("Correlate: {0}".format(pair))
cond0, cond1 = pair
x0 = dmdata[cond0]
x1 = dmdata[cond1]
tau, p_tau = kendalltau(x0, x1)
r, p_r = pearsonr(x0, x1)
rho, p_rho = spearmanr(x0, x1)
# then write it all out.
csvw.writerow([
tau, p_tau,
r, p_r,
rho, p_rho,
cond0,
cond1])
f.close()
