def compare_correlated_summary_stats(params, syn_current, pair, corr_mat, mag_of_change, selected_indices, using="C++", mode="auto", summary_func=None):
"""Simulates a reference and a correlated trace. Summarises both and returns an DataFrame
containing the selected summary statistics.
Parameters
----------
params : torch.Tensor
A 1D parameter tensor that holds the values of input parameters to the simulator model.
Contains the initial current "V0" [float], the stimulation current trace
"It" [ndarray], the product of the stimulation current trace and the area
of the soma "ItA": ndarray, the time axis "ts": ndarray, the step size
"dt": float and the simulus time course "StimOnset": float,
"StimEnd": float and "StimDur": float.
pair : tuple, list, numpy.ndarray
Specifies which correlation axis to change the parameters along.
corr_mat : torch.Tensor
Matrix containing the pairwise correlation coefficients of the different parameters.
mag_of_change : float
Value from the intervall of [-k, k], where k \in R that specifies by how much the
input parameters are altered.
selected_stats : None, tuple, list or array
Indices of summary stats to be used.
sumary_func : function or str
A function that takes voltage trace V(t), current trace I(t), time
points t and time step dt and outputs a dictionary with summary stats
and descriptions.
Alternatively a descriptive string like "v1" or "v2" can be provided to
use one of the summary methods already implemented as part of Trace().
mode : str
Specifies whether cell/experiment specific parameters like ASoma and V0
should be specified as part of the model parameters ("model_params")
or if ASoma and V0 should be considered part of the syn_current
or trace object ("trace").
Returns
-------
sum_df : pandas.DataFrame
Contains the summary statistics for both the base and correlated trace."""
# generate correlated parameters
pars = generate_correlated_parameters(params, corr_mat, pair=pair, mag_of_change=mag_of_change)
# run simulation with reference and correlated params
corr_trace = runHH(pars, syn_current=syn_current, using=using, mode=mode)
base_trace = runHH(params, syn_current=syn_current, using=using, mode=mode)
# collect results of summaries in DataFrame
sum_corr_trace_df = pd.Series(
corr_trace.summarise(selected_indices, summary_func=summary_func), name="correlated params")
sum_base_trace_df = pd.Series(base_trace.summarise(selected_indices,summary_func=summary_func), name="base params")
sum_df = pd.concat([sum_base_trace_df,sum_corr_trace_df], axis=1)
return sum_df
def plot_best_matches(thetas=None, stats_sim=None, stats_obs=None, simulated_traces=None,
trace_obs=None, samplesize=10, figsize=(10, 14), savefig=False,
fig_name="default", selected_stats=None, metric="MSE", mode="auto"):
"""Plot the best matching voltage traces compared with the observed trace,
based on the mean squared error of the summary statistics.
Parameters
----------
thetas : torch.tensor or numpy.ndarray
Holds the parameters that go along with stats_sim.
stats_sim = pandas.DataFrame
Contains the summarised simulation results.
stats_obs = pandas.DataFrame
Contains the summarised observation results.
simulated_traces : list[Trace]
The simulated data of the voltage traces in the form of Trace() objects.
trace_obs : Trace
Trace object for the observed voltage trace
selected_stats : None, tuple, list or array
Indices of summary stats to be used
samplesize : int
How many simulated traces to compare.
figsize : tuple, list or ndarray
Changes the figure size of the plot.
timewindow: tuple, list or ndarray
Contains start and end time of voltage traces that will be shown.
savefig : bool
Determines whether or not to save the plotted figure.
fig_name : str
Integrates figure name into the filename of the saved figure.
selected_stats : None, tuple, list or array
Indices of summary stats to be used.
metric : str
Determines which metric to sort the samples by. Currently "MSE"
and "std" are supported.
mode : str
Specifies whether cell/experiment specific parameters like ASoma and V0
should be specified as part of the model parameters ("model_params")
or if ASoma and V0 should be considered part of the syn_current
or trace object ("trace")."""
mins_idx, mins = best_matches(stats_sim, stats_obs, simulated_traces, trace_obs, selected_stats, metric)
if type(simulated_traces) == type(None):
if type(thetas) == pd.core.frame.DataFrame:
thetas = torch.tensor(thetas.values[mins_idx[:samplesize]])
simulated_traces = runHH(thetas, trace_obs, mode=mode)
if trace_obs == None:
raise ValueError("No observed trace was provided. Please provide Trace().")
plot_comparison(simulated_traces, trace_obs, samplesize, figsize=(10, 14))
if savefig:
plt.savefig('../data/figures/{}_8paramprior_support_best_sims_v_obs.png'.format(fig_name))
def show_correlated_traces(params, syn_current, corr_mat, pair=(0,1), start=0.01, stop=0.5,
N=5, figsize=(15,10), timewindow=[0.3,0.33], compare_changes=False,
savefig=False, fig_name="default",
mode="auto"):
"""Plots the effects of changing correlated parameters, for one parameter pair,
on the summary stats of the simulated voltage traces for different magnitudes of change.
Parameters
----------
params : torch.Tensor
A 1D parameter tensor that holds the values of input parameters to the simulator model.
syn_current : dict
Contains the initial current "V0" [float], the stimulation current trace
"It" [ndarray], the product of the stimulation current trace and the area
of the soma "ItA": ndarray, the time axis "ts": ndarray, the step size
"dt": float and the simulus time course "StimOnset": float,
"StimEnd": float and "StimDur": float.
corr_mat : torch.Tensor
Matrix containing the pairwise correlation coefficients of the different parameters.
pair : tuple, list, numpy.ndarray
Specifies which correlation axis to change the parameters along.
start : float
Minimum value of parameter change along correlation axis.
stop : float
Maximum value of parameter change along correlation axis.
N : int
Number of values inbetween min. and max. amount of change.
figsize : tuple
Specify size of each plt.figure.
timewindow : tuple, list or ndarray
The voltage and current trace will only be plotted between (t1,t2).
To be specified in secs.
compare_changes : bool
Whether or not to include a plot that compares the correlated traces for all
magnitudes of change.
savefig : bool
Determines whether or not to save the plotted figure.
fig_name : str
Integrates figure name into the filename of the saved figure.
mode : str
Specifies whether cell/experiment specific parameters like ASoma and V0
should be specified as part of the model parameters ("model_params")
or if ASoma and V0 should be considered part of the syn_current
or trace object ("trace")."""
base_trace = runHH(params, syn_current=syn_current, mode=mode)
corr_traces = []
fig = plt.figure(figsize=figsize)
for idx, change in enumerate(np.linspace(start, stop, N)):
# generate correlated params, simulate and store trace
pars = generate_correlated_parameters(params, corr_mat, pair=pair, mag_of_change=change)
corr_trace = runHH(pars, syn_current=syn_current)
corr_traces.append((change,corr_trace))
# plot both traces onto same axis
axs = base_trace.inspect(voltage_only=True, timewindow=timewindow, label="base params",
line_color="grey")
corr_trace.inspect(axes=axs, voltage_only=True, timewindow=timewindow, label="correlated params",
line_color=colormap(idx/N))
plt.ylim(min(base_trace.Vt*1e3),max(base_trace.Vt*1e3))
plt.legend(loc=1)
plt.title("change along correlation axis = {0:.1f}%".format(change*100))
if savefig:
plt.savefig('../data/figures/{}_8paramprior_support_corr_change_effect_on_traces_{}.png'.format(fig_name, pair))
# plots all changes into one set ox axes
if compare_changes:
axs = base_trace.inspect(voltage_only=True, timewindow=timewindow, label="base params",
line_color="grey")
for idx, (change,corr_trace) in enumerate(corr_traces):
corr_trace.inspect(axes=axs, voltage_only=True, timewindow=timewindow,
label="{0:.1f}%".format(change*100), line_color=colormap(idx/N))
plt.legend(loc=1)
if savefig:
plt.savefig('../data/figures/{}_8paramprior_support_corr_change_effect_on_traces_combined_{}.png'.format(fig_name, pair))
def plot_correlated_summary_stats(params, syn_current, corr_mat, pair=(0,1), start=0.01,
stop=0.5, N=5, figsize=(15,10), timewindow=[0.3,0.33],
selected_stats=None, change_only=False,
savefig=False, fig_name="default",
mode="auto", summary_func="v1"):
"""Plots the effects of changing correlated parameters, for one parameter pair,
on the summary stats of the simulated voltage traces for different magnitudes of change.
Parameters
----------
params : torch.Tensor
A 1D parameter tensor that holds the values of input parameters to the simulator model.
syn_current : dict
Contains the initial current "V0" [float], the stimulation current trace
"It" [ndarray], the product of the stimulation current trace and the area
of the soma "ItA": ndarray, the time axis "ts": ndarray, the step size
"dt": float and the simulus time course "StimOnset": float,
"StimEnd": float and "StimDur": float.
corr_mat : torch.Tensor
Matrix containing the pairwise correlation coefficients of the different parameters.
pair : tuple, list, numpy.ndarray
Specifies which correlation axis to change the parameters along.
start : float
Minimum value of parameter change along correlation axis.
stop : float
Maximum value of parameter change along correlation axis.
N : int
Number of values inbetween min. and max. amount of change.
figsize : tuple
Specify size of each plt.figure.
timewindow : tuple, list or ndarray
The voltage and current trace will only be plotted between (t1,t2).
To be specified in secs.
selected_stats : None, tuple, list or array
Indices of summary stats to be used.
change_only : bool
Whether or not to only plot the change in the difference of summary stats.
savefig : bool
Determines whether or not to save the plotted figure.
fig_name : str
Integrates figure name into the filename of the saved figure.
mode : str
Specifies whether cell/experiment specific parameters like ASoma and V0
should be specified as part of the model parameters ("model_params")
or if ASoma and V0 should be considered part of the syn_current
or trace object ("trace")."""
base_trace = runHH(params, syn_current=syn_current, mode=mode)
base_trace.summarise(selected_stats)
change_hist = pd.DataFrame(data=None,columns=base_trace.Summary.keys())
for idx, change in enumerate(np.linspace(start, stop, N)):
sum_df = compare_correlated_summary_stats(params, syn_current, pair, corr_mat, change, selected_stats, mode=mode, summary_func=summary_func)
rel_changes = abs((sum_df["base params"] - sum_df["correlated params"])/sum_df["base params"])*100
changes_df = pd.DataFrame(rel_changes, columns=[r"$\Delta_{corr}$"+" = {}%".format(change*100)])
change_hist = change_hist.append(changes_df.T)
if not change_only:
plot_change_of_corr_summaries(sum_df, changes_df, colormap(idx/N), savefig, fig_name)
change_hist.T.plot(kind="bar", figsize=(15,5), cmap=colormap)
plt.ylim(0,50)
if savefig:
plt.savefig('../data/figures/{}_8paramprior_support_corr_change_effect_on_summaries_change.png'.format(fig_name))
def plot_correlation_effects(params, syn_current, corr_mat, mag_of_change=0.1, figsize=(15,10),
timewindow=[0.3,0.33], effect_on="traces", selected_stats=None,
savefig=False, fig_name="default", summary_func="v1",
mode="auto"):
"""Plots the effects of changing correlated parameter pairs, for all pairs, on the simulated
voltage traces or the change in summary statistics. Starts with (0,1).
Parameters
----------
params : torch.Tensor
A 1D parameter tensor that holds the values of input parameters to the simulator model.
syn_current : dict
Contains the initial current "V0" [float], the stimulation current trace
"It" [ndarray], the product of the stimulation current trace and the area
of the soma "ItA": ndarray, the time axis "ts": ndarray, the step size
"dt": float and the simulus time course "StimOnset": float,
"StimEnd": float and "StimDur": float.
corr_mat : torch.Tensor
Matrix containing the pairwise correlation coefficients of the different parameters.
mag_of_change : float
Value from the intervall of [-k, k], where k \in R that specifies by how much the
input parameters are altered.
figsize : tuple
Specify size of plt.figure.
timewindow : tuple, list or ndarray
The voltage and current trace will only be plotted between (t1,t2).
To be specified in secs.
effect_on : str ("traces" or "summary stats")
Specify how the effect of a correlated change in parameters affects either the summary
statistics or the voltage traces.
selected_stats : None, tuple, list or array
Indices of summary stats to be used.
savefig : bool
Determines whether or not to save the plotted figure.
fig_name : str
Integrates figure name into the filename of the saved figure.
summary_func : str
A descriptive string like "v1" or "v2" can be provided to
use one of the summary methods already implemented as part of Trace().
mode : str
Specifies whether cell/experiment specific parameters like ASoma and V0
should be specified as part of the model parameters ("model_params")
or if ASoma and V0 should be considered part of the syn_current
or trace object ("trace")."""
N = np.max(params.shape)
fig, axes = plt.subplots(N-1,N-1, figsize=figsize)
# loop over pairs in upper triangle of matrix
for j in range(1,N):
for i in range(j):
pars = generate_correlated_parameters(params, corr_mat, pair=(i,j))
correlated_trace = runHH(pars, syn_current=syn_current, mode=mode)
base_trace = runHH(params, syn_current=syn_current, mode=mode)
axs = axes[i,j-1]
if "trace" in effect_on.lower():
# plot traces
base_trace.inspect(axes=axs, voltage_only=True, timewindow=timewindow)
correlated_trace.inspect(axes=axs, voltage_only=True, timewindow=timewindow)
# format axes
if i != j-1:
axs.get_xaxis().set_ticks([])
axs.get_yaxis().set_ticks([])
axs.set_xticklabels([])
axs.set_yticklabels([])
axs.set_xlabel(None)
axs.set_ylabel(None)
plt.subplots_adjust(hspace = .65, wspace = .2)
else:
if i == 0:
axs.set_ylabel("V [mV]", fontsize=10)
axs.set_xlabel("j = "+str(j), fontsize=10)
if j == 7:
axs.set_xlabel("t [ms]", fontsize=10)
axs.set_ylabel("i = "+ str(i), fontsize=10)
if i != 0 and j != 7:
axs.set_xlabel("j = "+ str(j), fontsize=10)
axs.set_ylabel("i =" + str(i), fontsize=10)
if "summary" in effect_on.lower():
summary_df = compare_correlated_summary_stats(params, syn_current, (i,j), corr_mat, mag_of_change, selected_stats, summary_func=summary_func, mode=mode)
changes_df = pd.DataFrame(abs((summary_df["base params"] - summary_df["correlated params"])/summary_df["base params"])*100, columns=[r"$\Delta_{corr}$"+" = {}\%".format(mag_of_change*100)])
# plot changes
changes_df.plot(kind="bar", ax=axs, color="black",legend=False, width=1)
# format axes and plots
axs.set_xticklabels([])
axs.set_yticklabels([])
axs.get_xaxis().set_ticks([])
axs.get_yaxis().set_ticks([])
axs.set_ylim([0,100])
plt.subplots_adjust(hspace = .65, wspace = .2)
plt.suptitle("Percentage change in the summary statistics")
if i == j-1:
axs.set_xlabel("j = "+ str(j), fontsize=10)
axs.set_ylabel("i =" + str(i), fontsize=10)
axs.get_yaxis().set_ticks([0,100])
axs.set_yticklabels([0,100])
if i == 0 and j == 1:
axs.set_ylabel(r"$\Delta$ [%]", fontsize=10)
axs.set_xlabel("j = "+str(j), fontsize=10)
if j == N-1 and i == N-2:
axs.set_xlabel("Summary Stats", fontsize=10)
axs.set_ylabel("i = "+ str(i), fontsize=10)
else:
axs.set_xlabel("j = "+ str(j), fontsize=10)
axs.set_ylabel("i =" + str(i), fontsize=10)
# remove diagonal and lowever triangle of matrix
for i in range(1,N):
for j in range(i-1):
axs = axes[i-1,j]
axs.get_xaxis().set_ticks([])
axs.get_yaxis().set_ticks([])
sns.despine(ax = axs, left = True, bottom = True)
plt.suptitle("")
if savefig:
plt.savefig('../data/figures/{}_8paramprior_support_corr_change_effect_on_{}.png'.format(fig_name, effect_on))