Source code for norse.torch.utils.plot.plot

"""
Utilities for plotting spikes in layers over time in 2D and 3D.
"""

from typing import Any, List, Optional

import matplotlib.pyplot as plt
import numpy as np
import torch

from norse.torch.functional.izhikevich import (
    IzhikevichSpikingBehavior,
    izhikevich_feed_forward_step,
)


def _detach_tensor(tensor: torch.Tensor):
    if isinstance(tensor, torch.Tensor):
        if tensor.requires_grad:
            return tensor.detach().cpu()
        else:
            return tensor.cpu()
    else:
        return tensor


[docs]def plot_heatmap_2d( data: torch.Tensor, axes: Optional[plt.Axes] = None, show_colorbar: bool = False, **kwargs, ): """ Plots a heatmat of two-dimensional data Example: >>> data = torch.randn(28, 28) >>> plot_heatmap_2d(data) .. plot:: import torch from norse.torch.utils.plot import plot_heatmap_2d data = torch.randn(28, 28) plot_heatmap_2d(data) plt.show() Arguments: data (torch.Tensor): A tensor of data to plot in two dimensions (typically spikes with time in the first dimension and neuron id in the second). axes (matplotlib.axes.Axes): The matplotlib axis to plot on, if any. Defaults to :meth:`matplotlib.pyplot.gca` show_colorbar (bool): Show a colorbar (True) or not (False). kwargs: Specific key-value arguments to style the figure fed to the :meth:`matplotlib.pyplot.imshow` function. Returns: A :class:`matplotlib.axes.Axes`. """ ax = axes if axes is not None else plt.gca() kwargs["aspect"] = kwargs.get("aspect", "auto") kwargs["interpolation"] = kwargs.get("interpolation", "none") pos = plt.imshow(_detach_tensor(data).T, **kwargs) if show_colorbar: plt.colorbar(pos, ax=ax) return ax
[docs]def plot_heatmap_3d(spikes: torch.Tensor, show_colorbar: bool = False, **kwargs): """ Plots heatmaps for some activity in several layers. Expects a named tensor with names=('L', 'X', 'Y'). Instead of using the :meth:`matplotlib.pyplot.imshow` matplotlib function, we make use of the :meth:`matplotlib.pyplot.scatter` function to disperse points in 3d. Example: >>> import torch >>> from norse.torch.utils.plot import plot_heatmap_3d >>> data = torch.randn(4, 28, 28, names=('L', 'X', 'Y')) >>> plot_heatmap_3d(data) >>> plt.show() .. plot:: import torch from norse.torch.utils.plot import plot_heatmap_3d data = torch.randn(4, 28, 28, names=('L', 'X', 'Y')) plot_heatmap_3d(data) plt.show() Arguments: spikes (torch.NamedTensor): A tensor named with four dimensions: T (time), L (layer), X, Y. Expected to be in the range :math:`[0, 1]`. show_colorbar (bool): Show a colorbar (True) or not (False). kwargs: Specific key-value arguments to style the figure fed to the :meth:`matplotlib.pyplot.scatter` function. Returns: An :class:`matplotlib.axes.Axes` object """ spikes = _detach_tensor(spikes).align_to(*"LXY") L = spikes.shape[0] ax = plt.gcf().add_subplot(1, 1, 1, projection="3d") unnamed = spikes + 1e-10 # Add infinitely small amount to "trigger" all pixels unnamed.names = None # Unnamed tensor required for to_sparse s = unnamed.to_sparse().coalesce() kwargs["c"] = kwargs.get("c", s.values()) ax.set_title(None) ax.invert_yaxis() ax.invert_zaxis() ax.set_xlim([0, L - 1]) ax.set_ylim([0, s.shape[2]]) ax.set_zlim([0, s.shape[1]]) plt.xticks(range(0, L), range(1, L + 1)) pos = ax.scatter(s.indices()[0], s.indices()[2], s.indices()[1], **kwargs) if show_colorbar: plt.gcf().colorbar(pos, ax=ax) return ax
[docs]def plot_neuron_states( states: List[Any], *variables: str, label: bool = True, axes: plt.Axes = None, **kwargs, ): """ Plots state variables in a line plot based on a list of states over time. Example: >>> cell = LIF() >>> _, states = cell(torch.ones(10, 3)) >>> plot_neuron_states(states, "i") Arguments: states (List[Any]): A list of state tuples containing state variables. We assume the list is ordered by time. *variables (str): A set of variables to plot. label (bool): Whether or not to render the label in the plot. axes (plt.Axes): An axes object to render on, if given. Returns: A plt.Axes object for further manipulation or rendering. """ assert len(variables) > 0, "0 variables were given to render, we require at least 1" if axes is None: axes = plt.gca() for variable in variables: values = [ np.array(_detach_tensor(getattr(state, variable))) for state in states ] if label: axes.plot(values, label=variable, **kwargs) else: axes.plot(values, **kwargs) if label: axes.legend() return axes
[docs]def plot_histogram_2d(spikes: torch.Tensor, axes: Optional[plt.Axes] = None, **kwargs): """ Plots a histogram of 1-dimensional data. Example: >>> cell = LIF() >>> data = torch.ones(10, 10) + torch.randn(10, 10) >>> spikes, state = cell(data) >>> plot_histogram_2d(state.v) .. plot:: import torch from norse.torch import LIF from norse.torch.utils.plot import plot_histogram_2d spikes, state = LIF()(torch.ones(10, 10) + torch.randn(10, 10)) plot_histogram_2d(state.v) plt.show() Arguments: data (torch.Tensor): A tensor of single-dimensional data. axes (matplotlib.axes.Axes): The matplotlib axis to plot on, if any. Defaults to :meth:`matplotlib.pyplot.gca` kwargs: Specific key-value arguments to style the figure fed to the :meth:`matplotlib.pyplot.hist` function. Returns: An :class:`matplotlib.axes.Axes`. """ ax = axes if axes is not None else plt.gca() plt.hist(_detach_tensor(spikes).numpy(), **kwargs) return ax
[docs]def plot_scatter_3d( spikes: torch.Tensor, axes: Optional[List[plt.Axes]] = None, show_colorbar: bool = True, **kwargs, ): """ Plots spike activity in time. If multiple layers are given, the layers will be shown in subplots. Expects a named tensor in three dimensions (L, X, Y) or four, with time (T, L, X, Y). Example: >>> distribution = torch.distributions.bernoulli.Bernoulli(torch.tensor([0.02])) >>> data = distribution.sample(sample_shape=(3, 100, 10, 10)).squeeze() >>> data.names = ('L', 'T', 'X', 'Y') >>> plot_scatter_3d(data) .. plot:: import torch from norse.torch.utils.plot import plot_scatter_3d plt.figure(figsize=(10, 3)) distribution = torch.distributions.bernoulli.Bernoulli(torch.tensor([0.02])) data = distribution.sample(sample_shape=(3, 100, 10, 10)).squeeze() data.names=('L', 'T', 'X', 'Y') plot_scatter_3d(data) plt.show() Arguments: spikes (torch.NamedTensor): A tensor named with four dimensions: T (time), L (layer), X, Y. Expected to be in the range :math:`[0, 1]`. axes (List[plt.Axes]): A list of Axes that should have the same length as the L dimension in the spike tensor. Defaults to None, which will generate a grid for you. show_colorbar (bool): Show a colorbar (True) or not (False). kwargs: Specific key-value arguments to style the figure fed to the :meth:`matplotlib.pyplot.scatter` function. Returns: A list of :class:`matplotlib.axes.Axes` """ if len(spikes.shape) > 3: spikes = _detach_tensor(spikes).align_to(*"TLXY") else: spikes = _detach_tensor(spikes).align_to(*"LXY").unsqueeze(0) T = spikes.shape[0] L = spikes.shape[1] axes = [None] * L if axes is None else axes for l, axis in enumerate(axes): if axis is not None: ax = axis else: ax = plt.gcf().add_subplot(1, L, l + 1, projection="3d") axes[l] = ax unnamed = spikes[:, l] unnamed.names = None # Unnamed tensor required for to_sparse s = unnamed.to_sparse().coalesce() ax.invert_yaxis() ax.invert_zaxis() ax.set_xlim([0, T]) pos = ax.scatter( s.indices()[0], s.indices()[2], s.indices()[1], c=s.values(), **kwargs ) if show_colorbar: plt.gcf().colorbar(pos, ax=axes, fraction=0.04 / len(axes)) return axes
[docs]def plot_spikes_2d(spikes: torch.Tensor, axes: plt.Axes = None, **kwargs): """ Plots a 2D diagram of spikes. Works similar to the :meth:`plot_heatmap_2d` but in black and white. Example: >>> import torch >>> from norse.torch import LIF >>> from norse.torch.utils.plot import plot_spikes_2d >>> spikes, _ = LIF()(torch.randn(200, 10)) >>> plot_spikes_2d(spikes) >>> plt.show() .. plot:: import torch from norse.torch import LIF from norse.torch.utils.plot import plot_spikes_2d plt.figure(figsize=(8, 4)) spikes, _ = LIF()(torch.randn(200, 10)) plot_spikes_2d(spikes) plt.show() Arguments: spikes (torch.Tensor): A tensor of spikes from a single layer in two dimensions. axes (matplotlib.axes.Axes): The matplotlib axis to plot on, if any. Defaults to :meth:`matplotlib.pyplot.gca` kwargs: Specific key-value arguments to style the figure fed to the :meth:`matplotlib.pyplot.imshow` function. Returns: An :class:`matplotlib.axes.Axes` object """ kwargs["cmap"] = kwargs.get("cmap", "binary") if axes is None: axes = plt.gca() ytick_step = max(1, spikes.shape[-1] // 10) axes.set_yticks(range(0, spikes.shape[-1], ytick_step)) return plot_heatmap_2d(spikes, axes=axes, **kwargs)
[docs]def plot_izhikevich( behavior: IzhikevichSpikingBehavior, current: float = 1, time_print: int = 250, time_current: int = 20, timestep_print: float = 0.1, ): """ Computes and plots a 2D visualisation of the behavior of an Izhikevich neuron model. By default, the time window is 250ms with a time step of 0.1ms Example : >>> import torch >>> from norse.torch import tonic_spiking >>> from norse.torch.utils.plot import plot_izhikevich >>> plot_izhikevich(tonic_spiking) .. plot:: import torch from norse.torch import tonic_spiking from norse.torch.utils.plot import plot_izhikevich plot_izhikevich(tonic_spiking) Arguments : behavior (IzhikevichSpikingBehavior) : behavior of an Izhikevich neuron current (float) : strengh of the input current, defaults to 1 time_print (float) : size of the time window for simulation (in ms) time_current (float) : time at which the input current goes from 0 to current (in ms) timestep_print (float) : timestep of the simulation (in ms) """ p, s = behavior T1 = time_current vs = [] us = [] cs = [] time = [] for t in np.arange(0, time_print, timestep_print): vs.append(s.v.item()) us.append(s.u.item()) time.append(t * timestep_print) if t > T1: input_current = current * torch.ones(1) else: input_current = 0 * torch.ones(1) _, s = izhikevich_feed_forward_step(input_current, s, p) cs.append(input_current) fig = plt.figure() ax1 = fig.add_subplot(111) ax1.set_ylabel("Membrane potential (mV)") ax1.set_xlabel("Time (ms)") ax1.plot(time, vs) ax1.plot(time, cs)