spikingjelly.timing_based package#

Encoding#

class spikingjelly.timing_based.encoding.GaussianTuning(n, m, x_min, x_max)[源代码]#

基类：object

API Language - 中文 | English

中文

使用高斯感受野将连续值编码为脉冲时间。每个输入维度由 m 个神经元编码，公开属性 mu 和 sigma2 分别保存感受野中心和方差。

参数:

n (int) -- 输入特征维度数量
m (int) -- 每个特征使用的神经元数量，必须大于 2
x_min (Tensor) -- 各输入特征的最小值，形状为 [n]
x_max (Tensor) -- 各输入特征的最大值，形状为 [n]

抛出:

ValueError -- n 或 m 非法、输入范围形状不匹配，或任一 x_min 不小于对应的 x_max

English

Encode continuous values as spike times with Gaussian receptive fields. Each input dimension uses m neurons. Public attributes mu and sigma2 contain the receptive-field centers and variances.

参数:

n (int) -- Number of input feature dimensions
m (int) -- Number of neurons per feature; must be greater than 2
x_min (Tensor) -- Per-feature minimum values with shape [n]
x_max (Tensor) -- Per-feature maximum values with shape [n]

抛出:

ValueError -- If n or m is invalid, the range tensors have incompatible shapes, or any minimum is not smaller than its maximum

encode(x, max_spike_time=50)[源代码]#

API Language - 中文 | English

中文

参数:

x (Tensor) -- 输入张量，形状为 [batch_size, n, samples_count]
max_spike_time (int) -- 非负编码时间窗长度；达到该值的神经元以 -1 表示不发放。当其为 0 时，所有神经元均不发放

返回:

形状为 [batch_size, n, samples_count, m] 的脉冲时间

返回类型:

Tensor

抛出:

AssertionError -- 输入不是三维张量或特征维不等于 n
ValueError -- max_spike_time 为负数

English

参数:

x (Tensor) -- Input tensor with shape [batch_size, n, samples_count]
max_spike_time (int) -- Non-negative encoding-window length; neurons reaching it are marked inactive with -1. When it is 0, all neurons are inactive

返回:

Spike times with shape [batch_size, n, samples_count, m]

返回类型:

Tensor

抛出:

AssertionError -- If the input is not three-dimensional or its feature dimension differs from n
ValueError -- If max_spike_time is negative

Neuron#

class spikingjelly.timing_based.neuron.Tempotron(in_features, out_features, T, tau=15.0, tau_s=3.75, v_threshold=1.0)[源代码]#

基类：Module

API Language - 中文 | English

中文

实现 Tempotron 脉冲时序分类神经元。输入中的每个值表示对应输入神经元的脉冲时刻，负值表示未发放。

参数:

in_features (int) -- 输入神经元数量
out_features (int) -- 输出神经元数量
T (int) -- 仿真时间窗口
tau (float) -- 膜电位时间常数
tau_s (float) -- 突触电流时间常数
v_threshold (float) -- 发放阈值

抛出:

ValueError -- 数量、时间常数或阈值非正，或两个时间常数相等

English

Implement a Tempotron neuron for spike-timing classification. Each input value is the spike time of one input neuron; a negative value means that neuron did not spike.

参数:

in_features (int) -- Number of input neurons
out_features (int) -- Number of output neurons
T (int) -- Simulation time window
tau (float) -- Membrane-potential time constant
tau_s (float) -- Synaptic-current time constant
v_threshold (float) -- Firing threshold

抛出:

ValueError -- If a size, time constant, or threshold is not positive, or if the two time constants are equal

mse_loss(v_max, label)[源代码]#

API Language - 中文 | English

中文

参数:

v_max (Tensor) -- 各输出神经元在时间窗口内的最大电压，形状为 [batch_size, out_features]
label (Tensor) -- 分类标签，形状为 [batch_size]

返回:

仅计算错误发放神经元的均方误差

返回类型:

Tensor

English

参数:

v_max (Tensor) -- Maximum voltage of each output neuron with shape [batch_size, out_features]
label (Tensor) -- Class labels with shape [batch_size]

返回:

Mean squared error over incorrectly firing neurons

返回类型:

Tensor

forward(in_spikes, ret_type)[源代码]#

API Language - 中文 | English

中文

参数:

in_spikes (Tensor) -- 输入脉冲时刻，形状为 [batch_size, in_features]
ret_type (str) -- "v"、"v_max" 或 "spikes"

返回:

完整电压轨迹、最大电压或输出脉冲时刻

返回类型:

Tensor

抛出:

ValueError -- ret_type 不受支持

English

参数:

in_spikes (Tensor) -- Input spike times with shape [batch_size, in_features]
ret_type (str) -- "v", "v_max", or "spikes"

返回:

Full voltage trace, maximum voltage, or output spike times

返回类型:

Tensor

抛出:

ValueError -- If ret_type is unsupported

Examples#

examples