spyx.data
#
Module Contents#
Functions#
|
Shift data augmentation tool. Rolls data along specified axes randomly up to a certain amount. |
|
Higher-order-function which builds a shuffle function for a dataset. |
|
Unrolls input data along axis 1 and converts to rate encoded spikes; the probability of spiking is based on the input value multiplied by a max rate, with each time step being a sample drawn from a Bernoulli distribution. |
|
Higher-order-function which returns an angle encoding function; given a continuous value, an angle converter generates a one-hot vector corresponding to where the value falls between a specified minimum and maximum. |
- spyx.data.shift_augment(max_shift=10, axes=(-1,))[source]#
Shift data augmentation tool. Rolls data along specified axes randomly up to a certain amount.
- Max_shift:
maximum to which values can be shifted.
- Axes:
the data axis or axes along which the input will be randomly shifted.
- spyx.data.shuffler(dataset, batch_size)[source]#
Higher-order-function which builds a shuffle function for a dataset.
- Dataset:
jnp.array [# samples, time, channels…]
- Batch_size:
desired batch size.
- spyx.data.rate_code(num_steps, max_r=0.75)[source]#
Unrolls input data along axis 1 and converts to rate encoded spikes; the probability of spiking is based on the input value multiplied by a max rate, with each time step being a sample drawn from a Bernoulli distribution. Currently Assumes input values have been rescaled to between 0 and 1.
- spyx.data.angle_code(neuron_count, min_val, max_val)[source]#
Higher-order-function which returns an angle encoding function; given a continuous value, an angle converter generates a one-hot vector corresponding to where the value falls between a specified minimum and maximum. To achieve non-linear descritization, apply a function to the continuous value before feeding it into the encoder.
- Neuron_count:
The number of output channels for the angle encoder
- Min_val:
A lower bound on the continuous input channel
- Max_val:
An upper bound on the continuous input channel.