Methods

CTRNN.eval( )

Description

Set the network to training mode, training will be performed and constraints will be enforced. Also, during training, the recurrent noises won’t be added.

Usage

import torch
from nn4n.model import CTRNN

ctrnn = CTRNN()
inputs = torch.rand(100, 1)

ctrnn = CTRNN()

for _ in range(100):
    ctrnn.eval()
    outputs, _ = ctrnn(inputs)

CTRNN.forward( )

Description

The forward pass of the CTRNN model. As CTRNN is a child class of torch.nn.Module, it can be called either using ctrnn.forward() or ctrnn(), where ctrnn is a instance of CTRNN.

Parameters

x: (torch.Tensor), required. The input tensor to the CTRNN model. The shape of the input tensor should be (seq_len, batch_size, input_dim).

Returns

outputs: (torch.Tensor) The output tensor of the CTRNN model. The shape of the output tensor is (seq_len, batch_size, output_dim). Note: when batch_size is 1, the shape of the output tensor is (seq_len, 1, output_dim).

hidden_states: (torch.Tensor) The hidden state tensor of the CTRNN model. The shape of the hidden state tensor is (seq_len, batch_size, hidden_dim). Note: when batch_size is 1, the shape of the hidden state tensor is (seq_len, 1, hidden_dim).

Usage

import torch
from nn4n.model import CTRNN

ctrnn = CTRNN()
outputs, hidden_states = ctrnn(torch.rand(100, 1))

# or

outputs, hidden_states = ctrnn.forward(torch.rand(100, 1))

CTRNN.load( )

Description

Load the CTRNN model from a .pth file. All model parameters and attributes will be loaded.

Parameters

path: (str), required. The path to save the model. Must end with .pth.

Usage

from nn4n.model import CTRNN

ctrnn = CTRNN()
ctrnn.load('ctrnn.pth')

CTRNN.print_layers( )

Description

Print out all layer details of the CTRNN.

Usage

from nn4n.model import CTRNN

ctrnn = CTRNN()
ctrnn.print_layers()

Output

Linear Layer: 
   | input_dim:  1
   | output_dim: 100
   | dist:       uniform
   | bias:       True
   | shape:      torch.Size([100, 1])
   | weight_min: -0.9917911291122437
   | weight_max: 0.9899972677230835
   | bias_min:   0.0
   | bias_max:   0.0
   | sparsity:   1

Recurrence: 
   | hidden_min:    0.0
   | hidden_max:    0.0
   | hidden_mean:   0.0
   | preact_noise:  0
   | postact_noise: 0
   | activation:    relu
   | alpha:         0.1

Hidden Layer: 
   | self_connections: False
   | input/output_dim: 100
   | distribution:     normal
   | bias:             True
   | dale:             False
   | shape:            torch.Size([100, 100])
   | weight_min:       -0.3491513133049011
   | weight_max:       0.3499620854854584
   | weight_mean:      -0.0004077071789652109
   | bias_min:         0.0
   | bias_max:         0.0
   | sparsity:         0.9900000095367432
   | scaling:          1.0

Linear Layer: 
   | input_dim:  100
   | output_dim: 1
   | dist:       uniform
   | bias:       True
   | shape:      torch.Size([1, 100])
   | weight_min: -0.09865634888410568
   | weight_max: 0.0993180200457573
   | bias_min:   0.0
   | bias_max:   0.0
   | sparsity:   1

CTRNN.save( )

Description

Save the CTRNN model to a .pth file. All model parameters and attributes will be saved.

Parameters

path: (str) required. The path to save the model. Must end with .pth.

Usage

from nn4n.model import CTRNN

ctrnn = CTRNN()
ctrnn.save('ctrnn.pth')

CTRNN.train()

Description

Set the network to training mode, training will be performed and constraints will be enforced. Also, during training, the recurrent noises (preact_noise and postact_noise) won’t be added.

Usage

import torch
from nn4n.model import CTRNN

inputs = torch.rand(100, 1)
targets = torch.rand(100, 1)

ctrnn = CTRNN()
optimizer = torch.optim.Adam(rnn.parameters(), lr=0.001)

for _ in range(100):
    ctrnn.train()
    outputs, _ = ctrnn(inputs)
    loss = torch.nn.MSELoss()(outputs, targets)
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

CTRNN.layers

Description

Get a list of the network layers.

Returns

layers: (list) A list of the network layers.

Usage

from nn4n.model import CTRNN

ctrnn = CTRNN()
layers = ctrnn.layers

for layer in layers:
    print(layer)

Output

LinearLayer()
HiddenLayer()
LinearLayer()