¿Cómo imprimo el resumen del modelo en PyTorch?
¿Cómo imprimo el resumen de un modelo en PyTorch como lo model.summary()hace en Keras?
Model Summary:
____________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
====================================================================================================
input_1 (InputLayer) (None, 1, 15, 27) 0
____________________________________________________________________________________________________
convolution2d_1 (Convolution2D) (None, 8, 15, 27) 872 input_1[0][0]
____________________________________________________________________________________________________
maxpooling2d_1 (MaxPooling2D) (None, 8, 7, 27) 0 convolution2d_1[0][0]
____________________________________________________________________________________________________
flatten_1 (Flatten) (None, 1512) 0 maxpooling2d_1[0][0]
____________________________________________________________________________________________________
dense_1 (Dense) (None, 1) 1513 flatten_1[0][0]
====================================================================================================
Total params: 2,385
Trainable params: 2,385
Non-trainable params: 0
Aceptado
Sí, puede obtener una representación exacta de Keras utilizando el paquete pytorch-summary .
Ejemplo para VGG16:
from torchvision import models
from torchsummary import summary
vgg = models.vgg16()
summary(vgg, (3, 224, 224))
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 64, 224, 224] 1,792
ReLU-2 [-1, 64, 224, 224] 0
Conv2d-3 [-1, 64, 224, 224] 36,928
ReLU-4 [-1, 64, 224, 224] 0
MaxPool2d-5 [-1, 64, 112, 112] 0
Conv2d-6 [-1, 128, 112, 112] 73,856
ReLU-7 [-1, 128, 112, 112] 0
Conv2d-8 [-1, 128, 112, 112] 147,584
ReLU-9 [-1, 128, 112, 112] 0
MaxPool2d-10 [-1, 128, 56, 56] 0
Conv2d-11 [-1, 256, 56, 56] 295,168
ReLU-12 [-1, 256, 56, 56] 0
Conv2d-13 [-1, 256, 56, 56] 590,080
ReLU-14 [-1, 256, 56, 56] 0
Conv2d-15 [-1, 256, 56, 56] 590,080
ReLU-16 [-1, 256, 56, 56] 0
MaxPool2d-17 [-1, 256, 28, 28] 0
Conv2d-18 [-1, 512, 28, 28] 1,180,160
ReLU-19 [-1, 512, 28, 28] 0
Conv2d-20 [-1, 512, 28, 28] 2,359,808
ReLU-21 [-1, 512, 28, 28] 0
Conv2d-22 [-1, 512, 28, 28] 2,359,808
ReLU-23 [-1, 512, 28, 28] 0
MaxPool2d-24 [-1, 512, 14, 14] 0
Conv2d-25 [-1, 512, 14, 14] 2,359,808
ReLU-26 [-1, 512, 14, 14] 0
Conv2d-27 [-1, 512, 14, 14] 2,359,808
ReLU-28 [-1, 512, 14, 14] 0
Conv2d-29 [-1, 512, 14, 14] 2,359,808
ReLU-30 [-1, 512, 14, 14] 0
MaxPool2d-31 [-1, 512, 7, 7] 0
Linear-32 [-1, 4096] 102,764,544
ReLU-33 [-1, 4096] 0
Dropout-34 [-1, 4096] 0
Linear-35 [-1, 4096] 16,781,312
ReLU-36 [-1, 4096] 0
Dropout-37 [-1, 4096] 0
Linear-38 [-1, 1000] 4,097,000
================================================================
Total params: 138,357,544
Trainable params: 138,357,544
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 218.59
Params size (MB): 527.79
Estimated Total Size (MB): 746.96
----------------------------------------------------------------
Si bien no obtendrá información tan detallada sobre el modelo como en el resumen del modelo de Keras, simplemente imprimir el modelo le dará una idea sobre las diferentes capas involucradas y sus especificaciones.
Por ejemplo:
from torchvision import models
model = models.vgg16()
print(model)
El resultado en este caso sería algo como lo siguiente:
VGG (
(features): Sequential (
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU (inplace)
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU (inplace)
(4): MaxPool2d (size=(2, 2), stride=(2, 2), dilation=(1, 1))
(5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(6): ReLU (inplace)
(7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(8): ReLU (inplace)
(9): MaxPool2d (size=(2, 2), stride=(2, 2), dilation=(1, 1))
(10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(11): ReLU (inplace)
(12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(13): ReLU (inplace)
(14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(15): ReLU (inplace)
(16): MaxPool2d (size=(2, 2), stride=(2, 2), dilation=(1, 1))
(17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(18): ReLU (inplace)
(19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(20): ReLU (inplace)
(21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(22): ReLU (inplace)
(23): MaxPool2d (size=(2, 2), stride=(2, 2), dilation=(1, 1))
(24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(25): ReLU (inplace)
(26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(27): ReLU (inplace)
(28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(29): ReLU (inplace)
(30): MaxPool2d (size=(2, 2), stride=(2, 2), dilation=(1, 1))
)
(classifier): Sequential (
(0): Dropout (p = 0.5)
(1): Linear (25088 -> 4096)
(2): ReLU (inplace)
(3): Dropout (p = 0.5)
(4): Linear (4096 -> 4096)
(5): ReLU (inplace)
(6): Linear (4096 -> 1000)
)
)
Ahora podrías, como menciona Kashyap , usar el state_dictmétodo para obtener los pesos de las diferentes capas. ¡Pero usar esta lista de capas tal vez proporcionaría más dirección: crear una función auxiliar para obtener ese resumen del modelo similar a Keras!
Para utilizar el tipo de resumen de antorcha:
from torchsummary import summary
Instálalo primero si no lo tienes.
pip install torchsummary
Y luego puedes probarlo, pero ten en cuenta que, por alguna razón, no funciona a menos que configure el modelo en cuda alexnet.cuda:
from torchsummary import summary
help(summary)
import torchvision.models as models
alexnet = models.alexnet(pretrained=False)
alexnet.cuda()
summary(alexnet, (3, 224, 224))
print(alexnet)
Debe summarytomar el tamaño de entrada y el tamaño del lote se establece en -1, lo que significa cualquier tamaño de lote que proporcionemos.
Si configuramos summary(alexnet, (3, 224, 224), 32)esto significa usar el bs=32.
summary(model, input_size, batch_size=-1, device='cuda')
Afuera:
Help on function summary in module torchsummary.torchsummary:
summary(model, input_size, batch_size=-1, device='cuda')
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [32, 64, 55, 55] 23,296
ReLU-2 [32, 64, 55, 55] 0
MaxPool2d-3 [32, 64, 27, 27] 0
Conv2d-4 [32, 192, 27, 27] 307,392
ReLU-5 [32, 192, 27, 27] 0
MaxPool2d-6 [32, 192, 13, 13] 0
Conv2d-7 [32, 384, 13, 13] 663,936
ReLU-8 [32, 384, 13, 13] 0
Conv2d-9 [32, 256, 13, 13] 884,992
ReLU-10 [32, 256, 13, 13] 0
Conv2d-11 [32, 256, 13, 13] 590,080
ReLU-12 [32, 256, 13, 13] 0
MaxPool2d-13 [32, 256, 6, 6] 0
AdaptiveAvgPool2d-14 [32, 256, 6, 6] 0
Dropout-15 [32, 9216] 0
Linear-16 [32, 4096] 37,752,832
ReLU-17 [32, 4096] 0
Dropout-18 [32, 4096] 0
Linear-19 [32, 4096] 16,781,312
ReLU-20 [32, 4096] 0
Linear-21 [32, 1000] 4,097,000
================================================================
Total params: 61,100,840
Trainable params: 61,100,840
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 18.38
Forward/backward pass size (MB): 268.12
Params size (MB): 233.08
Estimated Total Size (MB): 519.58
----------------------------------------------------------------
AlexNet(
(features): Sequential(
(0): Conv2d(3, 64, kernel_size=(11, 11), stride=(4, 4), padding=(2, 2))
(1): ReLU(inplace)
(2): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
(3): Conv2d(64, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): ReLU(inplace)
(5): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
(6): Conv2d(192, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(7): ReLU(inplace)
(8): Conv2d(384, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(9): ReLU(inplace)
(10): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(11): ReLU(inplace)
(12): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(avgpool): AdaptiveAvgPool2d(output_size=(6, 6))
(classifier): Sequential(
(0): Dropout(p=0.5)
(1): Linear(in_features=9216, out_features=4096, bias=True)
(2): ReLU(inplace)
(3): Dropout(p=0.5)
(4): Linear(in_features=4096, out_features=4096, bias=True)
(5): ReLU(inplace)
(6): Linear(in_features=4096, out_features=1000, bias=True)
)
)
El paquete torchinfo(anteriormente torchsummary) produce una salida análoga a Keras 1 (para una forma de entrada determinada): 2
from torchinfo import summary
model = ConvNet()
batch_size = 16
summary(model, input_size=(batch_size, 1, 28, 28))
==========================================================================================
Layer (type:depth-idx) Output Shape Param #
==========================================================================================
├─Conv2d (conv1): 1-1 [5, 10, 24, 24] 260
├─Conv2d (conv2): 1-2 [5, 20, 8, 8] 5,020
├─Dropout2d (conv2_drop): 1-3 [5, 20, 8, 8] --
├─Linear (fc1): 1-4 [5, 50] 16,050
├─Linear (fc2): 1-5 [5, 10] 510
==========================================================================================
Total params: 21,840
Trainable params: 21,840
Non-trainable params: 0
Total mult-adds (M): 7.69
==========================================================================================
Input size (MB): 0.05
Forward/backward pass size (MB): 0.91
Params size (MB): 0.09
Estimated Total Size (MB): 1.05
==========================================================================================
Notas:
-
Torchinfo proporciona información complementaria a la que proporciona
print(your_model)PyTorch, similar a Tensorflowmodel.summary()... A diferencia de Keras, PyTorch tiene un gráfico computacional dinámico que puede adaptarse a cualquier forma de entrada compatible a través de múltiples llamadas, por ejemplo, cualquier tamaño de imagen suficientemente grande (para una red totalmente convolucional).
Como tal, no puede presentar un conjunto inherente de formas de entrada/salida para cada capa, ya que dependen de la entrada, y por eso en el paquete anterior debe especificar las dimensiones de entrada.
