We will intialize model using the MultiLayerNetwork$()$ and

train the model with the train$()$ method We will display the training and plot the traing loss

# You should be able to get a train accuracy of $>91\%$ and a test accuracy $>90\%$ # The settings below gave $>92\%$ train accuracy and $>91\%$ test accuracy # Feel free to adjust the values and the architecture # Keep the network to $4$ layers and num$\_$epochs $500$

$|$

seed $=1$

torch.manual$\_$seed$($seed$)$

num$\_$epochs $=300$

learning$\_$rate $=0\mathrm{.}1$

net$\_$dims $=[784,200,10]$

# $2$ layer network

#$784$ is for image dimensions

#$200$ is arbitrary hidden layer

#$10$ is for number of categories

#activation$\_$list 'relu', 'softmax'$]$ #default activation.

## Initialize model

## Display model

## Estimate cost as the output of train$(...)$

### BEGIN SOLUTION ###

### END SOLUTION ###

yhat $=$ model. forward $(trx)$

ypred $=$ torch. yhat, axis $=0.$ reshape $(1,-1)$

trAcc $=$ torch.mean ypred $==$trY float $()$