代码代写｜Assignment 4: Introduction to neural networks
This assignment will have you building simple neural networks to perform a variety of machine learning tasks.
This project was developed at UC Berkeley (http://ai.berkeley.edu), and modified for CISC-352 (including a new rubric).
For this project, you will need to install the following two libraries:
conda activate [your environment name]
conda install -c anaconda numpy
conda install -c conda-forge matplotlib
pip install numpy
pip install matplotlib
Note: If you are using Python 3.x, you may need to change pip to pip3.
To test that everything has been installed, run:
python autograder.py –check-dependencies
If numpy and matplotlib are installed correctly, you should see a window pop up where a line segment spins in a circle:
Files to Edit and Submit: You will fill in portions of models.py during the assignment. Please do not change the other files in this distribution.
Files you should read but NOT
nn.py Neural network mini-library
Files you will not edit:
backend.py Backend code for various machine
autograder.py Project autograder
To test your implementation, run the autograder in command line:
python autograder.py -q q1
python autograder.py -q q2
python autograder.py -q q3
Your grade will most likely correspond to the reported score on the auto-grader
(we will use our own auto-grader code that mirrors what you are provided).
Important: Marks will be deducted for poor code quality. Anywhere from 0.5- 1.5pt per problem, depending on the severity of the issues. Please use comments where it makes sense, and try to keep your code clear and easy for the TAs to understand.
Note: Read Neural Network Tips and Tutorial before attempting the question.
Question 1 (1.5 points): Perceptron
In this part, you will implement a binary perceptron. Your task will be to complete the implementation of the PerceptronModel class in models.py.
For the perceptron, the output labels will be either 1 or -1 , meaning that data points (x, y) from the dataset will have y be a nn.Constant node that contains either 1 or -1 as its entries.
We have already initialized the perceptron weights self.w to be a 1 × dimensions parameter node. The provided code will include a bias feature inside x_point when needed, so you will not need a separate parameter for the bias.
Your tasks are to: 1. Implement the run(self, x_point) method. This should compute the dot product of the stored weight vector and the given input,returning an nn.DotProduct object.
Write the train_model(self) method. This should repeatedly loop over the data set and make updates on examples
In this project, the only way to change the value of a parameter is by calling parameter.update(multiplier, direction), which will perform the update to the weights:
weights ← weights + multiplier * direction
The direction argument is a Node with the same shape as the parameter, and the multiplier argument is a Python scalar.
Note: the autograder should take at most 20 seconds or so to run for a correct implementation. If the autograder is taking forever to run, your code probably has a bug.
Question 2 (2 points): Non-linear Regression
For this question, you will train a neural network to approximate sin(x) over [-2*Pi , 2*Pi]
You will need to complete the implementation of the RegressionModel class in models.py. For this problem, a relatively simple architecture should suffice (see Neural Network Tips for architecture tips. Use nn.SquareLoss as your loss.
Your implementation will receive full points if it gets a loss of 0.02 or better, averaged across all examples in the dataset. You may use the training loss to determine when to stop training (use nn.as_scalar to convert a loss node to a Python number). Note that it should take the model a few minutes to train.
Question 3 (2.5 points): Digit Classification
For this question, you will train a network to classify handwritten digits from the MNIST dataset.
Each digit is of size 28 × 28 pixels, the values of which are stored in a 784- dimensional vector of floating point numbers. Each output we provide is a 10-dimensional vector which has zeros in all positions, except for a one in the position corresponding to the correct class of the digit.
Complete the implementation of the DigitClassificationModel class in models.py. The return value from DigitClassificationModel.run() should be a batch_size × 10 node containing scores, where higher scores indicate a higher probability of a digit belonging to a particular class (0-9). You should use nn.SoftmaxLoss as your loss. Do not put a ReLU activation after the last layer of the network.
In addition to training data, there is also validation data and a test set. You can use dataset.get_validation_accuracy() to compute validation accuracy for your model, which can be useful when deciding whether to stop training. The test set will be used by the autograder.
To receive points for this question, your model should achieve an accuracy of at least 97% on the test set. For reference, our implementation consistently achieves an accuracy of near 98% on the validation data after training for around 10-15 epochs. Note that the test grades you on test accuracy, while you only have access to validation accuracy – so if your validation accuracy meets the 97% threshold, you may still fail the test if your test accuracy does not meet the threshold. Therefore, it may help to set a slightly higher stopping threshold on validation accuracy, such as 97.5% or 98%.
Neural Network Tips
Building Neural Nets
Throughout the applications portion of the project, you’ll use the framework provided in nn.py to create neural networks to solve a variety of machine learning problems. A simple neural network has layers, where each layer performs a linear operation (just like perceptron). Layers are separated by a non-linearity,which allows the network to approximate general functions. We’ll use the ReLU operation for our non-linearity, defined as relu(x) = max(x,0). For example, a simple two-layer neural network for mapping an input row vector x to an output vector f(x) would be given by the function:
f(x) = relu(x * W1 + b1) * W2 + b2
where we have parameter matrices W1 and W2 and parameter vectors b1 and b2 to learn during gradient descent. W1 will be an i × h matrix, where i s the dimension of our input vectors x , and h is the hidden layer size. b1 will be a size h vector. We are free to choose any value we want for the hidden size ( we will just need to make sure the dimensions of the other matrices and vectors agree so that we can perform the operations). Using a larger hidden size will usually make the network more powerful (able to fit more training data), but can make the network harder to train (since it adds more parameters to all the matrices and vectors we need to learn), or can lead to overfitting on the training data.
We can also create deeper networks by adding more layers, for example a three-layer net:
f(x) = relu( relu(x * W1 + b1) * W2 + b2) * W3 + b3
Note on Batching For efficiency,
You will be required to process whole batches of data at once rather than a single example at a time. This means that instead of a single input row vector x with size i , you will be presented with a batch of b inputs represented as a b × i matrix X . We provide an example for linear regression to demonstrate how a linear layer can be implemented in the batched setting.
Note on Randomness
The parameters of your neural network will be randomly initialized, and data in some tasks will be presented in shuffled order. Due to this randomness,it’s possible that you will still occasionally fail some tasks even with a strong architecture – this is the problem of local optima! This should happen very rarely, though – if when testing your code you fail the autograder twice in a row for a question, you should explore other architectures.
Deeper networks have exponentially more hyperparameter combinations, and getting even a single one wrong can ruin your performance. Use the small network to find a good learning rate and layer size; afterwards you can consider adding more layers of similar size.
– Hidden layer sizes: between 10 and 400.
– Batch size: between 1 and the size of the dataset. For Q2 and Q3, we require that total size of the dataset be evenly divisible by the batch size.
– Learning rate: between 0.001 and 1.0.
– Number of hidden layers: between 1 and 3.