def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct labels are known during training, but not at test time.
When correct labels are available, `y` is a (batch_size x 10) numpy
array. Each row in the array is a one-hot vector encoding the correct
class.
Inputs:
x: a (batch_size x 784) numpy array
y: a (batch_size x 10) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 10) numpy array of scores (aka logits)
"""
graph = nn.Graph([self.m, self.b, self.m2, self.b2])
input_x = nn.Input(graph, x)
#============= LAYER 01 ===============#
xm = nn.MatrixMultiply(graph, input_x, self.m)
xm_plus_b = nn.MatrixVectorAdd(graph, xm, self.b)
#============= LAYER 02 ===============#
relu = nn.ReLU(graph, xm_plus_b)
xm2 = nn.MatrixMultiply(graph, relu, self.m2)
xm_plus_b2 = nn.MatrixVectorAdd(graph, xm2, self.b2)
if y is not None:
input_y = nn.Input(graph, y)
loss = nn.SoftmaxLoss(graph, xm_plus_b2, input_y)
return graph
else:
return graph.get_output(xm_plus_b2)
def run(self, states, Q_target=None):
"""
"""
"*** YOUR CODE HERE ***"
graph = nn.Graph(
[self.m1, self.b1, self.m2, self.b2, self.m3, self.b3])
input_x = nn.Input(graph, states)
xm1 = nn.MatrixMultiply(graph, input_x, self.m1)
xm1_plus_b1 = nn.MatrixVectorAdd(graph, xm1, self.b1)
relu1 = nn.ReLU(graph, xm1_plus_b1)
xm2 = nn.MatrixMultiply(graph, relu1, self.m2)
xm2_plus_b2 = nn.MatrixVectorAdd(graph, xm2, self.b2)
relu2 = nn.ReLU(graph, xm2_plus_b2)
xm3 = nn.MatrixMultiply(graph, relu2, self.m3)
m3x_plus_b3 = nn.MatrixVectorAdd(graph, xm3, self.b3)
if Q_target is not None:
"*** YOUR CODE HERE ***"
input_y = nn.Input(graph, Q_target)
loss = nn.SquareLoss(graph, m3x_plus_b3, input_y)
graph.add(loss)
return graph
else:
"*** YOUR CODE HERE ***"
result = graph.get_output(m3x_plus_b3)
return result
def run(self, states, Q_target=None):
"""
Runs the DQN for a batch of states.
The DQN takes the state and computes Q-values for all possible actions
that can be taken. That is, if there are two actions, the network takes
as input the state s and computes the vector [Q(s, a_1), Q(s, a_2)]
When Q_target == None, return the matrix of Q-values currently computed
by the network for the input states.
When Q_target is passed, it will contain the Q-values which the network
should be producing for the current states. You must return a nn.Graph
which computes the training loss between your current Q-value
predictions and these target values, using nn.SquareLoss.
Inputs:
states: a (batch_size x 4) numpy array
Q_target: a (batch_size x 2) numpy array, or None
Output:
(if Q_target is not None) A nn.Graph instance, where the last added
node is the loss
(if Q_target is None) A (batch_size x 2) numpy array of Q-value
scores, for the two actions
"""
"*** YOUR CODE HERE ***"
#size of the input vector
i = states.shape[1] * 4
#to test and modify
h = 200
if not self.w1:
self.w1 = nn.Variable(self.state_size, h)
if not self.w2:
self.w2 = nn.Variable(h, self.num_actions)
if not self.b1:
self.b1 = nn.Variable(h)
if not self.b2:
self.b2 = nn.Variable(self.num_actions)
graph = nn.Graph([self.w1, self.w2, self.b1, self.b2])
inputNodeX = nn.Input(graph, states)
mult1 = nn.MatrixMultiply(graph, inputNodeX, self.w1)
add1 = nn.MatrixVectorAdd(graph, mult1, self.b1)
relu = nn.ReLU(graph, add1)
mult2 = nn.MatrixMultiply(graph, relu, self.w2)
add2 = nn.MatrixVectorAdd(graph, mult2, self.b2)
if Q_target is not None:
inputNodeY = nn.Input(graph, Q_target)
lossNode = nn.SquareLoss(graph, add2, inputNodeY)
graph.add(lossNode)
# print("q target is not none")
return graph
else:
# return [get_action(state, .5),
# print("q target is none")
return graph.get_output(add2)
def run(self, states, Q_target=None):
"""
Runs the DQN for a batch of states.
The DQN takes the state and computes Q-values for all possible actions
that can be taken. That is, if there are two actions, the network takes
as input the state s and computes the vector [Q(s, a_1), Q(s, a_2)]
When Q_target == None, return the matrix of Q-values currently computed
by the network for the input states.
When Q_target is passed, it will contain the Q-values which the network
should be producing for the current states. You must return a nn.Graph
which computes the training loss between your current Q-value
predictions and these target values, using nn.SquareLoss.
Inputs:
states: a (batch_size x 4) numpy array
Q_target: a (batch_size x 2) numpy array, or None
Output:
(if Q_target is not None) A nn.Graph instance, where the last added
node is the loss
(if Q_target is None) A (batch_size x 2) numpy array of Q-value
scores, for the two actions
"""
"*** YOUR CODE HERE ***"
graph = nn.Graph([self.W1, self.W2, self.b1, self.b2])
input_x = nn.Input(graph, states)
xW1mt = nn.MatrixMultiply(graph, input_x, self.W1)
xW1b1Add = nn.MatrixVectorAdd(graph, xW1mt, self.b1)
relu = nn.ReLU(graph, xW1b1Add)
reluMult = nn.MatrixMultiply(graph, relu, self.W2)
total = nn.MatrixVectorAdd(graph, reluMult, self.b2)
# graph = nn.Graph([self.W1])
#Q(s,a) = W1Feature1 + W2feature2 + W3feature3 + W4feature4
# input_x = nn.Input(graph, states)
# W1mult = nn.MatrixMultiply(graph, input_x, self.W1)
#W2mult = nn.MatrixMultiply(graph, input_x, self.W2)
#W3mult = nn.MatrixMultiply(graph, input_x, self.W3)
#W4mult = nn.MatrixMultiply(graph, input_x, self.W4)
#W1W2Add = nn.Add(graph, W1mult, W2mult)
#W3W4Add = nn.Add(graph, W3mult, W4mult)
#total = nn.Add(graph, W1W2Add, W3W4Add)
if Q_target is not None:
"*** YOUR CODE HERE ***"
input_y = nn.Input(graph, Q_target)
loss_node = nn.SquareLoss(graph, total, input_y)
return graph
else:
"*** YOUR CODE HERE ***"
return graph.get_output(total)
def execute_layer(self, input_x, y, graph):
xw1 = nn.MatrixMultiply(graph, input_x, self.w1)
xw1_plus_b1 = nn.MatrixVectorAdd(graph, xw1, self.b1)
relu = nn.ReLU(graph, xw1_plus_b1)
relu_w2 = nn.MatrixMultiply(graph, relu, self.w2)
relu_w2_plus_b2 = nn.MatrixVectorAdd(graph, relu_w2, self.b2)
return graph, relu_w2_plus_b2
def createNN():
if not self.graph:
for i in range(0, self.depth):
#make weight matrix with every layer X by X size
self.weights.append(nn.Variable(len(x), len(x)))
#make bias matrix with each layer being a vector of X by 1 size
self.bias.append(nn.Variable(len(x), 1))
#create graph with initialized weights and bias variables
self.graph = nn.Graph(self.weights +
self.bias) #weight + bias is variable vector
#create input nodes:
input_x = nn.Input(self.graph, x)
input_y = nn.Input(self.graph, y)
#create first layer:
xm = nn.MatrixMultiply(self.graph, self.weights[0], input_x)
xm_plus_b = nn.MatrixVectorAdd(self.graph, xm, self.bias[0])
#create the remaining layers:
for i in range(1, self.depth):
#add nonlinearity for previous layer:
relu = nn.ReLU(self.graph, xm_plus_b)
#create new hidden layer:
xm = nn.MatrixMultiply(self.graph, self.weights[i], relu)
xm_plus_b = nn.MatrixVectorAdd(self.graph, xm, self.bias[i])
#create loss node:
loss = nn.SquareLoss(self.graph, xm_plus_b, input_y)
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
#batch_size = x.shape[0]
# set up the graph
oddRegressionGraph = nn.Graph([self.W1, self.b1, self.W2, self.b2])
input_x = nn.Input(oddRegressionGraph, x)
xW1 = nn.MatrixMultiply(oddRegressionGraph, input_x, self.W1)
xW1_plus_b1 = nn.MatrixVectorAdd(oddRegressionGraph, xW1, self.b1)
ReLU_1 = nn.ReLU(oddRegressionGraph, xW1_plus_b1)
R1W2 = nn.MatrixMultiply(oddRegressionGraph, ReLU_1, self.W2)
R1W2_plus_b2 = nn.MatrixVectorAdd(oddRegressionGraph, R1W2, self.b2)
negx = nn.Input(oddRegressionGraph, x * -1)
negxW1 = nn.MatrixMultiply(oddRegressionGraph, negx, self.W1)
negxW1_plus_b1 = nn.MatrixVectorAdd(oddRegressionGraph, negxW1,
self.b1)
ReLU_2 = nn.ReLU(oddRegressionGraph, negxW1_plus_b1)
R2W2 = nn.MatrixMultiply(oddRegressionGraph, ReLU_2, self.W2)
R2W2_plus_b2 = nn.MatrixVectorAdd(oddRegressionGraph, R2W2, self.b2)
#neg2R2W2_plus_b2 = nn.Input(oddRegressionGraph, oddRegressionGraph.get_output(R2W2_plus_b2)*-2)
#negR2W2_plus_b2 = nn.Add(oddRegressionGraph, R2W2_plus_b2, neg2R2W2_plus_b2)
negR2W2_plus_b2 = nn.Input(
oddRegressionGraph,
oddRegressionGraph.get_output(R2W2_plus_b2) * -1)
sumMatrix = nn.Add(oddRegressionGraph, R1W2_plus_b2, negR2W2_plus_b2)
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
input_y = nn.Input(oddRegressionGraph, y)
sumMatrix_SL_y = nn.SquareLoss(oddRegressionGraph, sumMatrix,
input_y)
return oddRegressionGraph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
return oddRegressionGraph.get_output(sumMatrix)
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
# graph = nn.Graph([self.W1, self.b1, self.W2, self.b2])
# input_x = nn.Input(graph, x)
# xW1 = nn.MatrixMultiply(graph, input_x, self.W1)
# xW1_plus_b1 = nn.MatrixVectorAdd(graph, xW1, self.b1)
# relu = nn.ReLU(graph, xW1_plus_b1)
# xW2 = nn.MatrixMultiply(graph, relu, self.W2)
# sin = nn.MatrixVectorAdd(graph, xW2, self.b2)
"trying 3 layers"
graph = nn.Graph([self.W1, self.b1, self.W2, self.b2, self.W3, self.b3])
input_x = nn.Input(graph, x)
xW1 = nn.MatrixMultiply(graph, input_x, self.W1)
xW1_plus_b1 = nn.MatrixVectorAdd(graph, xW1, self.b1)
relu = nn.ReLU(graph, xW1_plus_b1)
xW2 = nn.MatrixMultiply(graph, relu, self.W2)
xW2_plus_b2 = nn.MatrixVectorAdd(graph, xW2, self.b2)
relu_second = nn.ReLU(graph, xW2_plus_b2)
xW3 = nn.MatrixMultiply(graph, relu_second, self.W3)
sin = nn.MatrixVectorAdd(graph, xW3, self.b3)
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
input_y = nn.Input(graph, y)
loss_node = nn.SquareLoss(graph, sin, input_y)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
return graph.get_output(sin)
def run(self, states, Q_target=None):
"""
Runs the DQN for a batch of states.
The DQN takes the state and computes Q-values for all possible actions
that can be taken. That is, if there are two actions, the network takes
as input the state s and computes the vector [Q(s, a_1), Q(s, a_2)]
When Q_target == None, return the matrix of Q-values currently computed
by the network for the input states.
When Q_target is passed, it will contain the Q-values which the network
should be producing for the current states. You must return a nn.Graph
which computes the training loss between your current Q-value
predictions and these target values, using nn.SquareLoss.
Inputs:
states: a (batch_size x 4) numpy array
Q_target: a (batch_size x 2) numpy array, or None
Output:
(if Q_target is not None) A nn.Graph instance, where the last added
node is the loss
(if Q_target is None) A (batch_size x 2) numpy array of Q-value
scores, for the two actions
"""
"*** YOUR CODE HERE ***"
if not self.graph:
w1 = nn.Variable(4, 50)
w2 = nn.Variable(50, 50)
w3 = nn.Variable(50, 2)
b1 = nn.Variable(1, 50)
b2 = nn.Variable(1, 50)
b3 = nn.Variable(1, 2)
self.list = [w1, w2, w3, b1, b2, b3]
self.graph = nn.Graph(self.l)
self.graph = nn.Graph(self.l)
input_x = nn.Input(self.graph, states)
if Q_target is not None:
input_y = nn.Input(self.graph, Q_target)
mult = nn.MatrixMultiply(self.graph, input_x, self.list[0])
add = nn.MatrixVectorAdd(self.graph, mult, self.list[3])
relu = nn.ReLU(self.graph, add)
mult2 = nn.MatrixMultiply(self.graph, relu, self.list[1])
add2 = nn.MatrixVectorAdd(self.graph, mult2, self.list[4])
relu2 = nn.ReLU(self.graph, add2)
mult3 = nn.MatrixMultiply(self.graph, relu2, self.list[2])
add3 = nn.MatrixVectorAdd(self.graph, mult3, self.list[5])
if Q_target is not None:
"* YOUR CODE HERE *"
loss = nn.SquareLoss(self.graph, add3, input_y)
return self.graph
else:
"* YOUR CODE HERE *"
return self.graph.get_output(self.graph.get_nodes()[-1])
def run(self, states, Q_target=None):
"""
TODO: Question 7 - [Application] Reinforcement Learning
Runs the DQN for a batch of states.
The DQN takes the state and computes Q-values for all possible actions
that can be taken. That is, if there are two actions, the network takes
as input the state s and computes the vector [Q(s, a_1), Q(s, a_2)]
When Q_target == None, return the matrix of Q-values currently computed
by the network for the input states.
When Q_target is passed, it will contain the Q-values which the network
should be producing for the current states. You must return a nn.Graph
which computes the training loss between your current Q-value
predictions and these target values, using nn.SquareLoss.
Inputs:
states: a (batch_size x 4) numpy array
Q_target: a (batch_size x 2) numpy array, or None
Output:
(if Q_target is not None) A nn.Graph instance, where the last added
node is the loss
(if Q_target is None) A (batch_size x 2) numpy array of Q-value
scores, for the two actions
"""
"*** YOUR CODE HERE ***"
#to implement f(x) = relu(x.w1+b1).w2 + b2
graph = nn.Graph([self.w1, self.b1, self.w2, self.b2])
input_x = nn.Input(graph, states)
#input_y = Input(graph, y)
#a = x.w1
a = nn.MatrixMultiply(graph, input_x, self.w1)
#relu(a+b1).w2 + b2
#b = a + b1
b = nn.MatrixVectorAdd(graph, a, self.b1)
#relu(b).w2 + b2
two_layer_relu = nn.ReLU(graph, b)
#c = relu(b).w2
c = nn.MatrixMultiply(graph, two_layer_relu, self.w2)
#d = c + b2
d = nn.MatrixVectorAdd(graph, c, self.b2)
#loss = SquareLoss(graph, xm_plus_b, input_y)
if Q_target is not None:
"*** YOUR CODE HERE ***"
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
input_y = nn.Input(graph, Q_target)
loss = nn.SquareLoss(graph, d, input_y)
return graph
else:
"*** YOUR CODE HERE ***"
return graph.get_output(d)
def run(self, x, y=None):
"""
TODO: Question 6 - [Application] Digit Classification
Runs the model for a batch of examples.
The correct labels are known during training, but not at test time.
When correct labels are available, `y` is a (batch_size x 10) numpy
array. Each row in the array is a one-hot vector encoding the correct
class.
Your model should predict a (batch_size x 10) numpy array of scores,
where higher scores correspond to greater probability of the image
belonging to a particular class. You should use `nn.SoftmaxLoss` as your
training loss.
Inputs:
x: a (batch_size x 784) numpy array
y: a (batch_size x 10) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 10) numpy array of scores (aka logits)
"""
"*** YOUR CODE HERE ***"
#to implement f(x) = relu(x.w1+b1).w2 + b2
graph = nn.Graph([self.w1, self.b1, self.w2, self.b2])
input_x = nn.Input(graph, x)
#input_y = Input(graph, y)
#a = x.w1
a = nn.MatrixMultiply(graph, input_x, self.w1)
#relu(a+b1).w2 + b2
#b = a + b1
b = nn.MatrixVectorAdd(graph, a, self.b1)
#relu(b).w2 + b2
two_layer_relu = nn.ReLU(graph, b)
#c = relu(b).w2
c = nn.MatrixMultiply(graph, two_layer_relu, self.w2)
#d = c + b2
d = nn.MatrixVectorAdd(graph, c, self.b2)
#loss = SquareLoss(graph, xm_plus_b, input_y)
if y is not None:
"*** YOUR CODE HERE ***"
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
input_y = nn.Input(graph, y)
loss = nn.SoftmaxLoss(graph, d, input_y)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
return graph.get_output(d)
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
# positive
graph = nn.Graph([self.w1, self.b1, self.w2, self.b2])
input_x = nn.Input(graph, x)
mul1 = nn.MatrixMultiply(graph, input_x, self.w1)
add1 = nn.MatrixVectorAdd(graph, mul1, self.b1)
hidden_output = nn.ReLU(graph, add1)
mul2 = nn.MatrixMultiply(graph, hidden_output, self.w2)
add2 = nn.MatrixVectorAdd(graph, mul2, self.b2)
# negative
neg_input_x = nn.Input(graph, np.dot(x, np.array([[-1]])))
neg_mul1 = nn.MatrixMultiply(graph, neg_input_x, self.w1)
neg_add1 = nn.MatrixVectorAdd(graph, neg_mul1, self.b1)
neg_hidden_output = nn.ReLU(graph, neg_add1)
neg_mul2 = nn.MatrixMultiply(graph, neg_hidden_output, self.w2)
neg_add2 = nn.MatrixVectorAdd(graph, neg_mul2, self.b2)
neg_mul3 = nn.MatrixMultiply(graph, neg_add2,
nn.Input(graph, np.array([[-1.0]])))
multiLayer = nn.MatrixVectorAdd(graph, add2, neg_mul3)
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
inputY = nn.Input(graph, y)
loss = nn.SquareLoss(graph, multiLayer, inputY)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
return graph.get_output(multiLayer)
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
graph = nn.Graph([self.w1, self.w2, self.b1, self.b2])
input_x = nn.Input(graph, x)
xw1 = nn.MatrixMultiply(graph, input_x, self.w1)
xw1_b1 = nn.MatrixVectorAdd(graph, xw1, self.b1)
relu = nn.ReLU(graph, xw1_b1)
xw2 = nn.MatrixMultiply(graph, relu, self.w2)
xw2_b2 = nn.MatrixVectorAdd(graph, xw2, self.b2)
negate = nn.Input(graph, np.array([[-1.]]))
neg_x = nn.MatrixMultiply(graph, input_x, negate)
neg_xw1 = nn.MatrixMultiply(graph, neg_x, self.w1)
neg_xw1_b1 = nn.MatrixVectorAdd(graph, neg_xw1, self.b1)
neg_relu = nn.ReLU(graph, neg_xw1_b1)
neg_xw2 = nn.MatrixMultiply(graph, neg_relu, self.w2)
neg_xw2_b2 = nn.MatrixVectorAdd(graph, neg_xw2, self.b2)
neg_neg = nn.MatrixMultiply(graph, neg_xw2_b2, negate)
final = nn.MatrixVectorAdd(graph, xw2_b2, neg_neg)
# print graph.get_output(graph.get_nodes()[-1])
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
input_y = nn.Input(graph, y)
loss = nn.SquareLoss(graph, final, input_y)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
# # print graph.get_output(graph.get_nodes()[-1])
return graph.get_output(graph.get_nodes()[-1])
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct labels are known during training, but not at test time.
When correct labels are available, `y` is a (batch_size x 10) numpy
array. Each row in the array is a one-hot vector encoding the correct
class.
Your model should predict a (batch_size x 10) numpy array of scores,
where higher scores correspond to greater probability of the image
belonging to a particular class. You should use `nn.SoftmaxLoss` as your
training loss.
Inputs:
x: a (batch_size x 784) numpy array
y: a (batch_size x 10) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 10) numpy array of scores (aka logits)
"""
"*** YOUR CODE HERE ***"
if len(x) == 1:
return 0
if not self.graph:
w1 = nn.Variable(784, 500)
w2 = nn.Variable(500, 500)
w3 = nn.Variable(500, 10)
b1 = nn.Variable(1, 500)
b2 = nn.Variable(1, 500)
b3 = nn.Variable(1, 10)
self.l = [w1, w2, w3, b1, b2, b3]
self.graph = nn.Graph(self.l)
self.graph = nn.Graph(self.l)
input_x = nn.Input(self.graph, x) #Tx784
if y is not None: #<--- THIS LITTLE CONDITIONAL SO IMPORTANT HFS
input_y = nn.Input(self.graph, y)
mult = nn.MatrixMultiply(self.graph, input_x, self.l[0]) #Tx50
add = nn.MatrixVectorAdd(self.graph, mult, self.l[3])
relu = nn.ReLU(self.graph, add)
mult2 = nn.MatrixMultiply(self.graph, relu, self.l[1]) #Tx50
add2 = nn.MatrixVectorAdd(self.graph, mult2, self.l[4]) #Tx50
relu2 = nn.ReLU(self.graph, add2)
mult3 = nn.MatrixMultiply(self.graph, relu2, self.l[2])
add3 = nn.MatrixVectorAdd(self.graph, mult3, self.l[5])
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
loss = nn.SoftmaxLoss(self.graph, add3, input_y)
return self.graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
#print(self.graph.get_output(self.graph.get_nodes()[-1]))
return self.graph.get_output(self.graph.get_nodes()[-1])
def run(self, x, y=None):
"""
TODO: Question 5 - [Application] OddRegression
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
graph = nn.Graph([self.w1, self.w2, self.b1, self.b2])
input_x = nn.Input(graph, x)
xm = nn.MatrixMultiply(graph, input_x, self.w1)
xm_plus_b = nn.MatrixVectorAdd(graph, xm, self.b1)
relu = nn.ReLU(graph, xm_plus_b)
w2mul = nn.MatrixMultiply(graph, relu, self.w2)
b2add = nn.MatrixVectorAdd(graph, w2mul, self.b2)
graph2 = nn.Graph([self.w1, self.w2, self.b1, self.b2])
input_x2 = nn.Input(graph2, x * -1)
inputsmall = nn.Input(graph2, np.array([-1.0]))
xm2 = nn.MatrixMultiply(graph2, input_x2, self.w1)
xm_plus_b2 = nn.MatrixVectorAdd(graph2, xm2, self.b1)
relu2 = nn.ReLU(graph2, xm_plus_b2)
w2mul2 = nn.MatrixMultiply(graph2, relu2, self.w2)
b2add2 = nn.MatrixVectorAdd(graph2, w2mul2, self.b2)
#final = nn.MatrixMultiply(graph2, b2add2, inputsmall)
fuckinput = nn.Input(graph,
graph2.get_output(graph2.get_nodes()[-1]) * -1)
realfinal = nn.MatrixVectorAdd(graph, b2add, fuckinput)
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
input_y = nn.Input(graph, y)
loss = nn.SquareLoss(graph, realfinal, input_y)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
#print graph.get_output(graph.get_nodes()[-1])
#print graph.get_output(input_x)
return graph.get_output(graph.get_nodes()[-1])
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
self.graph = nn.Graph([self.m1, self.m2, self.b1, self.b2])
input_x = nn.Input(self.graph, x)
matmult = nn.MatrixMultiply(self.graph, input_x, self.m1)
matadd = nn.MatrixVectorAdd(self.graph, matmult, self.b1)
relu = nn.ReLU(self.graph, matadd)
matmult = nn.MatrixMultiply(self.graph, relu, self.m2)
matadd1 = nn.MatrixVectorAdd(self.graph, matmult, self.b2)
matrixNeg = nn.Input(self.graph, np.array([[-1.]]))
matmultNeg = nn.MatrixMultiply(self.graph, input_x, matrixNeg)
matmult = nn.MatrixMultiply(self.graph, matmultNeg, self.m1)
matadd = nn.MatrixVectorAdd(self.graph, matmult, self.b1)
relu = nn.ReLU(self.graph, matadd)
matmult = nn.MatrixMultiply(self.graph, relu, self.m2)
matadd2 = nn.MatrixVectorAdd(self.graph, matmult, self.b2)
negMat = nn.MatrixMultiply(self.graph, matadd2, matrixNeg)
func = nn.MatrixVectorAdd(self.graph, matadd1, negMat)
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
input_y = nn.Input(self.graph, y)
loss = nn.SquareLoss(self.graph, func, input_y)
# self.graph.add(loss)
return self.graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
return self.graph.get_output(func)
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct labels are known during training, but not at test time.
When correct labels are available, y is a (batch_size x 10) numpy
array. Each row in the array is a one-hot vector encoding the correct
class.
Your model should predict a (batch_size x 10) numpy array of scores,
where higher scores correspond to greater probability of the image
belonging to a particular class. You should use nn.SoftmaxLoss as your
training loss.
Inputs:
x: a (batch_size x 784) numpy array
y: a (batch_size x 10) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 10) numpy array of scores (aka logits)
"""
"* YOUR CODE HERE *"
graph = nn.Graph([
self.weight1, self.bias1, self.weight2, self.bias2, self.weight3,
self.bias3, self.weight4, self.bias4, self.weight5, self.bias5
])
input_x = nn.Input(graph, x)
xw1 = nn.MatrixMultiply(graph, input_x, self.weight1)
plus1b1 = nn.MatrixVectorAdd(graph, xw1, self.bias1)
relu1 = nn.ReLU(graph, plus1b1)
relu1_2 = nn.MatrixMultiply(graph, relu1, self.weight2)
plus2b2 = nn.MatrixVectorAdd(graph, relu1_2, self.bias2)
relu2 = nn.ReLU(graph, plus2b2)
relu2_3 = nn.MatrixMultiply(graph, relu2, self.weight3)
plus3b3 = nn.MatrixVectorAdd(graph, relu2_3, self.bias3)
relu3 = nn.ReLU(graph, plus3b3)
relu3_4 = nn.MatrixMultiply(graph, relu3, self.weight4)
plus4b4 = nn.MatrixVectorAdd(graph, relu3_4, self.bias4)
relu4 = nn.ReLU(graph, plus4b4)
relu4_5 = nn.MatrixMultiply(graph, relu4, self.weight5)
plus5b5 = nn.MatrixVectorAdd(graph, relu4_5, self.bias5)
if y is not None:
"* YOUR CODE HERE *"
input_y = nn.Input(graph, y)
loss = nn.SoftmaxLoss(graph, plus5b5, input_y)
return graph
else:
"* YOUR CODE HERE *"
return graph.get_output(plus5b5)
def run(self, states, Q_target=None):
"""
TODO: Question 7 - [Application] Reinforcement Learning
Runs the DQN for a batch of states.
The DQN takes the state and computes Q-values for all possible actions
that can be taken. That is, if there are two actions, the network takes
as input the state s and computes the vector [Q(s, a_1), Q(s, a_2)]
When Q_target == None, return the matrix of Q-values currently computed
by the network for the input states.
When Q_target is passed, it will contain the Q-values which the network
should be producing for the current states. You must return a nn.Graph
which computes the training loss between your current Q-value
predictions and these target values, using nn.SquareLoss.
Inputs:
states: a (batch_size x 4) numpy array
Q_target: a (batch_size x 2) numpy array, or None
Output:
(if Q_target is not None) A nn.Graph instance, where the last added
node is the loss
(if Q_target is None) A (batch_size x 2) numpy array of Q-value
scores, for the two actions
"""
"*** YOUR CODE HERE ***"
hidden_layer_size = 150
if not self.w1:
self.w1 = nn.Variable(self.state_size, hidden_layer_size)
if not self.w2:
self.w2 = nn.Variable(hidden_layer_size, self.num_actions)
if not self.b1:
self.b1 = nn.Variable(hidden_layer_size)
if not self.b2:
self.b2 = nn.Variable(self.num_actions)
g = nn.Graph([self.w1, self.w2, self.b1, self.b2])
result = nn.MatrixVectorAdd(
g,
nn.MatrixMultiply(
g,
nn.ReLU(
g,
nn.MatrixVectorAdd(
g, nn.MatrixMultiply(g, nn.Input(g, states), self.w1),
self.b1)), self.w2), self.b2)
if Q_target is not None:
"*** YOUR CODE HERE ***"
g.add(nn.SquareLoss(g, result, nn.Input(g, Q_target)))
return g
else:
"*** YOUR CODE HERE ***"
return g.get_output(result)
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
graph = nn.Graph(self.variables)
negative1 = -1*np.ones((1,1))
input_x = nn.Input(graph, x)
neg_1 = nn.Input(graph, negative1)
"""First we do the positives"""
xw1 = nn.MatrixMultiply(graph, input_x, self.variables[0])
sumxw1b1 = nn.MatrixVectorAdd(graph, xw1, self.variables[1])
relu = nn.ReLU(graph, sumxw1b1)
reluW2 = nn.MatrixMultiply(graph, relu, self.variables[2])
"""Now we do the negatives"""
negx = nn.MatrixMultiply(graph, input_x, neg_1)
nxw1 = nn.MatrixMultiply(graph, negx, self.variables[0])
sumnxw1 = nn.MatrixVectorAdd(graph, nxw1, self.variables[1])
nrelu = nn.ReLU(graph, sumnxw1)
nreluW2 = nn.MatrixMultiply(graph, nrelu, self.variables[2])
"""Set the negative value of negative x to negative"""
nsumNRW2b2 = nn.MatrixMultiply(graph, nreluW2, neg_1)
"""Add the two sums together"""
totalSum = nn.Add(graph, reluW2, nsumNRW2b2)
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
input_y = nn.Input(graph, y)
loss = nn.SquareLoss(graph, totalSum, input_y)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
nodes = graph.get_nodes()
lastnode = nodes[-1]
out = graph.get_output(lastnode)
return out
def run(self, x, y=None):
"""
TODO: Question 5 - [Application] OddRegression
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
# calculates g(x)
graph = nn.Graph([self.w_one, self.b_one, self.w_two, self.b_two])
input_x = nn.Input(graph, x)
xw = nn.MatrixMultiply(graph, input_x, self.w_one)
xw_plus_b = nn.MatrixVectorAdd(graph, xw, self.b_one)
relu = nn.ReLU(graph, xw_plus_b)
reluw = nn.MatrixMultiply(graph, relu, self.w_two)
reluw_plus_b = nn.MatrixVectorAdd(graph, reluw, self.b_two)
# calculates g(-x)
negone = nn.Input(graph, np.array([[-1.0]]))
neg_x = nn.MatrixMultiply(graph, input_x, negone)
negxw = nn.MatrixMultiply(graph, neg_x, self.w_one)
negxw_plus_b = nn.MatrixVectorAdd(graph, negxw, self.b_one)
negrelu = nn.ReLU(graph, negxw_plus_b)
negreluw = nn.MatrixMultiply(graph, negrelu, self.w_two)
negreluw_plus_b = nn.MatrixVectorAdd(graph, negreluw, self.b_two)
#g(x)-(g(-x))
negG = nn.MatrixMultiply(graph, negreluw_plus_b, negone)
oddFunc = nn.Add(graph, reluw_plus_b, negG)
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
input_y = nn.Input(graph, y)
loss = nn.SquareLoss(graph, oddFunc, input_y)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
return graph.get_output(oddFunc)
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
graph = nn.Graph([self.w_one, self.w_two, self.b_one, self.b_two])
input_i = nn.Input(graph, x)
mul = nn.MatrixMultiply(graph, input_i, self.w_one)
add = nn.MatrixVectorAdd(graph, mul, self.b_one)
relu = nn.ReLU(graph, add)
# -x
tempx = np.negative(x)
input_i2 = nn.Input(graph, tempx)
multiply = nn.MatrixMultiply(graph, input_i2, self.w_one)
addition = nn.MatrixVectorAdd(graph, multiply, self.b_one)
relu2 = nn.ReLU(graph, addition)
# -f(-x)
arr = np.negative(np.identity(self.hidden_size))
negativeOne = nn.Input(graph, arr)
negatedRelu = nn.MatrixMultiply(graph, relu2, negativeOne)
# f(x) + (-f(-x))
inside = nn.MatrixVectorAdd(graph, relu, negatedRelu)
mul2 = nn.MatrixMultiply(graph, inside, self.w_two)
#add2 = nn.MatrixVectorAdd(graph, mul2, self.b_two)
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
y_in = nn.Input(graph, y)
loss = nn.SquareLoss(graph, mul2, y_in)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
return graph.get_output(mul2)
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
graph = nn.Graph([self.w1,self.b1,self.w2])
xInput = nn.Input(graph,x)
negXInput = nn.Input(graph,-x)
yInput = nn.Input(graph,y)
layer1 = nn.MatrixMultiply(graph, xInput, self.w1)
negLayer1 = nn.MatrixMultiply(graph, negXInput, self.w1)
layer2 = nn.MatrixVectorAdd(graph, layer1, self.b1)
negLayer2 = nn.MatrixVectorAdd(graph, negLayer1, self.b1)
layer3 = nn.ReLU(graph, layer2)
negLayer3 = nn.ReLU(graph, negLayer2)
combinedLayer = nn.Subtract(graph, layer3, negLayer3)
layer4 = nn.MatrixMultiply(graph, combinedLayer, self.w2)
layer5 = nn.SquareLoss(graph, layer4, yInput)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
graph = nn.Graph([self.w1,self.b1,self.w2])
xInput = nn.Input(graph,x)
negXInput = nn.Input(graph,-x)
layer1 = nn.MatrixMultiply(graph, xInput, self.w1)
negLayer1 = nn.MatrixMultiply(graph, negXInput, self.w1)
layer2 = nn.MatrixVectorAdd(graph, layer1, self.b1)
negLayer2 = nn.MatrixVectorAdd(graph, negLayer1, self.b1)
layer3 = nn.ReLU(graph, layer2)
negLayer3 = nn.ReLU(graph, negLayer2)
combinedLayer = nn.Subtract(graph, layer3, negLayer3)
layer4 = nn.MatrixMultiply(graph, combinedLayer, self.w2)
return graph.get_output(layer4)
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
graph = nn.Graph([self.w1, self.b1, self.w2, self.b2])
input_x = nn.Input(graph, x)
nn_mm_w1 = nn.MatrixMultiply(graph, input_x, self.w1)
nn_mm_b1 = nn.MatrixVectorAdd(graph, nn_mm_w1, self.b1)
nn_rl_b1 = nn.ReLU(graph, nn_mm_b1)
nn_mm_w2 = nn.MatrixMultiply(graph, nn_rl_b1, self.w2)
nn_mm_b2 = nn.MatrixVectorAdd(graph, nn_mm_w2, self.b2)
mx = np.dot(-1, x)
nn_mx = nn.Input(graph, mx)
mm_mx = nn.MatrixMultiply(graph, nn_mx, self.w1)
mm_mx_b1 = nn.MatrixVectorAdd(graph, mm_mx, self.b1)
rl_mx_b1 = nn.ReLU(graph, mm_mx_b1)
mm_mx_w2 = nn.MatrixMultiply(graph, rl_mx_b1, self.w2)
mm_mx_b2 = nn.MatrixVectorAdd(graph, mm_mx_w2, self.b2)
new_x = nn.Subtract(graph, nn_mm_b2, mm_mx_b2)
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
nn_y = nn.Input(graph, y)
loss = nn.SquareLoss(graph, new_x, nn_y)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
return graph.get_output(new_x)
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
sign = x < 0
x = np.where(sign, -x, x)
graph = nn.Graph(
[self.m0, self.b0, self.m1, self.b1, self.m2, self.b2])
input_x = nn.Input(graph, x)
t = nn.MatrixMultiply(graph, input_x, self.m0)
t = nn.MatrixVectorAdd(graph, t, self.b0)
t = nn.ReLU(graph, t)
t = nn.MatrixMultiply(graph, t, self.m1)
t = nn.MatrixVectorAdd(graph, t, self.b1)
t = nn.ReLU(graph, t)
t = nn.MatrixMultiply(graph, t, self.m2)
t = nn.MatrixVectorAdd(graph, t, self.b2)
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
y = np.where(sign, -y, y)
input_y = nn.Input(graph, y)
loss = nn.SquareLoss(graph, t, input_y)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
res = graph.outputs[graph.get_nodes()[-1]]
res = np.where(sign, -res, res)
res = np.where(x == 0, np.zeros_like(res), res)
return res
def add_three_edges(input0, graph, lists):
mult = nn.MatrixMultiply(graph, input0, lists[0])
add = nn.MatrixVectorAdd(graph, mult, lists[3])
relu = nn.ReLU(graph, add)
mult2 = nn.MatrixMultiply(graph, relu, lists[1])
add2 = nn.MatrixVectorAdd(graph, mult2, lists[4])
relu2 = nn.ReLU(graph, add2)
mult3 = nn.MatrixMultiply(graph, relu2, lists[2])
add3 = nn.MatrixVectorAdd(graph, mult3, lists[5])
return add3
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
"*** YOUR CODE HERE ***"
graph = nn.Graph(self.param_w + self.param_b)
inX = [nn.Input(graph, x), nn.Input(graph, -1 * x)]
last = inX
for i in range(self.num_layers):
newLast = last
for j in range(len(last)):
multNode = nn.MatrixMultiply(graph, last[j], self.param_w[i])
addNode = nn.MatrixVectorAdd(graph, multNode, self.param_b[i])
if i != self.num_layers - 1:
reluNode = nn.ReLU(graph, addNode)
newLast[j] = reluNode
else:
newLast[j] = addNode
last = newLast
# f(x) = g(x) + -1 * g(-x)
neg = nn.Input(graph, np.array(-1.0))
multNode = nn.MatrixMultiply(graph, inX[1], neg)
addNode = nn.MatrixVectorAdd(graph, inX[0], multNode)
final = addNode
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
"*** YOUR CODE HERE ***"
inY = nn.Input(graph, y)
loss = nn.SquareLoss(graph, final, inY)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
"*** YOUR CODE HERE ***"
return graph.get_output(final)
def run(self, x, y=None):
"""
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
graph = nn.Graph([self.w1, self.w2, self.b1, self.b2])
#pos
a_graph = nn.Graph([self.w1, self.w2, self.b1, self.b2])
a_input_x = nn.Input(graph, x)
a_mult1 = nn.MatrixMultiply(graph, a_input_x, self.w1)
a_add1 = nn.MatrixVectorAdd(graph, a_mult1, self.b1)
a_relu1 = nn.ReLU(graph, a_add1)
a_mult2 = nn.MatrixMultiply(graph, a_relu1, self.w2)
a_add2 = nn.MatrixVectorAdd(graph, a_mult2, self.b2)
#neg
b_input_x = nn.Input(graph, np.dot(-1, x))
b_mult1 = nn.MatrixMultiply(graph, b_input_x, self.w1)
b_add1 = nn.MatrixVectorAdd(graph, b_mult1, self.b1)
b_relu1 = nn.ReLU(graph, b_add1)
b_mult2 = nn.MatrixMultiply(graph, b_relu1, self.w2)
b_add2 = nn.MatrixVectorAdd(graph, b_mult2, self.b2)
b_output = graph.get_output(b_add2)
neg_b_output = -1 * b_output
neg_matrix = np.zeros(np.shape(b_output)[1])
neg = np.negative(np.identity(np.shape(b_output)[1]))
neg_b_add2 = nn.Input(graph, neg)
b_mult3 = nn.MatrixMultiply(graph, b_add2, neg_b_add2)
result = nn.Add(graph, a_add2, b_mult3)
if y is not None:
input_y = nn.Input(graph, y)
loss = nn.SquareLoss(graph, result, input_y)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
return graph.get_output(result)
def run(self, x, y=None):
"""
TODO: Question 4 - [Application] Regression
Runs the model for a batch of examples.
The correct outputs `y` are known during training, but not at test time.
If correct outputs `y` are provided, this method must construct and
return a nn.Graph for computing the training loss. If `y` is None, this
method must instead return predicted y-values.
Inputs:
x: a (batch_size x 1) numpy array
y: a (batch_size x 1) numpy array, or None
Output:
(if y is not None) A nn.Graph instance, where the last added node is
the loss
(if y is None) A (batch_size x 1) numpy array of predicted y-values
Note: DO NOT call backprop() or step() inside this method!
"""
graph = nn.Graph(
[self.W1, self.b1, self.W2, self.b2, self.W3, self.b3])
#build x input node
input_x = nn.Input(graph, x)
#build the graph
W1x = nn.MatrixMultiply(graph, input_x, self.W1)
W1b = nn.MatrixVectorAdd(graph, W1x, self.b1)
W1Relu = nn.ReLU(graph, W1b)
W2x = nn.MatrixMultiply(graph, W1Relu, self.W2)
W2b = nn.MatrixVectorAdd(graph, W2x, self.b2)
W2Relu = nn.ReLU(graph, W2b)
W3x = nn.MatrixMultiply(graph, W2Relu, self.W3)
W3b = nn.MatrixVectorAdd(graph, W3x, self.b3)
yHat = W3b
if y is not None:
# At training time, the correct output `y` is known.
# Here, you should construct a loss node, and return the nn.Graph
# that the node belongs to. The loss node must be the last node
# added to the graph.
#build y input node
input_y = nn.Input(graph, y)
Loss = nn.SquareLoss(graph, yHat, input_y)
return graph
else:
# At test time, the correct output is unknown.
# You should instead return your model's prediction as a numpy array
return graph.get_output(yHat)
def run(self, states, Q_target=None):
"""
Runs the DQN for a batch of states.
The DQN takes the state and computes Q-values for all possible actions
that can be taken. That is, if there are two actions, the network takes
as input the state s and computes the vector [Q(s, a_1), Q(s, a_2)]
When Q_target == None, return the matrix of Q-values currently computed
by the network for the input states.
When Q_target is passed, it will contain the Q-values which the network
should be producing for the current states. You must return a nn.Graph
which computes the training loss between your current Q-value
predictions and these target values, using nn.SquareLoss.
Inputs:
states: a (batch_size x 4) numpy array
Q_target: a (batch_size x 2) numpy array, or None
Output:
(if Q_target is not None) A nn.Graph instance, where the last added
node is the loss
(if Q_target is None) A (batch_size x 2) numpy array of Q-value
scores, for the two actions
"""
"*** YOUR CODE HERE ***"
graph = nn.Graph([
self.m0, self.b0, self.m1, self.b1, self.m2, self.b2, self.m3,
self.b3
])
input_x = nn.Input(graph, states)
t = nn.MatrixMultiply(graph, input_x, self.m0)
t = nn.MatrixVectorAdd(graph, t, self.b0)
t = nn.ReLU(graph, t)
t = nn.MatrixMultiply(graph, t, self.m1)
t = nn.MatrixVectorAdd(graph, t, self.b1)
t = nn.ReLU(graph, t)
t = nn.MatrixMultiply(graph, t, self.m2)
t = nn.MatrixVectorAdd(graph, t, self.b2)
t = nn.ReLU(graph, t)
t = nn.MatrixMultiply(graph, t, self.m3)
t = nn.MatrixVectorAdd(graph, t, self.b3)
if Q_target is not None:
"*** YOUR CODE HERE ***"
input_y = nn.Input(graph, Q_target)
loss = nn.SquareLoss(graph, t, input_y)
return graph
else:
"*** YOUR CODE HERE ***"
res = graph.outputs[graph.get_nodes()[-1]]
return res
def nueral_net(graph, pass_in):
xdotWpos = nn.MatrixMultiply(graph, pass_in, self.W1)
xWplusb1pos = nn.MatrixVectorAdd(graph, xdotWpos, self.b1)
relu_xdotWpos = nn.ReLU(graph, xWplusb1pos)
second_layerpos = nn.MatrixMultiply(graph, relu_xdotWpos, self.W2)
second_layerplusbpos = nn.MatrixVectorAdd(graph, second_layerpos,
self.b2)
return graph, second_layerplusbpos