A2C not converge as the loss explode - pytorch

I'm experimenting with Advantage Actor Critic algorithm, and the loss explode exponentially.
like
iteration
actor_loss
critic_loss
17
-0.072878
0.003239
78
-256202.2
254041.0
428
-1.02e+17
7.17e+16
The actor network and critic network share the same base network and have separate consequent layers.
I've checked that the predicted value is also exploding to (negative) the same order of magnitude as the losses.
The update session is like this:
for log_prob, value, R in zip(log_prob_list, value_list, returns):
advantage = R - value
# actor loss
policy = -log_prob * advantage
policy_losses.append(policy.mean())
# critic loss
value_losses.append(F.smooth_l1_loss(value, R))
loss1 = torch.stack(policy_losses).sum()
loss2 = torch.stack(value_losses).sum()
loss = loss1 + beta * loss2 - 0.001 * total_entropy
optimizer.zero_grad()
loss.backward()
optimizer.step()
I'm wondering what's wrong, somehow I think the explosion is reasonable as the network outputs low value close to negative infinity which makes advantage close to positive infinity and policy_loss -> negative infinity.
I've tried increase beta to even 1e4 but this didn't help.
Can anyone correct/help me?

Related

Pytorch Repeating loss and AUC- When using cumulative loss

I am using PyTorch to accumulate and add losses, and then implement backpropagation(loss.backward()) at the end.
At this time, the loss is not updated and remains almost the same, and the AUC repeats exactly the same. Are there any points I haven't considered when using cumulative losses?
Thank you so much for any reply. :)
Below is the loss calculation that occurs in one batch.
opt.zero_grad()
for s in range(len(qshft)):
for a in range(len(qshft[0])):
if(m[s][a]):
y_pred = (y[s][a] * one_hot(qshft[s].long(), self.num_q)).sum(-1)
y_pred = torch.masked_select(y_pred, m[s])
t = torch.masked_select(rshft[s], m[s])
loss += binary_cross_entropy(y_pred, t).clone().detach().requires_grad_(True)
count += 1
loss = torch.tensor(loss/count,requires_grad=True)
loss.backward()
opt.step()
loss_mean.append(loss.detach().cpu().numpy())
Your following operation of detach removes the computation graph, so the loss.backward() and opt.step() won't update your weights which results in repeating loss and AUC.
loss += binary_cross_entropy(y_pred, t).clone().detach().requires_grad_(True)
You can do
loss += binary_cross_entropy(y_pred, t)
and change
loss = torch.tensor(loss/count,requires_grad=True)
to
loss = loss/count
But make sure you reset count and loss to 0 every time you go into this part.

Trying to accumulate gradients in Pytorch, but getting RuntimeError when calling loss.backward

I'm trying to train a model in Pytorch, and I'd like to have a batch size of 8, but due to memory limitations, I can only have a batch size of at most 4. I've looked all around and read a lot about accumulating gradients, and it seems like the solution to my problem.
However, I seem to have trouble implementing it. Every time I run the code I get RuntimeError: Trying to backward through the graph a second time. I don't understand why since my code looks like all these other examples I've seen (unless I'm just missing something major):
https://stackoverflow.com/a/62076913/1227353
https://medium.com/huggingface/training-larger-batches-practical-tips-on-1-gpu-multi-gpu-distributed-setups-ec88c3e51255
https://discuss.pytorch.org/t/why-do-we-need-to-set-the-gradients-manually-to-zero-in-pytorch/4903/20
One caveat is that the labels for my images are all different size, so I can't send the output batch and the label batch into the loss function; I have to iterate over them together. This is what an epoch looks like (it's been pared down for the sake of brevity):
# labels_batch contains labels of different sizes
for batch_idx, (inputs_batch, labels_batch) in enumerate(dataloader):
outputs_batch = model(inputs_batch)
# have to do this because labels can't be stacked into a tensor
for output, label in zip(outputs_batch, labels_batch):
output_scaled = interpolate(...) # make output match label size
loss = train_criterion(output_scaled, label) / (BATCH_SIZE * 2)
loss.backward()
if batch_idx % 2 == 1:
optimizer.step()
optimizer.zero_grad()
Is there something I'm missing? If I do the following I also get an error:
# labels_batch contains labels of different sizes
for batch_idx, (inputs_batch, labels_batch) in enumerate(dataloader):
outputs_batch = model(inputs_batch)
# CHANGE: we're gonna accumulate losses manually
batch_loss = 0
# have to do this because labels can't be stacked into a tensor
for output, label in zip(outputs_batch, labels_batch):
output_scaled = interpolate(...) # make output match label size
loss = train_criterion(output_scaled, label) / (BATCH_SIZE * 2)
batch_loss += loss # CHANGE: accumulate!
# CHANGE: do backprop outside for loop
batch_loss.backward()
if batch_idx % 2 == 1:
optimizer.step()
optimizer.zero_grad()
The error I get in this case is RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn. This happens when the next epoch starts though... (INCORRECT, SEE EDIT BELOW)
How can I train my model with gradient accumulation? Or am I doomed to train with a batch size of 4 or less?
Oh and as a side question, does the location of where I put loss.backward() affect what I need to normalize the loss by? Or is it always normalized by BATCH_SIZE * 2?
EDIT:
The second code segment was getting an error due to the fact that I was doing torch.set_grad_enabled(phase == 'train') but I had forgotten to wrap the call to batch_loss.backward() with an if phase == 'train'... my bad
So now the second segment of code seems to work and do gradient accumulation, but why doesn't the first bit of code work? It feel equivalent to setting BATCH_SIZE as 1. Furthermore, I'm creating a new loss object each time, so shouldn't the calls to backward() operate on different graphs entirely?
It seems you have two issues here, you said you couldn't have batch_size=8 because of memory limitations but later state that your labels are not of the same size. The latter seems much more important than the former. Anyway, I will try to answer your questions best I can.
How can I train my model with gradient accumulation? Or am I doomed to train with a batch size of 4 or less?
You want to call .backward() on every loop cycle otherwise the batch will have no effect on the training. You can then call step() and zero_grad() only when batch_idx % 2 is True (i.e. for every other batch).
Here's an example which accumulates the gradient, not the loss:
model = nn.Linear(10, 3)
optim = torch.optim.SGD(model.parameters(), lr=0.1)
ds = TensorDataset(torch.rand(100, 10), torch.rand(100, 3))
dl = DataLoader(ds, batch_size=4)
for i, (x, y) in enumerate(dl):
y_hat = model(x)
loss = F.l1_loss(y_hat, y) / 2
loss.backward()
if i % 2:
optim.step()
optim.zero_grad()
Note this approach is different to accumulating the loss, and back-propagating only all batches (or part of the batches) have gone through the network. In the example above we backpropagate every 4 datapoints and updating the model every 8 datapoints.
Oh and as a side question, does the location of where I put loss.backward() affect what I need to normalize the loss by? Or is it always normalized by BATCH_SIZE * 2?
Usually torch's built-in losses have reduction='mean' set as default. This means the loss gets averaged over all batch elements that contributed to calculating the loss. So this will depend on your loss implementation.
However if you are using gradient accumalation, then yes you will need to average your loss by the number of accumulation steps (here loss = F.l1_loss(y_hat, y) / 2). Since your gradients will be accumulated twice.
To read more about this, I recommend taking a look at this other SO post.

Pytorch: Custom Loss involving Norm of End-to-End Jacobian

Cross posting from Pytorch discussion boards
I want to train a network using a modified loss function that has both a typical classification loss (e.g. nn.CrossEntropyLoss) as well as a penalty on the Frobenius norm of the end-to-end Jacobian (i.e. if f(x) is the output of the network, \nabla_x f(x)).
I’ve implemented a model that can successfully learn using nn.CrossEntropyLoss. However, when I try adding the second loss function (by doing two backwards passes), my training loop runs, but the model never learns. Furthermore, if I calculate the end-to-end Jacobian, but don’t include it in the loss function, the model also never learns. At a high level, my code does the following:
Forward pass to get predicted classes, yhat, from inputs x
Call yhat.backward(torch.ones(appropriate shape), retain_graph=True)
Jacobian norm = x.grad.data.norm(2)
Set loss equal to classification loss + scalar coefficient * jacobian norm
Run loss.backward()
I suspect that I’m misunderstanding how backward() works when run twice, but I haven’t been able to find any good resources to clarify this.
Too much is required to produce a working example, so I’ve tried to extract the relevant code:
def train_model(model, train_dataloader, optimizer, loss_fn, device=None):
if device is None:
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.train()
train_loss = 0
correct = 0
for batch_idx, (batch_input, batch_target) in enumerate(train_dataloader):
batch_input, batch_target = batch_input.to(device), batch_target.to(device)
optimizer.zero_grad()
batch_input.requires_grad_(True)
model_batch_output = model(batch_input)
loss = loss_fn(model_output=model_batch_output, model_input=batch_input, model=model, target=batch_target)
train_loss += loss.item() # sum up batch loss
loss.backward()
optimizer.step()
and
def end_to_end_jacobian_loss(model_output, model_input):
model_output.backward(
torch.ones(*model_output.shape),
retain_graph=True)
jacobian = model_input.grad.data
jacobian_norm = jacobian.norm(2)
return jacobian_norm
Edit 1: I swapped my previous implementation with .backward() to autograd.grad and it apparently works! What's the difference?
def end_to_end_jacobian_loss(model_output, model_input):
jacobian = autograd.grad(
outputs=model_output['penultimate_layer'],
inputs=model_input,
grad_outputs=torch.ones(*model_output['penultimate_layer'].shape),
retain_graph=True,
only_inputs=True)[0]
jacobian_norm = jacobian.norm(2)
return jacobian_norm

How to properly update the weights in PyTorch?

I'm trying to implement the gradient descent with PyTorch according to this schema but can't figure out how to properly update the weights. It is just a toy example with 2 linear layers with 2 nodes in hidden layer and one output.
Learning rate = 0.05;
target output = 1
https://hmkcode.github.io/ai/backpropagation-step-by-step/
Forward
Backward
My code is as following:
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
class MyNet(nn.Module):
def __init__(self):
super(MyNet, self).__init__()
self.linear1 = nn.Linear(2, 2, bias=None)
self.linear1.weight = torch.nn.Parameter(torch.tensor([[0.11, 0.21], [0.12, 0.08]]))
self.linear2 = nn.Linear(2, 1, bias=None)
self.linear2.weight = torch.nn.Parameter(torch.tensor([[0.14, 0.15]]))
def forward(self, inputs):
out = self.linear1(inputs)
out = self.linear2(out)
return out
losses = []
loss_function = nn.L1Loss()
model = MyNet()
optimizer = optim.SGD(model.parameters(), lr=0.05)
input = torch.tensor([2.0,3.0])
print('weights before backpropagation = ', list(model.parameters()))
for epoch in range(1):
result = model(input )
loss = loss_function(result , torch.tensor([1.00],dtype=torch.float))
print('result = ', result)
print("loss = ", loss)
model.zero_grad()
loss.backward()
print('gradients =', [x.grad.data for x in model.parameters()] )
optimizer.step()
print('weights after backpropagation = ', list(model.parameters()))
The result is following :
weights before backpropagation = [Parameter containing:
tensor([[0.1100, 0.2100],
[0.1200, 0.0800]], requires_grad=True), Parameter containing:
tensor([[0.1400, 0.1500]], requires_grad=True)]
result = tensor([0.1910], grad_fn=<SqueezeBackward3>)
loss = tensor(0.8090, grad_fn=<L1LossBackward>)
gradients = [tensor([[-0.2800, -0.4200], [-0.3000, -0.4500]]),
tensor([[-0.8500, -0.4800]])]
weights after backpropagation = [Parameter containing:
tensor([[0.1240, 0.2310],
[0.1350, 0.1025]], requires_grad=True), Parameter containing:
tensor([[0.1825, 0.1740]], requires_grad=True)]
Forward pass values:
2x0.11 + 3*0.21=0.85 ->
2x0.12 + 3*0.08=0.48 -> 0.85x0.14 + 0.48*0.15=0.191 -> loss =0.191-1 = -0.809
Backward pass: let's calculate w5 and w6 (output node weights)
w = w - (prediction-target)x(gradient)x(output of previous node)x(learning rate)
w5= 0.14 -(0.191-1)*1*0.85*0.05= 0.14 + 0.034= 0.174
w6= 0.15 -(0.191-1)*1*0.48*0.05= 0.15 + 0.019= 0.169
In my example Torch doesn't multiply the loss by derivative so we get wrong weights after updating. For the output node we got new weights w5,w6 [0.1825, 0.1740] , when it should be [0.174, 0.169]
Moving backward to update the first weight of the output node (w5) we need to calculate: (prediction-target)x(gradient)x(output of previous node)x(learning rate)=-0.809*1*0.85*0.05=-0.034. Updated weight w5 = 0.14-(-0.034)=0.174. But instead pytorch calculated new weight = 0.1825. It forgot to multiply by (prediction-target)=-0.809. For the output node we got gradients -0.8500 and -0.4800. But we still need to multiply them by loss 0.809 and learning rate 0.05 before we can update the weights.
What is the proper way of doing this?
Should we pass 'loss' as an argument to backward() as following: loss.backward(loss) .
That seems to fix it. But I couldn't find any example on this in documentation.
You should use .zero_grad() with optimizer, so optimizer.zero_grad(), not loss or model as suggested in the comments (though model is fine, but it is not clear or readable IMO).
Except that your parameters are updated fine, so the error is not on PyTorch's side.
Based on gradient values you provided:
gradients = [tensor([[-0.2800, -0.4200], [-0.3000, -0.4500]]),
tensor([[-0.8500, -0.4800]])]
Let's multiply all of them by your learning rate (0.05):
gradients_times_lr = [tensor([[-0.014, -0.021], [-0.015, -0.0225]]),
tensor([[-0.0425, -0.024]])]
Finally, let's apply ordinary SGD (theta -= gradient * lr), to get exactly the same results as in PyTorch:
parameters = [tensor([[0.1240, 0.2310], [0.1350, 0.1025]]),
tensor([[0.1825, 0.1740]])]
What you have done is taken the gradients calculated by PyTorch and multiplied them with the output of previous node and that's not how it works!.
What you've done:
w5= 0.14 -(0.191-1)*1*0.85*0.05= 0.14 + 0.034= 0.174
What should of been done (using PyTorch's results):
w5 = 0.14 - (-0.85*0.05) = 0.1825
No multiplication of previous node, it's done behind the scenes (that's what .backprop() does - calculates correct gradients for all of the nodes), no need to multiply them by previous ones.
If you want to calculate them manually, you have to start at the loss (with delta being one) and backprop all the way down (do not use learning rate here, it's a different story!).
After all of them are calculated, you can multiply each weight by optimizers learning rate (or any other formula for that matter, e.g. Momentum) and after this you have your correct update.
How to calculate backprop
Learning rate is not part of backpropagation, leave it alone until you calculate all of the gradients (it confuses separate algorithms together, optimization procedures and backpropagation).
1. Derivative of total error w.r.t. output
Well, I don't know why you are using Mean Absolute Error (while in the tutorial it is Mean Squared Error), and that's why both those results vary. But let's go with your choice.
Derivative of | y_true - y_pred | w.r.t. to y_pred is 1, so IT IS NOT the same as loss. Change to MSE to get equal results (here, the derivative will be (1/2 * y_pred - y_true), but we usually multiply MSE by two in order to remove the first multiplication).
In MSE case you would multiply by the loss value, but it depends entirely on the loss function (it was a bit unfortunate that the tutorial you were using didn't point this out).
2. Derivative of total error w.r.t. w5
You could probably go from here, but... Derivative of total error w.r.t to w5 is the output of h1 (0.85 in this case). We multiply it by derivative of total error w.r.t. output (it is 1!) and obtain 0.85, as done in PyTorch. Same idea goes for w6.
I seriously advise you not to confuse learning rate with backprop, you are making your life harder (and it's not easy with backprop IMO, quite counterintuitive), and those are two separate things (can't stress that one enough).
This source is nice, more step-by-step, with a little more complicated network idea (activations included), so you can get a better grasp if you go through all of it.
Furthermore, if you are really keen (and you seem to be), to know more ins and outs of this, calculate the weight corrections for other optimizers (say, nesterov), so you know why we should keep those ideas separated.

Tensorflow- How to display accuracy rate for a linear regression model

I have a linear regression model that seems to work. I first load the data into X and the target column into Y, after that I implement the following...
X_train, X_test, Y_train, Y_test = train_test_split(
X_data,
Y_data,
test_size=0.2
)
rng = np.random
n_rows = X_train.shape[0]
X = tf.placeholder("float")
Y = tf.placeholder("float")
W = tf.Variable(rng.randn(), name="weight")
b = tf.Variable(rng.randn(), name="bias")
pred = tf.add(tf.multiply(X, W), b)
cost = tf.reduce_sum(tf.pow(pred-Y, 2)/(2*n_rows))
optimizer = tf.train.GradientDescentOptimizer(FLAGS.learning_rate).minimize(cost)
init = tf.global_variables_initializer()
init_local = tf.local_variables_initializer()
with tf.Session() as sess:
sess.run([init, init_local])
for epoch in range(FLAGS.training_epochs):
avg_cost = 0
for (x, y) in zip(X_train, Y_train):
sess.run(optimizer, feed_dict={X:x, Y:y})
# display logs per epoch step
if (epoch + 1) % FLAGS.display_step == 0:
c = sess.run(
cost,
feed_dict={X:X_train, Y:Y_train}
)
print("Epoch:", '%04d' % (epoch + 1), "cost=", "{:.9f}".format(c))
print("Optimization Finished!")
accuracy, accuracy_op = tf.metrics.accuracy(labels=tf.argmax(Y_test, 0), predictions=tf.argmax(pred, 0))
print(sess.run(accuracy))
I cannot figure out how to print out the model's accuracy. For example, in sklearn, it is simple, if you have a model you just print model.score(X_test, Y_test). But I do not know how to do this in tensorflow or if it is even possible.
I think I'd be able to calculate the Mean Squared Error. Does this help in any way?
EDIT
I tried implementing tf.metrics.accuracy as suggested in the comments but I'm having an issue implementing it. The documentation says it takes 2 arguments, labels and predictions, so I tried the following...
accuracy, accuracy_op = tf.metrics.accuracy(labels=tf.argmax(Y_test, 0), predictions=tf.argmax(pred, 0))
print(sess.run(accuracy))
But this gives me an error...
FailedPreconditionError (see above for traceback): Attempting to use uninitialized value accuracy/count
[[Node: accuracy/count/read = IdentityT=DT_FLOAT, _class=["loc:#accuracy/count"], _device="/job:localhost/replica:0/task:0/device:CPU:0"]]
How exactly does one implement this?
Turns out, since this is a multi-class Linear Regression problem, and not a classification problem, that tf.metrics.accuracy is not the right approach.
Instead of displaying the accuracy of my model in terms of percentage, I instead focused on reducing the Mean Square Error (MSE) instead.
From looking at other examples, tf.metrics.accuracy is never used for Linear Regression, and only classification. Normally tf.metric.mean_squared_error is the right approach.
I implemented two ways of calculating the total MSE of my predictions to my testing data...
pred = tf.add(tf.matmul(X, W), b)
...
...
Y_pred = sess.run(pred, feed_dict={X:X_test})
mse = tf.reduce_mean(tf.square(Y_pred - Y_test))
OR
mse = tf.metrics.mean_squared_error(labels=Y_test, predictions=Y_pred)
They both do the same but obviously the second approach is more concise.
There's a good explanation of how to measure the accuracy of a Linear Regression model here.
I didn't think this was clear at all from the Tensorflow documentation, but you have to declare the accuracy operation, and then initialize all global and local variables, before you run the accuracy calculation:
accuracy, accuracy_op = tf.metrics.accuracy(labels=tf.argmax(Y_test, 0), predictions=tf.argmax(pred, 0))
# ...
init_global = tf.global_variables_initializer
init_local = tf.local_variables_initializer
sess.run([init_global, init_local])
# ...
# run accuracy calculation
I read something on Stack Overflow about the accuracy calculation using local variables, which is why the local variable initializer is necessary.
After reading the complete code you posted, I noticed a couple other things:
In your calculation of pred, you use
pred = tf.add(tf.multiply(X, W), b). tf.multiply performs element-wise multiplication, and will not give you the fully connected layers you need for a neural network (which I am assuming is what you are ultimately working toward, since you're using TensorFlow). To implement fully connected layers, where each layer i (including input and output layers) has ni nodes, you need separate weight and bias matrices for each pair of successive layers. The dimensions of the i-th weight matrix (the weights between the i-th layer and the i+1-th layer) should be (ni, ni + 1), and the i-th bias matrix should have dimensions (ni + 1, 1). Then, going back to the multiplication operation - replace tf.multiply with tf.matmul, and you're good to go. I assume that what you have is probably fine for a single-class linear regression problem, but this is definitely the way you want to go if you plan to solve a multiclass regression problem or implement a deeper network.
Your weight and bias tensors have a shape of (1, 1). You give the variables the initial value of np.random.randn(), which according to the documentation, generates a single floating point number when no arguments are given. The dimensions of your weight and bias tensors need to be supplied as arguments to np.random.randn(). Better yet, you can actually initialize these to random values in Tensorflow: W = tf.Variable(tf.random_normal([dim0, dim1], seed = seed) (I always initialize random variables with a seed value for reproducibility)
Just a note in case you don't know this already, but non-linear activation functions are required for neural networks to be effective. If all your activations are linear, then no matter how many layers you have, it will reduce to a simple linear regression in the end. Many people use relu activation for hidden layers. For the output layer, use softmax activation for multiclass classification problems where the output classes are exclusive (i.e., where only one class can be correct for any given input), and sigmoid activation for multiclass classification problems where the output classes are not exlclusive.

Resources