Basically as this thread discusses here, you cannot use python list to wrap your sub-modules (for example your layers); otherwise, Pytorch is not going to update the parameters of the sub-modules inside the list. Instead you should use nn.ModuleList to wrap your sub-modules to make sure their parameters are going to be updated. Now I have also seen codes like following where the author uses python list to calculate the loss and then do loss.backward() to do the update (in reinforce algorithm of RL). Here is the code:
policy_loss = []
for log_prob in self.controller.log_probability_slected_action_list:
policy_loss.append(- log_prob * (average_reward - b))
self.optimizer.zero_grad()
final_policy_loss = (torch.cat(policy_loss).sum()) * gamma
final_policy_loss.backward()
self.optimizer.step()
Why using the list in this format works for updating the parameters of modules but the first case does not work? I am very confused now. If I change in the previous code policy_loss = nn.ModuleList([]), it throws an exception saying that tensor float is not sub-module.
You are misunderstanding what Modules are. A Module stores parameters and defines an implementation of the forward pass.
You're allowed to perform arbitrary computation with tensors and parameters resulting in other new tensors. Modules need not be aware of those tensors. You're also allowed to store lists of tensors in Python lists. When calling backward it needs to be on a scalar tensor thus the sum of the concatenation. These tensors are losses and not parameters so they should not be attributes of a Module nor wrapped in a ModuleList.
Related
I have a regression that I can run for example as
reghdfe y, a(x1_est=x1 x2_est=x2)
which will store the estimated coefficients in x1_est and x2_est. Now, the issue is that using absorb() does not allow me to get the standard errors for these coefficients. If I understand it correctly, no postestimation method of reghdfe allows me to retrieve those.
Luckily, I only care about the standard errors of x1. So, I could instead run
reg y i.x1, a(x2)
and inspect _se[x1]. Unfortunately, x1 has so many different levels that it is not possible to store it as integer, it has to be double. The previous regression hence will fail with x1: factor variables may not contain noninteger values.
What could be another approach to get standard errors for x1?
With large number of fixed effects, STATA's default approaches won't work. One angle is to bootstrap fixed effects and generate standard errors. Again, the issue is that there are so many FE, such that standard bootstrapping methods won't work (cannot return such a large matrix in each bootstrap).
Essentially, to bootstrap the FE, one would (for a large number of iterations)
preserve
bsample
run the regression, reghdfe y, a(x1_est=x1 x2_est-x2)
Store x1_est in a .dta file
restore
After the loop is done, iteratively append all the .dta files, and compute standard errors.
I have a function defined and I just want to know if it is possible to perform it batchwise. For instance,
def function():
Some processes here
return x
def forward():
encode = self._encoding(embedded_premises,premises_lengths)
Now since, the encode will be 3D tensor, which will be batch size, seq_length, hidden size I want to perform function() batchwise and return x also batchwise.
Is there any other way than looping over all batches?
If you're working with pytorch functions inside your function, which you most likely are if you want your method to work with autograd, it can work batchwise. That means most pytorch operations respect or can be made to respect the first dimension as batch dimension (for instance convolutions, linear layers, etc). Sometimes it's more complex to express your operation such that it is both correct and fast, but in general pytorch is built with the assumption that operations will be used on batched data and it is made as simple as reasonably possible. If you have a more specific example of your function, please post it.
I have a huge lib of math functions, like pdf or cdf of statistical distributions. But often e.g. the inverse cdf can be only calculated numerically, e.g. using Newton-Raphson or bisection, in the latter we would need to check if cdf(x) is > or < then the target y0.
However, many functions have further parameters like a Gaussian distribution having certain mean and sigma, so cdf is cdf(x,mean,sigma). Whereas other functions, such as standard normal cdf, have no further parameters, or some have even 3 or 4 further parameters.
A similar problem would happen if you want to apply bisection for either linear functions (2 parameters) or parabolas (3 parameters). Or if you want not the inverse function, but e.g. the integral of f.
The easiest implementation would be to define cdf as global function f(x); and to check for >y0 or global variables.
However, this is a very old-fashioned way, and Freepascal also supports procedural parameters, for calls like x=icdf(0.9987,#cdfStdNorm)
Even overloading is supported to allow calls like x2=icdf(0.9987,0,2,#cdfNorm) to pass also mean and sigma.
But this ends up still in two separate code blocks (even whole functions), because in one case we need to call cdf only with x, and in 2nd example also with mean and sigma.
Is there an elegant solution for this problem in Freepascal? Maybe using variant records? Or an object-oriented approach? I have no glue about OO, but I know the variant object style would require to change at least the headers of many functions because I want to apply the technique not only for inverse cdf calculation, but also to numerical integration, root finding, optimization, etc.
Or is it "best" just to define a real function type with e.g. x + 5 parameters (maybe as array), and to ignore the unused parameters? But for me it looks that then I would need many "wrapper" functions or to re-code all the existing functions (to use the arrays, even if they are sometimes not needed!).
Maybe macros can help as well? Any Freepascal hints are very welcome!
If you make it a (function .. of object), mean and sigma could be part of the class, and the function could internally just access it. Only the really changing parameters during the iteration would be parameters. (read: x)
Anonymous methods as talked about by David and Rudy is a further step to avoid having to declare a class for each such invocation, but that is convenience thing and IMHO not the core of the question. At the expense of declaring the class, your core code is free of global variable use and anonymous methods might also come with a performance cost, depending on usage.
Free Pascal also supports nested functions (function... is nested), which is the original Pascal closure-like way which was never adopted by Pascal compilers from Borland. A nested procedure passed as callback can access local variables in the procedure where it was declared. The Free Pascal numlib numeric math package uses this in some cases for similar cases like yours. For math it is even more natural.
Delphi never implements old constructs because borrowing syntax from other languages looks better on bulletlists and keeps the subscriptions flowing.
I am new to keras and despite reading the documentation and the examples folder in keras, I'm still struggling with how to fit everything together.
In particular, I want to start with a simple task: I have a sequence of tokens, where each token has exactly one label. I have a lot training data like this - practically infinite, as I can generate more (token, label) training pairs as needed.
I want to build a network to predict labels given tokens. The number of tokens must always be the same as the number of labels (one token = one label).
And I want this to be based on all surrounding tokens, say within the same line or sentence or window -- not just on the preceding tokens.
How far I got on my own:
created the training numpy vectors, where I converted each sentence into a token-vector and label-vector (of same length), using a token-to-int and label-to-int mappings
wrote a model using categorical_crossentropy and one LSTM layer, based on https://github.com/fchollet/keras/blob/master/examples/lstm_text_generation.py.
Now struggling with:
All the input_dim and input_shape parameters... since each sentence has a different length (different number of tokens and labels in it), what should I put as input_dim for the input layer?
How to tell the network to use the entire token sentence for prediction, not just one token? How to predict a whole sequence of labels given a sequence of tokens, rather than just label based on previous tokens?
Does splitting the text into sentences or windows make any sense? Or can I just pass a vector for the entire text as a single sequence? What is a "sequence"?
What are "time slices" and "time steps"? The documentation keeps mentioning that and I have no idea how that relates to my problem. What is "time" in keras?
Basically I have trouble connecting the concepts from the documentation like "time" or "sequence" to my problem. Issues like Keras#40 didn't make me any wiser.
Pointing to relevant examples on the web or code samples would be much appreciated. Not looking for academic articles.
Thanks!
If you have sequences of different length you can either pad them or use a stateful RNN implementation in which the activations are saved between batches. The former is the easiest and most used.
If you want to use future information when using RNNs you want to use a bidirectional model where you concatenate two RNN's moving in opposite directions. RNN will use a representation of all previous information when e.g. predicting.
If you have very long sentences it might be useful to sample a random sub-sequence and train on that. Fx 100 characters. This also helps with overfitting.
Time steps are your tokens. A sentence is a sequence of characters/tokens.
I've written an example of how I understand your problem but it's not tested so it might not run. Instead of using integers to represent your data I suggest one-hot encoding if it is possible and then use binary_crossentropy instead of mse.
from keras.models import Model
from keras.layers import Input, LSTM, TimeDistributed
from keras.preprocessing import sequence
# Make sure all sequences are of same length
X_train = sequence.pad_sequences(X_train, maxlen=maxlen)
# The input shape is your sequence length and your token embedding size (which is 1)
inputs = Input(shape=(maxlen, 1))
# Build a bidirectional RNN
lstm_forward = LSTM(128)(inputs)
lstm_backward = LSTM(128, go_backwards=True)(inputs)
bidirectional_lstm = merge([lstm_forward, lstm_backward], mode='concat', concat_axis=2)
# Output each timestep into a fully connected layer with linear
# output to map to an integer
sequence_output = TimeDistributed(Dense(1, activation='linear'))(bidirectional_lstm)
# Dense(n_classes, activation='sigmoid') if you want to classify
model = Model(inputs, sequence_output)
model.compile('adam', 'mse')
model.fit(X_train, y_train)
I often cross this kind of code transformation (or even mathematical transformation). (Python example, but applies to any language.)
I've go a function
def f(x):
return x
I use it into another one.
def g(x):
return f(x)*f(x)
print g(2)
leads to 4
But I want to remove the functional dependency, and I change the function g into
def g(f):
return f*f
print g( f(2) )
leads to 4 too
How do you call this kind of transformation, locally turning a function into a scalar ?
I'm not sure there is a specific term for it.
In general terms for functional programming there usually isn't a distinction made between passing scalar arguments and passing functions as arguments.
In the first example I could still call g(f(2)) and it should calculate f(f(2))*f(f(2)), which (since f(x) is the identity transformation) will also result in 4 as the answer.