I am using estimator in Tensorflow(1.8) and python3.6 to build up Neural network for my Reinforcement learning project. And I notice everytime you use estimator.predict(), the tensorflow will load up the checkpoint under the model_dir. But It's extremely inefficient if you have to use this function multiple times for the same checkpoint, e.g. in Reinforcement learning, I may need to predict next action based on current state and next state will be realized only after you choose a specific action. So it's commonplace to call this function thousands of times.
So my question is, how to call this function without loading checkpoint(same checkpoint) everytime.
Thank you.
Well, I think I've just found a good answer to my own question. A good solution to this problem is to construct a tf.dataset by generator. Link is here.
Generator will keep your estimator.predict open and in this way you won't need to keep loading the checkpoint. Only thing you need to do is to change the yielded object in this fastpredict object(self.next_feature in this case) if necessary.
However, I need to mention if your ultimate goal is to make the whole thing a service or something like that. You may need something like tf.serving. So I suggest you go in that way directly. I waste a lot time in the process. So I hope this answer helps you save yours.
Related
I'm training a mask-r-cnn network, which is built on tensorflow and keras. I'm searching for a way to reduce training time, so I thought implementing it with tensorflow-distributed.
I've been working with mask-r-cnn for some time, but it seems what I'm trying to do will require me to modify the source code of mask-r-cnn, which is above my current skills.
So, my question is, has someone ever done it, or something similar? is it possible at all, or am I misunderstand the use of tensorflow-distributed.
Thanks ahead.
Even without distributed training, the tensorpack implementation of Mask R-CNN in tensorflow runs 5x faster (and more accurate as well) than the one you linked to.
It also supports distributed training with MPI.
Several days ago, I used the sklearn multilayer perceptron module for predictions.
Now I try to change the cost function while I using the neural network method which may make the prediction results more accurate. I have added the new cost function to ‘_base.py’ and also I have changed some codes in ‘multilayer_perceptron.py’. However, when I try to call the package and the module, there is a ‘no module named...’ problem. I tried several methods to solve this problem, like check the ‘init.py’ file and check the ‘PYTHONPATH’, but these methods don’t work.
So, could you please give me some guide on how to change the cost function? I would appreciate that and thank you so much.
I have model saved in graph (.pb file). But now the model is inaccurate and I would like to develop it. I have pictures of additional data to learn, but I don't if it's possible or if it's how to do it? The result must be the modified of new data pb graph.
It's a good question. Actually it would be nice, if someone could explain how to do this. But in addition i can say you, that it would come to "catastrophic forgetting", so it wouldn't work out. You had to train all your data again.
But anyway, i also would like to know that espacially for ssd, just for test reasons.
The mozzila/DeepSpeech community has contributed a way to initialize training from a frozen graph(.pb). It does not restores optimizer parameters, so adjusting the learning rate is necessary.
You could find the code at:
https://github.com/mozilla/DeepSpeech/blob/master/DeepSpeech.py#L1562
Hope this helps!
Maybe this is a stupid question, but I switched from basic TensorFlow recently to tflearn and while I knew little of TensorFlow, I know even less of tflearn as I have just begun to experiment with it. I was able to create a network, train it, and generate a model that achieved a satisfactory metric. I did this all without using a TensorFlow session because a) none of the documentation I was looking at necessarily suggested it and b) I didn't even think to use it.
However, I would like to predict a value for a single input (the model performs regression on images, so I'm trying to get a value for a single image) and now I'm getting an error that the convolutional layers need to be initialized (Specifically "FailedPreconditionError: Attempting to use uninitialized value Conv2D/W").
The only thing I've added, though, are two lines:
model = Evaluator(network)
model.predict(feed_dict={input_placeholder: image_data})
I'm asking this as a general question because my actual code is a bit troublesome to just post here because admittedly I've been very sloppy in writing it. I will mention, however, that even if I start a session and initialize all variables before that second line, then run the line in the session, I get the same error.
Succinctly put, does tflearn require a session if I've not used TensorFlow stuff directly anywhere in my code? If so, does the model need to be trained in the session? And if not, what about those two lines would cause such an error?
I'm hoping it isn't necessary for more code to be posted, but if this isn't a general issue and is actually specific to my code then I can try to format it to be understandable here and then edit the post.
I'd like to have 2 separate networks running in Theano at the same time, where the first network trains on the results of the second. I could embed both networks in the same structure but that would be a real mess in the entire forward pass (and probably won't even work because of the shared variables etc.)
The problem is that when I define a theano function I don't specify the model it's applied on, meaning if I'm having a predict and a train function they'll both work on the first model I define.
Is there a way to overcome that issue?
In a rather simplified way I've managed to find a nice solution. The trick was to create one model, define its function and then create the other model and define the second function. Works like a charm