I need assistance as a ML beginner. I have retrained TF2's model using make_image_classifier through tensorflow hub (command line approach).
My very first training:
I have immediately retrained the model once again as this one's predictions did not satisfy me -> using 4 classes instead and achieved 85% val.accuracy in Epoch 4/5 and 81% val.accuracy in Epoch 5/5. Because the problem is complex, I have decided to expand upon my db, increasing the number of data fed to the model.
I have expanded the database more than twofold! The results are shocking and I have no clue what to do. I can't believe this.
If anything - the added data is of much better quality, relevance and diversity, and it is actually obtained by myself - a human "discriminator" to the 2nd model's preformance (the 81% acc. one). If I knew how to do it, which is a second question - I would simply add this "feedback data" to my already working model. But I have no idea how to make it happen - ideally, I'd want to give feedback to the bot, allowing it to train once again on the additional data, it understanding that this is a feedback rather than a random additional set of data. However training a new model with updated data from scratch I'd like to try out anyways.
How do I interpret those numbers? What can be wrong with the data? Is it the data what's wrong at all, which is my assumption? Why is the training accuracy below 0.5?? What could have caused the drop between Epoch 2/5' and Epoch 3/5 starting the downfall? What are common issues and similar situations when something like this happens?
I'd love to understand what happened behind the curtains on a more lower level here but I have difficulty, I need guidance. This is as discouraging as the first & second training were encouraging.
Possible problems with the data I can see - but I can't believe they could cause such a drop especially because they were there during 1st&2nd training that went okay:
1-5% images can be in multiple classes (2 classes or maximum 3) - in my opinion this is even encouraged for the problem, as the model imitates me, and I want it to struggle with finding out what it is about certain elements that caused me to classify them as two+ simultaneous classes. Finding out features in them that made me think they'd satisfy both.
1-3% can be duplicates.
There's white border around the images (since it didn't seem to affect the first trainings I didn't bother to remove it)
The drop is 40%.
I will now try to use Tensorflow's tutorial to do it through a script rather than using command line, but seeing such a huge drop is very discouraging, I was hoping to see an increase, after all that's a common suggestion to feed a model more data and I have made sure to feed it quality additional data.. I appreciate every single suggestion for fine-tuning the model, as I am a beginner and I have no experience with what might work and what most probably won't. Thank you.
EDIT: I have cleaned my data by:
removing borders(now they are tiny, white, sometimes not present at all)
removing dust
removing artifacts which there were quite many in the added data!
Convinced the last point was the issue, I retrained. Results are even worse!
2 classes binary
4 classes
Related
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about programming within the scope defined in the help center.
Closed 2 years ago.
Improve this question
I am trying to tune a basic neural network as practice. (Based on an example from a coursera course: Neural Networks and Deep Learning - DeepLearning.AI)
I face the issue of the random weight initialization. Lets say I try to tune the number of layers in the network.
I have two options:
1.: set the random seed to a fixed value
2.: run my experiments more times without setting the seed
Both version has pros and cons.
My biggest concern is that if I use a random seed (e.g.: tf.random.set_seed(1)) then the determined values can be "over-fitted" to the seed and may not work well without the seed or if the value is changed (e.g.: tf.random.set_seed(1) -> tf.random.set_seed(2). On the other hand, if I run my experiments more times without random seed then I can inspect less option (due to limited computing capacity) and still only inspect a subset of possible random weight initialization.
In both cases I feel that luck is a strong factor in the process.
Is there a best practice how to handle this topic?
Has TensorFlow built in tools for this purpose? I appreciate any source of descriptions or tutorials. Thanks in advance!
Tuning hyperparameters in deep learning (generally in machine learning) is a common issue. Setting the random seed to a fixed number ensures reproducibility and fair comparison. Repeating the same experiment will lead to the same outcomes. As you probably know, best practice to avoid over-fitting is to do a train-test split of your data and then use k-fold cross-validation to select optimal hyperparameters. If you test multiple values for a hyperparameter, you want to make sure other circumstances that might influence the performance of your model (e.g. train-test-split or weight initialization) are the same for each hyperparameter in order to have a fair comparison of the performance. Therefore I would always recommend to fix the seed.
Now, the problem with this is, as you already pointed out, the performance for each model will still depend on the random seed, like the particular data split or weight initialization in your case. To avoid this, one can do repeated k-fold-cross validation. That means you repeat the k-fold cross-validation multiple times, each time with a different seed, select best parameters of that run, test on test data and average the final results to get a good estimate of performance + variance and therefore eliminate the influence the seed has in the validation process.
Alternatively you can perform k-fold cross validation a single time and train each split n-times with a different random seed (eliminating the effect of weight initialization, but still having the effect of the train-test-split).
Finally TensorFlow has no build-in tool for this purpose. You as practitioner have to take care of this.
There is no an absolute right or wrong answer to your question. You are almost answered your own question already. In what follows, however, I will try to expand more, via the following points:
The purpose of random initialization is to break the symmetry that makes neural networks fail to learn:
... the only property known with complete certainty is that the
initial parameters need to “break symmetry” between different units.
If two hidden units with the same activation function are connected to
the same inputs, then these units must have different initial
parameters. If they have the same initial parameters, then a
deterministic learning algorithm applied to a deterministic cost and
model will constantly update both of these units in the same way...
Deep Learning (Adaptive Computation and Machine Learning series)
Hence, we need the neural network components (especially weights) to be initialized by different values. There are some rules of thumb of how to choose those values, such as the Xavier initialization, which samples from normal distribution with mean of 0 and special variance based on the number of the network layer. This is a very interesting article to read.
Having said so, the initial values are important but not extremely critical "if" proper rules are followed, as per mentioned in point 2. They are important because large or improper ones may lead to vanishing or exploding gradient problems. On the other hand, different "proper" weights shall not hugely change the final results, unless they are making the aforementioned problems, or getting the neural network stuck at some local maxima. Please note, however, the the latter depends also on many other aspects, such as the learning rate, the activation functions used (some explode/vanish more than others: this is a great comparison), the architecture of the neural network (e.g. fully connected, convolutional ..etc: this is a cool paper) and the optimizer.
In addition to point 2, bringing a good learning optimizer into the bargain, other than the standard stochastic one, shall in theory not let a huge influence of the initial values to affect the final results quality, noticeably. A good example is Adam, which provides a very adaptive learning technique.
If you still get a noticeably-different results, with different "proper" initialized weights, there are some ways that "might help" to make neural network more stable, for example: use a Train-Test split, use a GridSearchCV for best parameters, and use k-fold cross validation...etc.
At the end, obviously the best scenario is to train the same network with different random initial weights many times then get the average results and variance, for more specific judgement on the overall performance. How many times? Well, if can do it hundreds of times, it will be better, yet that clearly is almost impractical (unless you have some Googlish hardware capability and capacity). As a result, we come to the same conclusion that you had in your question: There should be a tradeoff between time & space complexity and reliability on using a seed, taking into considerations some of the rules of thumb mentioned in previous points. Personally, I am okay to use the seed because I believe that, "It’s not who has the best algorithm that wins. It’s who has the most data". (Banko and Brill, 2001). Hence, using a seed with enough (define enough: it is subjective, but the more the better) data samples, shall not cause any concerns.
I'm kinda new to Neural Networks and just started to learn coding them by trying some examples.
Two weeks ago I was searching for an interesting challenge and I found one. But I'm about to give up because it seems to be too hard for me... But I was curious to know if anyone of you is able to solve this?
The Problem: Assume there are ".htm"-files that contain tables about the same topic. But the table structure isn't the same for every file. For example: We have a lot ".htm"-files containing information about teachers substitutions per day per school. Because the structure of those ".htm"-files isn't the same for every file it would be hard to program a parser that could extract the data from those tables. So my thought was that this is a task for a Neural Network.
First Question: Is it a task a Neural Network can/should handle or am I mistaken by that?
Because for me a Neural Network seemed to fit for this kind of a challenge I tried to thing of an Input. I came up with two options:
First Input Option: Take the HTML Code (only from the body-tag) as string and convert it as Tensor
Second Input Option: Convert the HTML Tables into Images (via Canvas maybe) and feed this input to the DNN through Conv2D-Layers.
Second Question: Are those Options any good? Do you have any better solution to this?
After that I wanted to figure out how I would make a DNN output this heavily dynamic data for me? My thought was to convert my desired JSON-Output into Tensors and feed them to the DNN while training and for every prediction i would expect the DNN to return a Tensor that is convertible into a JSON-Output...
Third Question: Is it even possible to get such a detailed Output from a DNN? And if Yes: Do you think the Output would be suitable for this task?
Last Question: Assuming all my assumptions are correct - Wouldn't training this DNN take for ever? Let's say you have a RTX 2080 ti for it. What would you guess?
I guess that's it. I hope i can learn a lot from you guys!
(I'm sorry about my bad English - it's not my native language)
Addition:
Here is a more in-depth Example. Lets say we have a ".htm"-file that looks like this:
The task would be to get all the relevant informations from this table. For example:
All Students from Class "9c" don't have lessons in their 6th hour due to cancellation.
1) This is not particularly suitable problem for a Neural Network, as you domain is a structured data with clear dependcies inside. Tree based ML algorithms tend to show much better results on such problems.
2) Both you choices of input are very unstructured. To learn from such data would be nearly impossible. The are clear ways to give more knowledge to the model. For example, you have the same data in different format, the difference is only the structure. It means that a model needs to learn a mapping from one structure to another, it doesn't need to know any data. Hence, words can be Tokenized with unique identifiers to remove unnecessary information. Htm data can be parsed to a tree, as well as json. Then, there are different ways to represent graph structures, which can be used in a ML model.
3) It seems that the only adequate option for output is a sequence of identifiers pointing to unique entities from text. The whole problem then is similar to Seq2Seq best solved by RNNs with an decoder-encoder architecture.
I believe that, if there is enough data and htm files don't have huge amount of noise, the task can be completed. Training time hugely depends on selected model and its complexity, as well as diversity of initial data.
I'm using Tensorflow-GPU 1.8 API on Windows 10. For many projects I use the tf.Estimator's, which really work great. It takes care of a bunch of steps including writting summaries for Tensorboard. But right now the 'events.out.tfevents' file getting way to big and I am running into "out of space" errors. For that reason I want to disable the summary writting or at least reduce the amount of summaries written.
Going along with that mission I found out about the RunConfig you can pass over at construction of tf.Estimator. Apparently the parameter 'save_summary_steps' (which by default is 200) controls the way summaries are wrtitten out. Unfortunately changing this parameter seems to have no effect at all. It won't disable (using None value) the summary or reducing (choosing higher values, e.g. 3000) the file size of 'events.out.tfevents'.
I hope you guys can help me out here. Any help is appreciated.
Cheers,
Tobs.
I've observed the following behavior. It doesn't make sense to me so I hope we get a better answer:
When the input_fn gets data from tf.data.TFRecordDataset then the number of steps between saving events is the minimum of save_summary_steps and (number of training examples divided by batch size). That means it does it a minimum of once per epoch.
When the input_fn gets data from tf.TextLineReader, it follows save_summary_steps as you'd expect and I can give it a large value for infrequent updates.
I implemented the A3C network in https://arxiv.org/abs/1602.01783 in TensorFlow.
At this point I'm 90% sure the algorithm is implemented correctly. However, the network diverges after convergence. See the attached image that I got from a toy example where the maximum episode reward is 7.
When it diverges, policy network starts giving a single action very high probability (>0.9) for most states.
What should I check for this kind of problem? Is there any reference for it?
Note that in Figure 1 of the original paper the authors say:
For asynchronous methods we average over the best 5
models from 50 experiments.
That can mean that in lot of cases the algorithm does not work that well. From my experience, A3C often diverges, even after convergence. Carefull learning-rate scheduling can help. Or do what the authors did - learn several agents with different seed and pick the one performing the best on your validation data. You could also employ early stopping when validation error becomes to increase.
having read this article about a guy who uses tensorflow to sort cucumber into nine different classes I was wondering if this type of process could be applied to a large number of classes. My idea would be to use it to identify Lego parts.
At the moment, a site like Bricklink describes more than 40,000 different parts so it would be a bit different than the cucumber example but I am wondering if it sounds suitable. There is no easy way to get hundreds of pictures for each part but does the following process sound feasible :
take pictures of a part ;
try to identify the part using tensorflow ;
if it does not identify the correct part, take more pictures and feed the neural network with them ;
go on with the next part.
That way, each time we encounter a new piece we "teach" the network its reference so that it can better be recognized the next time. Like that and after hundreds of iterations monitored by a human, could we imagine tensorflow to be able to recognize the parts? At least the most common ones?
My question might sound stupid but I am not into neural networks so any advice is welcome. At the moment I have not found any way to identify a lego part based on pictures and this "cucumber example" sounds promising so I am looking for some feedback.
Thanks.
You can read about the work of Jacques Mattheij, he actually uses a customized version of Xception1 running on https://keras.io/.
The introduction is Sorting 2 Metric Tons of Lego.
In Sorting 2 Tons of Lego, The software Side you can read:
The hard challenge to deal with next was to get a training set large
enough to make working with 1000+ classes possible. At first this
seemed like an insurmountable problem. I could not figure out how to
make enough images and to label them by hand in acceptable time, even
the most optimistic calculations had me working for 6 months or longer
full-time in order to make a data set that would allow the machine to
work with many classes of parts rather than just a couple.
In the end the solution was staring me in the face for at least a week
before I finally clued in: it doesn’t matter. All that matters is that
the machine labels its own images most of the time and then all I need
to do is correct its mistakes. As it gets better there will be fewer
mistakes. This very rapidly expanded the number of training images.
The first day I managed to hand-label about 500 parts. The next day
the machine added 2000 more, with about half of those labeled wrong.
The resulting 2500 parts where the basis for the next round of
training 3 days later, which resulted in 4000 more parts, 90% of which
were labeled right! So I only had to correct some 400 parts, rinse,
repeat… So, by the end of two weeks there was a dataset of 20K images,
all labeled correctly.
This is far from enough, some classes are severely under-represented
so I need to increase the number of images for those, perhaps I’ll
just run a single batch consisting of nothing but those parts through
the machine. No need for corrections, they’ll all be labeled
identically.
A recent update is Sorting 2 Tons of Lego, Many Questions, Results.
1CHOLLET, François. Xception: Deep Learning with Depthwise Separable Convolutions. arXiv preprint arXiv:1610.02357, 2016.
I have started this using IBM Watson's Visual Recognition.
I had six different bricks to be recognized on the transport belt background.
I am actually thinking about tensorflow, since I can have it running locally.
The codelab : TensorFlow for Poets, describes almost exactly what you want to achieve,
For a demo of the Watson version:
https://www.ibm.com/developerworks/community/blogs/ibmandgoogle/entry/Lego_bricks_recognition_with_Watosn_lego_and_raspberry_pi?lang=en