We were given an assignment in which we were supposed to implement our own neural network, and two other already developed Neural Networks. I have done that and however, this isn't the requirement of the assignment but I still would want to know that what are the steps/procedure I can follow to improve the accuracy of my Models?
I am fairly new to Deep Learning and Machine Learning as a whole so do not have much idea.
The given dataset contains a total of 15 classes (airplane, chair etc.) and we are provided with about 15 images of each class in training dataset. The testing dataset has 10 images of each class.
Complete github repository of my code can be found here (Jupyter Notebook file): https://github.com/hassanashas/Deep-Learning-Models
I tried it out with own CNN first (made one using Youtube tutorials).
Code is as follows,
X_train = X_train/255.0
model = Sequential()
model.add(Conv2D(64, (3, 3), input_shape = X_train.shape[1:]))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(128, (3, 3)))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dense(64))
model.add(Dense(16)) # added 16 because it model.fit gave error on 15
model.add(Activation('softmax'))
For the compiling of Model,
from tensorflow.keras.optimizers import SGD
model.compile(loss='sparse_categorical_crossentropy',
optimizer=SGD(learning_rate=0.01),
metrics=['accuracy'])
I used sparse categorical crossentropy because my "y" label was intenger values, ranging from 1 to 15.
I ran this model with following way,
model_fit = model.fit(X_train, y_train, batch_size=32, epochs=30, validation_split=0.1)
It gave me an accuracy of 0.2030 on training dataset and only 0.0733 on the testing dataset (both the datasets are present in the github repository)
Then, I tried out the AlexNet CNN (followed a Youtube tutorial for its code)
I ran the AlexNet on the same dataset for 15 epochs. It improved the accuracy on training dataset to 0.3317, however accuracy on testing dataset was even worse than my own CNN, at only 0.06
Afterwards, I tried out the VGG16 CNN, again following a Youtube Tutorial.
I ran the code on Google Colab for 10 Epochs. It managed to improve to 100% accuracy on training dataset in the 8th epoch. But this model gave the worst accuracy of all three on testing dataset with only 0.0533
I am unable to understand this contrasting behavior of all these models. I have tried out different epoch values, loss functions etc. but the current ones gave the best result relatively. My own CNN was able to get to 100% accuracy when I ran it on 100 epochs (however, it gave very poor results on the testing dataset)
What can I do to improve the performance of these Models? And specifically, what are the few crucial things that one should always try to follow in order to improve efficiency of a Deep Learning Model? I have looked up multiple similar questions on Stackoverflow but almost all of them were working on datasets provided by the tensorflow like mnist dataset and etc. and I didn't find much help from those.
Disclaimer: it's been a few years since I've played with CNNs myself, so I can only pass on some general advice and suggestions.
First of all, I would like to talk about the results you've gotten so far. The first two networks you've trained seem to at least learn something from the training data because they perform better than just randomly guessing.
However: the performance on the test data indicates that the network has not learned anything meaningful because those numbers suggest the network is as good as (or only marginally better than) a random guess.
As for the third network: high accuracy for training data combined with low accuracy for testing data means that your network has overfitted. This means that the network has memorized the training data but has not learned any meaningful patterns.
There's no point in continuing to train a network that has started overfitting. So once the training accuracy increases and testing accuracy decreases for a few epochs consecutively, you can stop training.
Increase the dataset size
Neural networks rely on loads of good training data to learn patterns from. Your dataset contains 15 classes with 15 images each, that is very little training data.
Of course, it would be great if you could get hold of additional high-quality training data to expand your dataset, but that is not always feasible. So a different approach is to artificially expand your dataset. You can easily do this by applying a bunch of transformations to the original training data. Think about: mirroring, rotating, zooming, and cropping.
Remember to not just apply these transformations willy-nilly, they must make sense! For example, if you want a network to recognize a chair, do you also want it to recognize chairs that are upside down? Or for detecting road signs: mirroring them makes no sense because the text, numbers, and graphics will never appear mirrored in real life.
From the brief description of the classes you have (planes and chairs and whatnot...), I think mirroring horizontally could be the best transformation to apply initially. That will already double your training dataset size.
Also, keep in mind that an artificially inflated dataset is never as good as one of the same size that contains all authentic, real images. A mirrored image contains much of the same information as its original, we merely hope it will delay the network from overfitting and hope that it will learn the important patterns instead.
Lower the learning rate
This is a bit of side note, but try lowering the learning rate. Your network seems to overfit in only a few epochs which is very fast. Obviously, lowering the learning rate will not combat overfitting but it will happen more slowly. This means that you can hopefully find an epoch with better overall performance before overfitting takes place.
Note that a lower learning rate will never magically make a bad-performing network good. It's just one way to locate a set of parameters that performs a tad bit better.
Randomize the training data order
During training, the training data is presented in batches to the network. This often happens in a fixed order over all iterations. This may lead to certain biases in the network.
First of all, make sure that the training data is shuffled at least once. You do not want to present the classes one by one, for example first all plane images, then all chairs, etc... This could lead to the network unlearning much of the first class by the end of each epoch.
Also, reshuffle the training data between epochs. This will again avoid potential minor biases because of training data order.
Improve the network design
You've designed a convolutional neural network with only two convolution layers and two fully connected layers. Maybe this model is too shallow to learn to differentiate between the different classes.
Know that the convolution layers tend to first pick up small visual features and then tend to combine these in higher level patterns. So maybe adding a third convolution layer may help the network identify more meaningful patterns.
Obviously, network design is something you'll have to experiment with and making networks overly deep or complex is also a pitfall to watch out for!
Related
I'm training a classification model with custom layers on top of BERT. During this, the training performance of this model is going down with increasing epochs ( after the first epoch ) .. I'm not sure what to fix here - is it the model or the data?
( for the data it's binary labels, and balanced in the number of data points for each label).
Any quick pointers on what the problem could be? Has anyone come across this before?
Edit: Turns out there was a mismatch in the transformers library and tf version I was using. Once I fixed that, the training performance was fine!
Thanks!
Remember that fine-tuning a pre-trained model like Bert usually requires a much smaller number of epochs than models trained from scratch. In fact the authors of Bert recommend between 2 and 4 epochs. Further training often translates to overfitting to your data and forgetting the pre-trained weights (see catastrophic forgetting).
In my experience, this affects small datasets especially as it's easy to overfit on them, even at the 2nd epoch. Besides, you haven't commented on your custom layers on top of Bert, but adding much complexity there might increase overfitting also -- note that the common architecture for text classification only adds a linear transformation.
I'm new in everithing about CNN and tensorflow. Im training a pretrained ssd-mobilenev1-pets.config to detect columns of buildings, about one day but the loss is between 2-1 and doesnt decrease since 10 hours ago.
I realized that my input images are 128x128 and SSD resize de image to 300*300.
Does the size of the input images affect the training?
If that is the case, should I retrain the network with larger input images? or what would be another option to decrease the loss? my train dataset has 660 images and test 166 I dont Know if there are enough images
I really aprecciate your help ....
Loss values of ssd_mobilenet can be different from faster_rcnn. From EdjeElectronics' TensorFlow Object Detection Tutorial:
For my training on the Faster-RCNN-Inception-V2 model, it started at
about 3.0 and quickly dropped below 0.8. I recommend allowing your
model to train until the loss consistently drops below 0.05, which
will take about 40,000 steps, or about 2 hours (depending on how
powerful your CPU and GPU are). Note: The loss numbers will be
different if a different model is used. MobileNet-SSD starts with a
loss of about 20, and should be trained until the loss is consistently
under 2.
For more information: https://github.com/EdjeElectronics/TensorFlow-Object-Detection-API-Tutorial-Train-Multiple-Objects-Windows-10#6-run-the-training
The SSD Mobilnet architecture demands additional training to suffice
the loss accuracy values of the R-CNN model, however, offers
practicality, scalability, and easy accessibility on smaller devices
which reveals the SSD model as a promising candidate for further
assessment (Fleury and Fleury, 2018).
For more information: Fleury, D. & Fleury, A. (2018). Implementation of Regional-CNN and SSD machine learning object detection architectures for the real time analysis of blood borne pathogens in dark field microscopy. MDPI AG.
I would recommend you to take 15%-20% images for testing which cover all the variety present in training data. As you said you have 650+ images for training and 150+ for testing. That is roughly 25% of testing images. It looks like you have enough images to start with. I know the more, the merrier but make sure your model also has sufficient data to learn from!
Resizing the images does not contribute to the loss. It makes sure there is consistency across all images for the model to recognize them without bias. The loss has nothing to do with image resizing as long as every image is resized identically.
You have to make stops and recover checkpoints again and again if you want your model to be perfectly fit. Usually, you can get away with good accuracy by re-training the ssd mobilenet until the loss consistently becomes under 1.Ideally we want the loss to be as lower as possible but we want to make sure the model is not over-fitting. It is all about trial and error. (Loss between 0.5 and 1 seems to be doing the job well but again it all depends on you.)
The reason I think your model is underperforming is due to the fact that you have variety of testing data and not enough training data to suffice.
The model has not been given enough knowledge in training data to make the model learn for new variety of testing data. (For example : Your test data has some images of new angles of buildings which are not sufficiently present in training data). In that case, I recommend you to put variety of all images in training data and then picking images to test making sure you still have sufficient training data of new postures. That's why I recommend you to take 15%-20% test data.
I'm attempting to train a tensorflow model based on the popular slim implementation of mobilenet_v2 and am observing behaviour I cannot explain related (I think) to batch normalization.
Problem Summary
Model performance in inference mode improves initially but starts producing trivial inferences (all near-zeros) after a long period. Good performance continues when run in training mode, even on the evaluation dataset. Evaluation performance is impacted by batch normalization decay/momentum rate... somehow.
More extensive implementation details below, but I'll probably lose most of you with the wall of text, so here are some pictures to get you interested.
The curves below are from a model which I tweaked the bn_decay parameter of while training.
0-370k: bn_decay=0.997 (default)
370k-670k: bn_decay=0.9
670k+: bn_decay=0.5
Loss for (orange) training (in training mode) and (blue) evaluation (in inference mode). Low is good.
Evaluation metric of model on evaluation dataset in inference mode. High is good.
I have attempted to produce a minimal example which demonstrates the issue - classification on MNIST - but have failed (i.e. classification works well and the problem I experience is not exhibited). My apologies for not being able to reduce things further.
Implementation Details
My problem is 2D pose estimation, targeting Gaussians centered at the joint locations. It is essentially the same as semantic segmentation, except rather than using a softmax_cross_entropy_with_logits(labels, logits) I use tf.losses.l2_loss(sigmoid(logits) - gaussian(label_2d_points)) (I use the term "logits" to describe unactivated output of my learned model, though this probably isn't the best term).
Inference Model
After preprocessing my inputs, my logits function is a scoped call to the base mobilenet_v2 followed by a single unactivated convolutional layer to make the number of filters appropriate.
from slim.nets.mobilenet import mobilenet_v2
def get_logtis(image):
with mobilenet_v2.training_scope(
is_training=is_training, bn_decay=bn_decay):
base, _ = mobilenet_v2.mobilenet(image, base_only=True)
logits = tf.layers.conv2d(base, n_joints, 1, 1)
return logits
Training Op
I have experimented with tf.contrib.slim.learning.create_train_op as well as a custom training op:
def get_train_op(optimizer, loss):
global_step = tf.train.get_or_create_global_step()
opt_op = optimizer.minimize(loss, global_step)
update_ops = set(tf.get_collection(tf.GraphKeys.UPDATE_OPS))
update_ops.add(opt_op)
return tf.group(*update_ops)
I'm using tf.train.AdamOptimizer with learning rate=1e-3.
Training Loop
I'm using the tf.estimator.Estimator API for training/evaluation.
Behaviour
Training initially goes well, with an expected sharp increase in performance. This is consistent with my expectations, as the final layer is rapidly trained to interpret the high-level features output by the pretrained base model.
However, after a long period (60k steps with batch_size 8, ~8 hours on a GTX-1070) my model begins to output near-zero values (~1e-11) when run in inference mode, i.e. is_training=False. The exact same model continues to improve when run in *training mode, i.e.is_training=True`, even on the valuation set. I have visually verified this is.
After some experimentation I changed the bn_decay (batch normalization decay/momentum rate) from the default 0.997 to 0.9 at ~370k steps (also tried 0.99, but that didn't make much of a difference) and observed an immdeiate improvement in accuracy. Visual inspection of the inference in inference mode showed clear peaks in the inferred values of order ~1e-1 in the expected places, consistent with the location of peaks from training mode (though values much lower). This is why the accuracy increases significantly, but the loss - while more volative - does not improve much.
These effects dropped off after more training and reverted to all zero inference.
I further dropped the bn_decay to 0.5 at step ~670k. This resulted in improvements to both loss and accuracy. I'll likely have to wait until tomorrow to see the long-term effect.
Loss and an evaluation metric plots given below. Note the evaluation metric is based on the argmax of the logits and high is good. Loss is based on the actual values, and low is good. Orange uses is_training=True on the training set, while blue uses is_training=False on the evaluation set. The loss of around 8 is consistent with all zero outputs.
Other notes
I have also experimented with turning off dropout (i.e. always running the dropout layers with is_training=False), and observed no difference.
I have experimented with all versions of tensorflow from 1.7 to 1.10. No difference.
I have trained models from the pretrained checkpoint using bn_decay=0.99 from the start. Same behaviour as using default bn_decay.
Other experiments with a batch size of 16 result in qualitatively identical behaviour (though I can't evaluate and train simultaneously due to memory constraints, hence quantitatively analysing on batch size of 8).
I have trained different models using the same loss and using tf.layers API and trained from scratch. They have worked fine.
Training from scratch (rather than using pretrained checkpoints) results in similar behaviour, though takes longer.
Summary/my thoughts:
I am confident this is not an overfitting/dataset problem. The model makes sensible inferences on the evaluation set when run with is_training=True, both in terms of location of peaks and magnitude.
I am confident this is not a problem with not running update ops. I haven't used slim before, but apart from the use of arg_scope it doesn't look too much different to the tf.layers API which I've used extensively. I can also inspect the moving average values and observe that they are changing as training progresses.
Chaning bn_decay values significantly effected the results temporarily. I accept that a value of 0.5 is absurdly low, but I'm running out of ideas.
I have tried swapping out slim.layers.conv2d layers for tf.layers.conv2d with momentum=0.997 (i.e. momentum consistent with default decay value) and behaviour was the same.
Minimal example using pretrained weights and Estimator framework worked for classification of MNIST without modification to bn_decay parameter.
I've looked through issues on both the tensorflow and models github repositories but haven't found much apart from this. I'm currently experimenting with a lower learning rate and a simpler optimizer (MomentumOptimizer), but that's more because I'm running out of ideas rather than because I think that's where the problem lies.
Possible Explanations
The best explanation I have is that my model parameters are rapidly cycling in a manner such that the moving statistics are unable to keep up with the batch statistics. I've never heard of such behaviour, and it doesn't explain why the model reverts to poor behaviour after more time, but it's the best explanation I have.
There may be a bug in the moving average code, but it has worked perfectly for me in every other case, including a simple classification task. I don't want to file an issue until I can produce a simpler example.
Anyway, I'm running out of ideas, the debug cycle is long, and I've already spent too much time on this. Happy to provide more details or run experiments on demand. Also happy to post more code, though I'm worried that'll scare more people off.
Thanks in advance.
Both lowering the learning rate to 1e-4 with Adam and using Momentum optimizer (with learning_rate=1e-3 and momentum=0.9) resolved this issue. I also found this post which suggests the problem spans multiple frameworks and is an undocumented pathology of some networks due to the interaction between optimizer and batch-normalization. I do not believe it is a simple case of the optimizer failing to find a suitable minimum due to the learning rate being too high (otherwise performance in training mode would be poor).
I hope that helps others experiencing the same issue, but I'm a long way from satisfied. I'm definitely happy to hear other explanations.
To elaborate : Under what circumstances would fine tuning all layers of a small network (say SqueezeNet) perform better than feature extracting or fine tuning only last 1 or 2 Convolution layer of a big network (e.g inceptionV4)?
My understanding is computing resource required for both is somewhat comparable. And I remember reading in a paper that extreme options i.e fine tuning 90% or 10% of network is far better compared to more moderate like 50%. So, what should be the default choice when experimenting extensively is not an option?
Any past experiments and intuitive description of their result, research paper or blog would be specially helpful. Thanks.
I don't have much experience in training models like SqueezeNet, but I think it is much easier to finetune only the last 1 or 2 layers of a big network: you don't have to extensively search for many optimal hyperparameters. Transfer learning works amazingly well out of the box with the LR finder and the cyclical learning rate from fast.ai.
If you want fast inference after the training, then it is preferable to train SqueezeNet. It might also be the case if the new task is very different from ImageNet.
Some intuition from http://cs231n.github.io/transfer-learning/
New dataset is small and similar to original dataset. Since the data is small, it is not a good idea to fine-tune the ConvNet due to overfitting concerns. Since the data is similar to the original data, we expect higher-level features in the ConvNet to be relevant to this dataset as well. Hence, the best idea might be to train a linear classifier on the CNN codes.
New dataset is large and similar to the original dataset. Since we have more data, we can have more confidence that we won’t overfit if we were to try to fine-tune through the full network.
New dataset is small but very different from the original dataset. Since the data is small, it is likely best to only train a linear classifier. Since the dataset is very different, it might not be best to train the classifier form the top of the network, which contains more dataset-specific features. Instead, it might work better to train the SVM classifier from activations somewhere earlier in the network.
New dataset is large and very different from the original dataset. Since the dataset is very large, we may expect that we can afford to train a ConvNet from scratch. However, in practice it is very often still beneficial to initialize with weights from a pretrained model. In this case, we would have enough data and confidence to fine-tune through the entire network.
I'm using TensorFlow to train a Convolutional Neural Network (CNN) for a sign language application. The CNN has to classify 27 different labels, so unsurprisingly, a major problem has been addressing overfitting. I've taken several steps to accomplish this:
I've collected a large amount of high-quality training data (over 5000 samples per label).
I've built a reasonably sophisticated pre-processing stage to help maximize invariance to things like lighting conditions.
I'm using dropout on the fully-connected layers.
I'm applying L2 regularization to the fully-connected parameters.
I've done extensive hyper-parameter optimization (to the extent possible given HW and time limitations) to identify the simplest model that can achieve close to 0% loss on training data.
Unfortunately, even after all these steps, I'm finding that I can't achieve much better that about 3% test error. (It's not terrible, but for the application to be viable, I'll need to improve that substantially.)
I suspect that the source of the overfitting lies in the convolutional layers since I'm not taking any explicit steps there to regularize (besides keeping the layers as small as possible). But based on examples provided with TensorFlow, it doesn't appear that regularization or dropout is typically applied to convolutional layers.
The only approach I've found online that explicitly deals with prevention of overfitting in convolutional layers is a fairly new approach called Stochastic Pooling. Unfortunately, it appears that there is no implementation for this in TensorFlow, at least not yet.
So in short, is there a recommended approach to prevent overfitting in convolutional layers that can be achieved in TensorFlow? Or will it be necessary to create a custom pooling operator to support the Stochastic Pooling approach?
Thanks for any guidance!
How can I fight overfitting?
Get more data (or data augmentation)
Dropout (see paper, explanation, dropout for cnns)
DropConnect
Regularization (see my masters thesis, page 85 for examples)
Feature scale clipping
Global average pooling
Make network smaller
Early stopping
How can I improve my CNN?
Thoma, Martin. "Analysis and Optimization of Convolutional Neural Network Architectures." arXiv preprint arXiv:1707.09725 (2017).
See chapter 2.5 for analysis techniques. As written in the beginning of that chapter, you can usually do the following:
(I1) Change the problem definition (e.g., the classes which are to be distinguished)
(I2) Get more training data
(I3) Clean the training data
(I4) Change the preprocessing (see Appendix B.1)
(I5) Augment the training data set (see Appendix B.2)
(I6) Change the training setup (see Appendices B.3 to B.5)
(I7) Change the model (see Appendices B.6 and B.7)
Misc
The CNN has to classify 27 different labels, so unsurprisingly, a major problem has been addressing overfitting.
I don't understand how this is connected. You can have hundreds of labels without a problem of overfitting.