What does inconsistent test results mean? - machine-learning

I'm doing some research on CNN for text classification using tensorflow. When I run my model I get a very high training accuracy (arround 100%). However, on test split I get an inconsistent accuracy results (sometimes 11% and sometimes 90%).
Moreover, I noticed also that the loss in training is decreasing until it reaches small numbers like 0.000499564048368, while in testing it is not and sometimes it gets high values like 70. What does this mean? Any ideas?

If you get very high training accuracy and bad testing accuracy, you are almost definitely overfitting. To get a better picture of what your models real accuracy is, use cross-validation.
Cross validation splits the dataset into a training and validation set, and does this multiple times, slightly changing the training and validation data each time. This is beneficial because it can prevent scenarios where you train your model on one label, and it can't accurately identify another one. For example, picture a training set like this:
Feature1, Feature2, Label
x y 0
a y 0
b c 1
If we train the model only on the first two datapoints, it will not be able to identify the third datapoint because it is not built generally.

Related

Train Accuracy is very high, Validation accuracy is very high but the test set accuracy is very low

I have split the dataset ( around 28K images) into 75% trainset and 25% testset. Then I have taken randomly 15% of trainset and randomly 15% of testset to create a validation set. The goal is to classify the images into two categories. The exact image sample can't be shared. But its similar to the one attached. I'm using this model: VGG19 with imagenet weights with last two layers trainable and 4 dense layers appended. I am also using ImageDataGenerator to Augment the images. I trained the model for 30 epochs and found that the training accuracy is 95% and Validation accuracy is 96% and when trained on test dataset it fell down enormously to 75% only.
I have tried regularization and dropout to tackle the overfitting if it is suffering. I have also done one more thing to see what happens if I use the testset as Validation set and test the model on the same testset. The results were: Trainset Acc = 96% and Validation ACC = 96.3% and the testAcc = 68%. I don't understand what should I Do ?
image
First off, you need to make sure that when you split in data, the relative size of every class in the new datasets is equal. It can be imbalanced if that is the distribution of your initial data, but it must have the same imbalance in all datasets after the split.
Now, regarding the split. If you need a train, validation and test sets, they must all be independent of each-other (no-shared samples). This is important if you don't want to cheat yourself with the results that you are getting.
In general, in machine-learning we start from a training set and a test set. For choosing the best model architecture/hyper-parameters, we further divide the training set to get the validation set (the test set should not be touched).
After determining the best architecture/hyper-parameters for our model, we combine the training and validation set and train the best-case model from scratch with the combined full training set. Only now we get to test the results on the test set.
I had faced a similar issue in one of my practice projects.
My InceptionV3 model gave a high training accuracy (99%), a high validation accuracy (95%+) but a very low testing accuracy (55%).
The dataset was a subset of the popular Dogs vs. Cats dataset (https://www.kaggle.com/c/dogs-vs-cats/data), made by me, having 15k images split into 3 folders: train, valid, and test in the ratio of 60:20:20 (9000, 3000, 3000 each halved for cats folder and dogs folder).
The error in my case was actually in my code. It had nothing to do with the model or the data. The model had been defined inside a function and that was creating an untrained instance during the evaluation. Hence, an untrained model was being tested upon on the test dataset. After correcting the errors in my notebook I got a 96%+ testing accuracy.
Links:
https://colab.research.google.com/drive/1-PO1KJYvXdNC8LbvrdL70oG6QbHg_N-e?usp=sharing&fbclid=IwAR2k9ZCXvX_y_UNWpl4ljs1y0P3budKmlOggVrw6xI7ht0cgm03_VeoKVTI
https://drive.google.com/drive/u/3/folders/1h6jVHasLpbGLtu6Vsnpe1tyGCtR7bw_G?fbclid=IwAR3Xtsbm_EZA3TOebm5EfSvJjUmndHrWXm4Iet2fT3BjE6pPJmnqIwW8KWY
tyuhm
Other probable causes:
One possibility is that the testing set would have a different
distribution than the validation set (This could be excluded by
joining all the data, randomizing, and splitting again to train,
test, valid).
To swap valid and test with each other and see if it has an
effect (Sometimes if one set has relatively harder examples).
If the training somehow overfitted on the validation set (Is it
possible that during training, at one or more steps, the model giving the best score on the validation set is chosen).
Images overlapping, lack of shuffling.
In the deep learning world, if something seems way too odd to be
true, or even way too good to be even true, a good guess is that its
probably a bug unless proven otherwise!

What's a "good" value for the loss function of a DL model like yolo?

I collected ~1500 labelled data and trained with yolo v3, got a training loss of ~10, validation loss ~ 16. Obviously we can use real test data to evaluate the model performance, but I am wondering if there is a way to tell if this training loss = 10 is a "good" one? Or does it indicate I need to use more training data to see if I can push it down to 5 or even less?
Ultimately my question is, for a well-known model with a pre-defined loss function, is there a "good" standard value for the training loss?
thanks.
you need to train your weights until avg loss become 0.0XXXXX. It is minimal requirement to detect object with matching anchor IOU.
Update:28th Nov, 2018
while training object detection model, Loss might be vary sometimes with large data set. but all you need to calculate is Mean Average Precision(MAP) which exactly gave the accuracy criteria of trained model.
./darknet detector map .data .cfg .weights
If your MAP is near to 0.1 i.e. 100%, model performing well.
Follow link to know more about MAP:
https://medium.com/#jonathan_hui/map-mean-average-precision-for-object-detection-45c121a31173
Your validation loss is a good indicator of if the training loss can further alleviate, I mean i don't have any one-shot solutions ,you will have to tweak Hyper-parameters and check on the val test and iterate.You can also get a nice idea by looking at the loss curve, was it decreasing when you stopped training or was it flat, you can get a sense of how the training has progressed and make changes accordingly.GoodLuck

What does this learning curve show ? And how to handle non representativity of a sample?

==> to see learning curves
I am trying a random forest regressor for a machine learning problem (price estimation of spatial points). I have a sample of spatial points in a city. The sample is not randomly drawn since there are very few observations downtown. And I want to estimate prices for all addresses in the city.
I have a good cross validation score (absolute mean squared error) an also a good test score after splitting the training set. But predictions are very bad.
What could explain this results ?
I plotted the learning curve (link above) : cross validation score increases with number of instances (that sounds logical), training score remains high (should it decrease ?) ... What do these learning curves show ? And in general how do we "read" learning curves ?
Moreover, I suppose that the sample is not representative. I tried to make the dataset for which I want predictions spatially similar to the training set by drawing whitout replacement according to proportions of observations in each district for the training set. But this didn't change the result. How can I handle this non representativity ?
Thanks in advance for any help
There are a few common cases that pop up when looking at training and cross-validation scores:
Overfitting: When your model has a very high training score but a poor cross-validation score. Generally this occurs when your model is too complex, allowing it to fit the training data exceedingly well but giving it poor generalization to the validation dataset.
Underfitting: When neither the training nor the cross-validation scores are high. This occurs when your model is not complex enough.
Ideal fit: When both the training and cross-validation scores are fairly high. You model not only learns to represent the training data, but it generalizes well to new data.
Here's a nice graphic from this Quora post showing how model complexity and error relate to the type a fit a model exhibits.
In the plot above, the errors for a given complexity are the errors found at equilibrium. In contrast, learning curves show how the score progresses throughout the entire training process. Generally you never want to see the score decreasing during training, as this usually means your model is diverging. But the difference between the training and validation scores as they move forward in time (towards equilibrium) indicates how well your model is fitting.
Notice that even when you have an ideal fit (middle of complexity axis) it is common to see a training score that's higher than the cross-validation score, since the model's parameters are updated using the training data. But since you're getting poor predictions, and since validation score is ~10% lower than training score (assuming the score is out of 1), I would guess that your model is overfitting and could benefit from less complexity.
To answer your second point, models will generalize better if the training data is a better representation of validation data. So when splitting the data into training and validation sets, I recommend finding a way to randomly segregate the data. For example, you could generate a list of all the points in the city, iterate of the list, and for each point draw from a uniform distribution to decide which dataset that point belongs to.

Why test accuracy remains constant and do not increase in binary classification when test and train dataset are from different source

I have train dataset and test dataset from two different sources. I mean they are from two different experiments but the results of both of them are same biological images. I want to do binary classification using deep CNN and I have following results on test accuracy and train accuracy. The blue line shows train accuracy and the red line shows test accuracy after almost 250 epochs. Why the test accuracy is almost constant and not raising? Is that because Test and Train dataset are come from different distributions?
Edited:
After I have add dropout layer, reguralization terms and mean subtraction I still get following strange results which says the model is overfitting from the beginning!
There could be 2 reasons. First you overfit on the training data. This can be validated by using the validation score as a comparison metric to the test data. If so you can use standard techniques to combat overfitting, like weight decay and dropout.
The second one is that your data is too different to be learned like this. This is harder to solve. You should first look at the value spread of both images. Are they both normalized. Matplotlib normalizes automatically for plotted images. If this still does not work you might want to look into augmentation to make your training data more similar to the test data. Here I can not tell you what to use, without seeing both the trainset and the testset.
Edit:
For normalization the test set and the training set should have a similar value spread. If you do dataset normalization you calculate mean and std on training set. But you also need to use those calculated values on the test set and not calculate the test set values from the test set. This only makes sense if the value spread is similar for both the training and test set. If this is not the case you might want to do per sample normalization first.
Other augmentation that are commonly used for every dataset are oversampling, random channel shifts, random rotations, random translation and random zoom. This makes you invariante to those operations.

Machine Learning Training & Test data split method

I was running a random forest classification model and initially divided the data into train (80%) and test (20%). However, the prediction had too many False Positive which I think was because there was too much noise in training data, so I decided to split the data in a different method and here's how I did it.
Since I thought the high False Positive was due to the noise in the train data, I made the train data to have the equal number of target variables. For example, if I have data of 10,000 rows and the target variable is 8,000 (0) and 2,000 (1), I had the training data to be a total of 4,000 rows including 2,000 (0) and 2,000 (1) so that the training data now have more signals.
When I tried this new splitting method, it predicted way better by increasing the Recall Positive from 14 % to 70%.
I would love to hear your feedback if I am doing anything wrong here. I am concerned if I am making my training data biased.
When you have unequal number of data points in each classes in training set, the baseline (random prediction) changes.
By noisy data, I think you want to mean that number of training points for class 1 is more than other. This is not really called noise. It is actually bias.
For ex: You have 10000 data point in training set, 8000 of class 1 and 2000 of class 0. I can predict class 0 all the time and get 80% accuracy already. This induces a bias and baseline for 0-1 classification will not be 50%.
To remove this bias either you can intentionally balance the training set as you did or you can change the error function by giving weight inversely proportional to number of points in training set.
Actually, what you did is right and this process is something similar to "Stratified sampling".
In your first model,where accuracy was very low the model did not get enough correlations between features and target for positive class(1).Also it model might have somewhat over-fitted for negative class.This is called "High bias -High variance" situation.
"Stratified sampling" is nothing but when you are extracting a sample data from a big population,you make sure that all classes will have some what approximately equal proportion to make the model's training assumptions more accurate and reliable.
In the second case model was able to correlate relationships between features and target and positive and negative class characteristics was well distinguishable.
Eliminating noise is a part of data preparation that should be obviously done before putting data into a model.

Resources