I have used to below hyper parameters to train the model.
rcf.set_hyperparameters(
num_samples_per_tree=200,
num_trees=250,
feature_dim=1,
eval_metrics =["accuracy", "precision_recall_fscore"])
is there any best way to choose the num_samples_per_tree and num_trees parameters.
what are the best numbers for both num_samples_per_tree and num_trees.
There are natural interpretations for these two hyper-parameters that can help you determine good starting approximations for HPO:
num_samples_per_tree -- the reciprocal of this value approximates the density of anomalies in your data set/stream. For example, if you set this to 200 then the assumption is that approximately 0.5% of the data is anomalous. Try exploring your dataset to make an educated estimate.
num_trees -- the more trees in your RCF model the less noise in scores. That is, if more trees are reporting that the input inference point is an anomaly then the point is much more likely to be an anomaly than if few trees suggest so.
The total number of points sampled from the input dataset is equal to num_samples_per_tree * num_trees. You should make sure that the input training set is at least this size.
(Disclosure - I helped create SageMaker Random Cut Forest)
Related
I am comparing models for the detection of objects for maritime Search and Rescue (SAR) purposes. From the models that I used, I got the best results for the improved version of YOLOv3 for small object detection and for FASTER RCNN.
For YOLOv3 I got the best mAP#50, but for FASTER RCNN I got better all other metrics (precision, recall, F1 score). Now I am wondering how to read it and which model is really better in this case?
I would like to add that there are only two classes in the dataset: small and large objects. We chose this solution because the objects' distinction between classes is not as important to us as the detection of any human origin object.
However, small objects don't mean small GT bounding boxes. These are objects that actually have a small area - less than 2 square meters (e.g. people, buoys). Large objects are objects with a larger area (boats, ships, canoes, etc.).
Here are the results per category:
And two sample images from the dataset (with YOLOv3 detections):
The mAP for object detection is the average of the AP calculated for all the classes. mAP#0.5 means that it is the mAP calculated at IOU threshold 0.5.
The general definition for the Average Precision(AP) is finding the area under the precision-recall curve.
The process of plotting the model's precision and recall as a function of the model’s confidence threshold is the precision recall curve.
Precision measures how accurate is your predictions. i.e. the percentage of your predictions that are correct. Recall measures how good you find all the positives. F1 score is HM (Harmonic Mean) of precision and recall.
To answer your questions now.
How to read it and which model is really better in this case?
The mAP is a good measure of the sensitivity of the neural network. So good mAP indicates a model that's stable and consistent across difference confidence thresholds. In your case faster rcnn results indicate that precision-recall curve metric is bad compared to that of Yolov3, which means that either faster rcnn has very bad recall at higher confidence thresholds or very bad precision at lower confidence threshold compared to that of Yolov3 (especially for small objects).
Precision, Recall and F1 score are computed for given confidence threshold. I'm assuming you're running the model with default confidence threshold (could be 0.25). So higher Precision, Recall and F1 score of faster rcnn indicate that at that confidence threshold it's better in terms of all the 3 metric compared to that of Yolov3.
What metric should be more important?
In general to analyse better performing model, I would suggest you to use validation set (data set that is used to tune hyper-parameters) and test set (data set that is used to assess the performance of a fully-trained model).
Note: FP - False Positive FN - False Negative
On validation set:
Use mAP to select best performing model (model that is more stable and consistent) out of all the trained weights across iterations/epochs. Use mAP to understand whether model should be trained/tuned further or not.
Check class level AP values to ensure model is stable and good across the classes.
As per use-case/application, if you're completely tolerant to FNs and highly intolerant to FPs then to train/tune the model accordingly use Precision.
As per use-case/application, if you're completely tolerant to FPs and highly intolerant to FNs then to train/tune the model accordingly use Recall.
On test set:
If you're neutral towards FPs and FNs, then use F1 score to evaluate best performing model.
If FPs are not acceptable to you (without caring much about FNs) pick the model with higher Precision
If FNs are not acceptable to you (without caring much about FPs) pick the model with higher Recall
Once you decide metric you should be using, try out multiple confidence thresholds (say for example - 0.25, 0.35 and 0.5) for given model to understand for which confidence threshold value the metric you selected works in your favour and also to understand acceptable trade off ranges (say you want Precision of at least 80% and some decent Recall). Once confidence threshold is decided, you use it across different models to find out best performing model.
I have been using sklearn to learn on some data. This is a binary classifcation task and I am using a RBF kernel. My data set is quite unbalanced (80:20) and I'm using only 120 samples, with 10ish features (I've been experimenting with a few less). Since I set class_weight="auto" the accuracy I've calculated from a cross validated (10 folds) gridsearch has dropped dramatically. Why??
I will include a couple of validation accuracy heatmaps to demonstrate the difference.
NOTE: top heatmap is before classweight was changed to auto.
Accuracy is not the best metrics to use when dealing with unbalanced dataset. Let's say you have 99 positive examples and 1 negative example, and if you predict all outputs to be positive, still you will get 99% accuracy, whereas you have mis-classified the only negative example. You might have gotten high accuracy in the first case because your predictions will be on the side which has high number of samples.
When you do class weight = auto, it takes the imbalance into consideration and hence, your predictions might have moved towards center, you can cross-check it using plotting the histograms of predictions.
My suggestion is, don't use accuracy as performance metric, use something like F1 Score or AUC.
I am a frequent user of scikit-learn, I want some insights about the “class_ weight ” parameter with SGD.
I was able to figure out till the function call
plain_sgd(coef, intercept, est.loss_function,
penalty_type, alpha, C, est.l1_ratio,
dataset, n_iter, int(est.fit_intercept),
int(est.verbose), int(est.shuffle), est.random_state,
pos_weight, neg_weight,
learning_rate_type, est.eta0,
est.power_t, est.t_, intercept_decay)
https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/linear_model/stochastic_gradient.py
After this it goes to sgd_fast and I am not very good with cpython. Can you give some celerity on these questions.
I am having a class biased in the dev set where positive class is somewhere 15k and negative class is 36k. does the class_weight will resolve this problem. Or doing undersampling will be a better idea. I am getting better numbers but it’s hard to explain.
If yes then how it actually does it. I mean is it applied on the features penalization or is it a weight to the optimization function. How I can explain this to layman ?
class_weight can indeed help increasing the ROC AUC or f1-score of a classification model trained on imbalanced data.
You can try class_weight="auto" to select weights that are inversely proportional to class frequencies. You can also try to pass your own weights has a python dictionary with class label as keys and weights as values.
Tuning the weights can be achieved via grid search with cross-validation.
Internally this is done by deriving sample_weight from the class_weight (depending on the class label of each sample). Sample weights are then used to scale the contribution of individual samples to the loss function used to trained the linear classification model with Stochastic Gradient Descent.
The feature penalization is controlled independently via the penalty and alpha hyperparameters. sample_weight / class_weight have no impact on it.
I need to find naturally occurring classes of nouns based on their distribution with different preposition (like agentive, instrumental, time, place etc.). I tried using k-means clustering but of less help, it didn't work well, there was a lot of overlap over the classes that I was looking for (probably because of non-globular shape of classes and random initialisation in k-means).
I am now working on using DBSCAN, but I have trouble understanding the epsilon value and mini-points value in this clustering algorithm. Can I use random values or do I need to compute them. Can anybody help? Particularly with epsilon, at least how to compute it if I need to?
Use your domain knowledge to choose the parameters. Epsilon is a radius. You can think of it as a minimum cluster size.
Obviously random values won't work very well. As a heuristic, you can try to look at a k-distance plot; but it's not automatic either.
The first thing to do either way is to choose a good distance function for your data. And perform appropriate normalization.
As for "minPts" it again depends on your data and needs. One user may want a very different value than another. And of course minPts and Epsilon are coupled. If you double epsilon, you will roughly need to increase your minPts by 2^d (for Euclidean distance, because that is how the volume of a hypersphere increases!)
If you want lots of small and fine detailed clusters, choose a low minpts. If you want larger and fewer clusters (and more noise), use a larger minpts. If you don't want any clusters at all, choose minpts larger than your data set size...
It is highly important to select the hyperparameters of DBSCAN algorithm rightly for your dataset and the domain in which it belongs.
eps hyperparameter
In order to determine the best value of eps for your dataset, use the K-Nearest Neighbours approach as explained in these two papers: Sander et al. 1998 and Schubert et al. 2017 (both papers from the original DBSCAN authors).
Here's a condensed version of their approach:
If you have N-dimensional data to begin, then choose n_neighbors in sklearn.neighbors.NearestNeighbors to be equal to 2xN - 1, and find out distances of the K-nearest neighbors (K being 2xN - 1) for each point in your dataset. Sort these distances out and plot them to find the "elbow" which separates noisy points (with high K-nearest neighbor distance) from points (with relatively low K-nearest neighbor distance) which will most likely fall into a cluster. The distance at which this "elbow" occurs is your point of optimal eps.
Here's some python code to illustrate how to do this:
def get_kdist_plot(X=None, k=None, radius_nbrs=1.0):
nbrs = NearestNeighbors(n_neighbors=k, radius=radius_nbrs).fit(X)
# For each point, compute distances to its k-nearest neighbors
distances, indices = nbrs.kneighbors(X)
distances = np.sort(distances, axis=0)
distances = distances[:, k-1]
# Plot the sorted K-nearest neighbor distance for each point in the dataset
plt.figure(figsize=(8,8))
plt.plot(distances)
plt.xlabel('Points/Objects in the dataset', fontsize=12)
plt.ylabel('Sorted {}-nearest neighbor distance'.format(k), fontsize=12)
plt.grid(True, linestyle="--", color='black', alpha=0.4)
plt.show()
plt.close()
k = 2 * X.shape[-1] - 1 # k=2*{dim(dataset)} - 1
get_kdist_plot(X=X, k=k)
Here's an example resultant plot from the code above:
From the plot above, it can be inferred that the optimal value for eps can be assumed at around 22 for the given dataset.
NOTE: I would strongly advice the reader to refer to the two papers cited above (especially Schubert et al. 2017) for additional tips on how to avoid several common pitfalls when using DBSCAN as well as other clustering algorithms.
There are a few articles online –– DBSCAN Python Example: The Optimal Value For Epsilon (EPS) and CoronaVirus Pandemic and Google Mobility Trend EDA –– which basically use the same approach but fail to mention the crucial choice of the value of K or n_neighbors as 2xN-1 when performing the above procedure.
min_samples hyperparameter
As for the min_samples hyperparameter, I agree with the suggestions in the accepted answer. Also, a general guideline for choosing this hyperparameter's optimal value is that it should be set to twice the number of features (Sander et al. 1998). For instance, if each point in the dataset has 10 features, a starting point to consider for min_samples would be 20.
100 periods have been collected from a 3 dimensional periodic signal. The wavelength slightly varies. The noise of the wavelength follows Gaussian distribution with zero mean. A good estimate of the wavelength is known, that is not an issue here. The noise of the amplitude may not be Gaussian and may be contaminated with outliers.
How can I compute a single period that approximates 'best' all of the collected 100 periods?
Time-series, ARMA, ARIMA, Kalman Filter, autoregression and autocorrelation seem to be keywords here.
UPDATE 1: I have no idea how time-series models work. Are they prepared for varying wavelengths? Can they handle non-smooth true signals? If a time-series model is fitted, can I compute a 'best estimate' for a single period? How?
UPDATE 2: A related question is this. Speed is not an issue in my case. Processing is done off-line, after all periods have been collected.
Origin of the problem: I am measuring acceleration during human steps at 200 Hz. After that I am trying to double integrate the data to get the vertical displacement of the center of gravity. Of course the noise introduces a HUGE error when you integrate twice. I would like to exploit periodicity to reduce this noise. Here is a crude graph of the actual data (y: acceleration in g, x: time in second) of 6 steps corresponding to 3 periods (1 left and 1 right step is a period):
My interest is now purely theoretical, as http://jap.physiology.org/content/39/1/174.abstract gives a pretty good recipe what to do.
We have used wavelets for noise suppression with similar signal measured from cows during walking.
I'm don't think the noise is so much of a problem here and the biggest peaks represent actual changes in the acceleration during walking.
I suppose that the angle of the leg and thus accelerometer changes during your experiment and you need to account for that in order to calculate the distance i.e you need to know what is the orientation of the accelerometer in each time step. See e.g this technical note for one to account for angle.
If you need get accurate measures of the position the best solution would be to get an accelerometer with a magnetometer, which also measures orientation. Something like this should work: http://www.sparkfun.com/products/10321.
EDIT: I have looked into this a bit more in the last few days because a similar project is in my to do list as well... We have not used gyros in the past, but we are doing so in the next project.
The inaccuracy in the positioning doesn't come from the white noise, but from the inaccuracy and drift of the gyro. And the error then accumulates very quickly due to the double integration. Intersense has a product called Navshoe, that addresses this problem by zeroing the error after each step (see this paper). And this is a good introduction to inertial navigation.
Periodic signal without noise has the following property:
f(a) = f(a+k), where k is the wavelength.
Next bit of information that is needed is that your signal is composed of separate samples. Every bit of information you've collected are based on samples, which are values of f() function. From 100 samples, you can get the mean value:
1/n * sum(s_i), where i is in range [0..n-1] and n = 100.
This needs to be done for every dimension of your data. If you use 3d data, it will be applied 3 times. Result would be (x,y,z) points. You can find value of s_i from the periodic signal equation simply by doing
s_i(a).x = f(a+k*i).x
s_i(a).y = f(a+k*i).y
s_i(a).z = f(a+k*i).z
If the wavelength is not accurate, this will give you additional source of error or you'll need to adjust it to match the real wavelength of each period. Since
k*i = k+k+...+k
if the wavelength varies, you'll need to use
k_1+k_2+k_3+...+k_i
instead of k*i.
Unfortunately with errors in wavelength, there will be big problems keeping this k_1..k_i chain in sync with the actual data. You'd actually need to know how to regognize the starting position of each period from your actual data. Possibly need to mark them by hand.
Now, all the mean values you calculated would be functions like this:
m(a) :: R->(x,y,z)
Now this is a curve in 3d space. More complex error models will be left as an excersize for the reader.
If you have a copy of Curve Fitting Toolbox, localized regression might be a good choice.
Curve Fitting Toolbox supports both lowess and loess localized regression models for curve and curve fitting.
There is an option for robust localized regression
The following blog post shows how to use cross validation to estimate an optimzal spaning parameter for a localized regression model, as well as techniques to estimate confidence intervals using a bootstrap.
http://blogs.mathworks.com/loren/2011/01/13/data-driven-fitting/