I'm still a little unsure of whether questions like these belong on stackoverflow. Is this website only for questions with explicit code? The "How to Format" just tells me it should be a programming question, which it is. I will remove my question if the community thinks otherwise.
I have created a neural network and am predicting reasonable values for most of my data (the task is multi-variate time series forecasting).
I scale my data before inputting it using scikit-learn's MinMaxScaler(0,1) or MinMaxScaler(-1,1) (the two primary scalings I am using).
The model learns, predicts, and I inverse the scaling using MinMaxScaler()'s inverse_transform method to visually see how close my predictions were to the actual values. However, I notice that the inverse_scaled values for a particular part of the vector I predicted have now become very noisy. Here is what I mean (inverse_scaled prediction):
Left end: noisy; right end: not-so noisy.
I initially thought that perhaps my network didn't learn that part of the vector well, so is just outputting ~random values. BUT, I notice that the predicted values before the inverse scaling seem to match the actual values very well, but that these values are typically near 0 or -1 (lower limit of the feature scale) because of the fact that these values have a very large spread (unscaled mean= 1E-1, max= 1E+1 [not an outlier]). Example (scaled prediction):
So, when inverse transforming these values (again, often near -1 or 0), the transformed values exhibit loud noise, as shown in the images.
Questions:
1.) Should I be using a different scaler/scaling differently, perhaps one that exponentially/nonlinearly scales? MinMaxScaler() scales each column. Simply dropping the high-magnitude data isn't an option since they are real, meaningful data. 2.) What other solutions can help this?
Please let me know if you'd like anything else clarified.
Related
I am using the Learning from Data textbook by Yaser Abu-Mostafa et al. I am curious about the following statement in the linear regression chapter and would like to verify that my understanding is correct.
After talking about the "pseudo-inverse" way to get the "best weights" (best for minimizing squared error), i.e w_lin = (X^T X)^-1 X^T y
The statement is "The linear regression weight vector is an attempt to map the inputs X to the outputs y. However, w_lin does not produce y exactly, but produces an estimate X w_lin which differs from y due to in sample error.
If the data is on a hyper-plane, won't X w_lin exactly match y (i.e in-sample error = 0)? I.e above statement is only talking about data that is not linearly separable.
Here, 'w_lin' is the not the same for all data points (all pairs of (X,y)).
The linear regression model finds the best possible weight vector (or best possible 'w_lin') considering all data points such that X*w_lin gives a result very close to 'y' for any data point.
Hence the error will not be zero unless all data points line on a straight line.
The community might not get whole context unless the book is opened because not everything that the author of the book says might have been covered in your post. But let me try to answer.
Whenever any model is formed, there are certain constants used whose value is not known beforehand but are used to fit the line/curve as good as possible. Also, the equations, many a times, contain an element of randomness. Variables that take random values cause some errors when actual and expected outputs are computed.
Suggested reading: Errors and residuals
I need to recognize images with hand-written numerals with known values. Physical objects with the number are always identical but come in slight variations of positions/scale/lighting. They are about 100 in number, having about 100x500 px in size.
In the first pass, the code should "learn" possible inputs, and then recognize them (classify them as being close to one of the "training" images) when they come again.
I was mostly following the Feature Matching Python-OpenCV tutorial
Input images are analyzed first, keypoints & descriptors are remembered in the orbTrained list:
import cv2
import collections
ORBTrained=collections.namedtuple('ORBTrained',['kp','des','img'])
orbTrained=[]
for img in trainingImgs:
z2=preprocessImg(img)
orb=cv2.ORB_create(nfeatures=400,patchSize=30,edgeThreshold=0)
kp,des=orb.detectAndCompute(z2,None)
orbTrained.append(ORBTrained(kp=kp,des=des,img=z2))
z3=cv2.drawKeypoints(z2,kp,None,color=(0,255,0),flags=0)
A typical result of this first stage looks like this:
Then in the next loop, for each real input image, cycle through all training images to see which is matching the best:
ORBMatch=collections.namedtuple('ORBMatch',['dist','match','train'])
for img in inputImgs:
z2=preprocessNum(img)
orb=cv2.ORB_create(nfeatures=400,patchSize=30,edgeThreshold=0)
kp,des=orb.detectAndCompute(z2,None)
bf=cv2.BFMatcher(cv2.NORM_HAMMING,crossCheck=True)
mm=[]
for train in orbTrained:
m=bf.match(des,train.des)
dist=sum([m_.distance for m_ in m])
mm.append(ORBMatch(dist=dist,match=m,train=train))
# sort matching images based on score
mm.sort(key=lambda m: m.dist)
print([m.dist for m in mm[:5]])
best=mm[0]
best.match.sort(key=lambda x:x.distance) # sort matches in the best match
z3=cv2.drawMatches(z2,kp,best.train.img,best.train.kp,best.match[:50],None,flags=2)
The result I get is nonsensical, and consistently so (only when I run with pixel-identical input, the result is correct):
What is the problem? Am I completely misunderstanding what to do, or do I just need to tune some parameters?
First, are you sure you're not reinventing the wheel by creating your own OCR library? There are many free frameworks, some of which support training with custom character sets.
Second, you should understand what feature matching is. It will find similar small areas, but isn't aware of other feature pairs. It will match similar corners of characters, not the characters itself. You might experiment with larger patchSize so that it covers at least half of the digit.
You can minimize false pairs by running feature detection only on a single digit at a time using thresholding and contours to find character bounds.
If the text isn't rotated, using roation-invariant feature descriptor such as ORB isn't the best option, try rotation-variant descriptor, such as FAST.
According to authors of papers (ORB and ORB-SLAM) ORB is invariant to rotation and scale "in a certain range". May be you should first match for small scale or rotation change.
Data: When I have N rows of data like this: (x,y,z) where logically f(x,y)=z, that is z is dependent on x and y, like in my case (setting1, setting2 ,signal) . Different x's and y's can lead to the same z, but the z's wouldn't mean the same thing.
There are 30 unique setting1, 30 setting2 and 1 signal for each (setting1, setting2)-pairing, hence 900 signal values.
Data set: These [900,3] data points are considered 1 data set. I have many samples of these data sets.
I want to make a classification based on these data sets, but I need to flatten the data (make them all into one row). If I flatten it, I will duplicate all the setting values (setting1 and setting2) 30 times, i.e. I will have a row with 3x900 columns.
Question:
Is it correct to keep all the duplicate setting1,setting2 values in the data set? Or should I remove them and only include the unique values a single time?, i.e. have a row with 30 + 30 + 900 columns. I'm worried, that the logical dependency of the signal to the settings will be lost this way. Is this relevant? Or shouldn't I bother including the settings at all (e.g. due to correlations)?
If I understand correctly, you are training NN on a sample where each observation is [900,3].
You are flatning it and getting an input layer of 3*900.
Some of those values are a result of a function on others.
It is important which function, as if it is a liniar function, NN might not work:
From here:
"If inputs are linearly dependent then you are in effect introducing
the same variable as multiple inputs. By doing so you've introduced a
new problem for the network, finding the dependency so that the
duplicated inputs are treated as a single input and a single new
dimension in the data. For some dependencies, finding appropriate
weights for the duplicate inputs is not possible."
Also, if you add dependent variables you risk the NN being biased towards said variables.
E.g. If you are running LMS on [x1,x2,x3,average(x1,x2)] to predict y, you basically assign a higher weight to the x1 and x2 variables.
Unless you have a reason to believe that those weights should be higher, don't include their function.
I was not able to find any link to support, but my intuition is that you might want to decrease your input layer in addition to omitting the dependent values:
From professor A. Ng's ML Course I remember that the input should be the minimum amount of values that are 'reasonable' to make the prediction.
Reasonable is vague, but I understand it so: If you try to predict the price of a house include footage, area quality, distance from major hub, do not include average sun spot activity during the open home day even though you got that data.
I would remove the duplicates, I would also look for any other data that can be omitted, maybe run PCA over the full set of Nx[3,900].
I'm setting up the new Tensorflow Object Detection API to find small objects in large areas of satellite imagery. It works quite well - it finds all 10 objects I want, but I also get 50-100 false positives [things that look a little like the target object, but aren't].
I'm using the sample config from the 'pets' tutorial, to fine-tune the faster_rcnn_resnet101_coco model they offer. I've started small, with only 100 training examples of my objects (just 1 class). 50 examples in my validation set. Each example is a 200x200 pixel image with a labeled object (~40x40) in the center. I train until my precision & loss curves plateau.
I'm relatively new to using deep learning for object detection. What is the best strategy to increase my precision? e.g. Hard-negative mining? Increase my training dataset size? I've yet to try the most accurate model they offer faster_rcnn_inception_resnet_v2_atrous_coco as i'd like to maintain some speed, but will do so if needed.
Hard-negative mining seems to be a logical step. If you agree, how do I implement it w.r.t setting up the tfrecord file for my training dataset? Let's say I make 200x200 images for each of the 50-100 false positives:
Do I create 'annotation' xml files for each, with no 'object' element?
...or do I label these hard negatives as a second class?
If I then have 100 negatives to 100 positives in my training set - is that a healthy ratio? How many negatives can I include?
I've revisited this topic recently in my work and thought I'd update with my current learnings for any who visit in the future.
The topic appeared on Tensorflow's Models repo issue tracker. SSD allows you to set the ratio of how many negative:postive examples to mine (max_negatives_per_positive: 3), but you can also set a minimum number for images with no postives (min_negatives_per_image: 3). Both of these are defined in the model-ssd-loss config section.
That said, I don't see the same option in Faster-RCNN's model configuration. It's mentioned in the issue that models/research/object_detection/core/balanced_positive_negative_sampler.py contains the code used for Faster-RCNN.
One other option discussed in the issue is creating a second class specifically for lookalikes. During training, the model will attempt to learn class differences which should help serve your purpose.
Lastly, I came across this article on Filter Amplifier Networks (FAN) that may be informative for your work on aerial imagery.
===================================================================
The following paper describes hard negative mining for the same purpose you describe:
Training Region-based Object Detectors with Online Hard Example Mining
In section 3.1 they describe using a foreground and background class:
Background RoIs. A region is labeled background (bg) if its maximum
IoU with ground truth is in the interval [bg lo, 0.5). A lower
threshold of bg lo = 0.1 is used by both FRCN and SPPnet, and is
hypothesized in [14] to crudely approximate hard negative mining; the
assumption is that regions with some overlap with the ground truth are
more likely to be the confusing or hard ones. We show in Section 5.4
that although this heuristic helps convergence and detection accuracy,
it is suboptimal because it ignores some infrequent, but important,
difficult background regions. Our method removes the bg lo threshold.
In fact this paper is referenced and its ideas are used in Tensorflow's object detection losses.py code for hard mining:
class HardExampleMiner(object):
"""Hard example mining for regions in a list of images.
Implements hard example mining to select a subset of regions to be
back-propagated. For each image, selects the regions with highest losses,
subject to the condition that a newly selected region cannot have
an IOU > iou_threshold with any of the previously selected regions.
This can be achieved by re-using a greedy non-maximum suppression algorithm.
A constraint on the number of negatives mined per positive region can also be
enforced.
Reference papers: "Training Region-based Object Detectors with Online
Hard Example Mining" (CVPR 2016) by Srivastava et al., and
"SSD: Single Shot MultiBox Detector" (ECCV 2016) by Liu et al.
"""
Based on your model config file, the HardMinerObject is returned by losses_builder.py in this bit of code:
def build_hard_example_miner(config,
classification_weight,
localization_weight):
"""Builds hard example miner based on the config.
Args:
config: A losses_pb2.HardExampleMiner object.
classification_weight: Classification loss weight.
localization_weight: Localization loss weight.
Returns:
Hard example miner.
"""
loss_type = None
if config.loss_type == losses_pb2.HardExampleMiner.BOTH:
loss_type = 'both'
if config.loss_type == losses_pb2.HardExampleMiner.CLASSIFICATION:
loss_type = 'cls'
if config.loss_type == losses_pb2.HardExampleMiner.LOCALIZATION:
loss_type = 'loc'
max_negatives_per_positive = None
num_hard_examples = None
if config.max_negatives_per_positive > 0:
max_negatives_per_positive = config.max_negatives_per_positive
if config.num_hard_examples > 0:
num_hard_examples = config.num_hard_examples
hard_example_miner = losses.HardExampleMiner(
num_hard_examples=num_hard_examples,
iou_threshold=config.iou_threshold,
loss_type=loss_type,
cls_loss_weight=classification_weight,
loc_loss_weight=localization_weight,
max_negatives_per_positive=max_negatives_per_positive,
min_negatives_per_image=config.min_negatives_per_image)
return hard_example_miner
which is returned by model_builder.py and called by train.py. So basically, it seems to me that simply generating your true positive labels (with a tool like LabelImg or RectLabel) should be enough for the train algorithm to find hard negatives within the same images. The related question gives an excellent walkthrough.
In the event you want to feed in data that has no true positives (i.e. nothing should be classified in the image), just add the negative image to your tfrecord with no bounding boxes.
I think I was passing through the same or close scenario and it's worth it to share with you.
I managed to solve it by passing images without annotations to the trainer.
On my scenario I'm building a project to detect assembly failures from my client's products, at real time.
I successfully achieved very robust results (for production env) by using detection+classification for components that has explicity a negative pattern (e.g. a screw that has screw on/off(just the hole)) and only detection for things that doesn't has the negative pattens (e.g. a tape that can be placed anywhere).
On the system it's mandatory that the user record 2 videos, one containing the positive scenario and another containing the negative (or the n videos, containing n patterns of positive and negative so the algorithm can generalize).
After a while testing I found out that if I register to detected only tape the detector was giving very confident (0.999) false positive detections of tape. It was learning the pattern where the tape was inserted instead of the tape itself. When I had another component (like a screw on it's negative format) I was passing the negative pattern of tape without being explicitly aware of it, so the FPs didn't happen.
So I found out that, in this scenario, I had to necessarily pass the images without tape so it could differentiate between tape and no-tape.
I considered two alternatives to experiment and try to solve this behavior:
Train passing an considerable amount of images that doesn't has any annotation (10% of all my negative samples) along with all images that I have real annotations.
On the images that I don't have annotation I create a dummy annotation with a dummy label so I could force the detector to train with that image (thus learning the no-tape pattern). Later on, when get the dummy predictions, just ignore them.
Concluded that both alternatives worked perfectly on my scenario.
The training loss got a little messy but the predictions work with robustness for my very controlled scenario (the system's camera has its own box and illumination to decrease variables).
I had to make two little modifications for the first alternative to work:
All images that didn't had any annotation I passed a dummy annotation (class=None, xmin/ymin/xmax/ymax=-1)
When generating the tfrecord files I use this information (xmin == -1, in this case) to add an empty list for the sample:
def create_tf_example(group, path, label_map):
with tf.gfile.GFile(os.path.join(path, '{}'.format(group.filename)), 'rb') as fid:
encoded_jpg = fid.read()
encoded_jpg_io = io.BytesIO(encoded_jpg)
image = Image.open(encoded_jpg_io)
width, height = image.size
filename = group.filename.encode('utf8')
image_format = b'jpg'
xmins = []
xmaxs = []
ymins = []
ymaxs = []
classes_text = []
classes = []
for index, row in group.object.iterrows():
if not pd.isnull(row.xmin):
if not row.xmin == -1:
xmins.append(row['xmin'] / width)
xmaxs.append(row['xmax'] / width)
ymins.append(row['ymin'] / height)
ymaxs.append(row['ymax'] / height)
classes_text.append(row['class'].encode('utf8'))
classes.append(label_map[row['class']])
tf_example = tf.train.Example(features=tf.train.Features(feature={
'image/height': dataset_util.int64_feature(height),
'image/width': dataset_util.int64_feature(width),
'image/filename': dataset_util.bytes_feature(filename),
'image/source_id': dataset_util.bytes_feature(filename),
'image/encoded': dataset_util.bytes_feature(encoded_jpg),
'image/format': dataset_util.bytes_feature(image_format),
'image/object/bbox/xmin': dataset_util.float_list_feature(xmins),
'image/object/bbox/xmax': dataset_util.float_list_feature(xmaxs),
'image/object/bbox/ymin': dataset_util.float_list_feature(ymins),
'image/object/bbox/ymax': dataset_util.float_list_feature(ymaxs),
'image/object/class/text': dataset_util.bytes_list_feature(classes_text),
'image/object/class/label': dataset_util.int64_list_feature(classes),
}))
return tf_example
Part of the traning progress:
Currently I'm using tensorflow object detection along with tensorflow==1.15, using faster_rcnn_resnet101_coco.config.
Hope it will solve someone's problem as I didn't found any solution on the internet. I read a lot of people telling that faster_rcnn is not adapted for negative training for FPs reduction but my tests proved the opposite.
This my sound as very naive question. I checked on google and many YouTube videos for beginners and pretty much, all explain data weighting as something the most obvious. I still do not understand why data is being weighted.
Let's assume I have four features:
a b c d
1 2 1 4
If I pass each value to Sigmond function, I'll receive -1 >< 1 value already.
I really don't understand why data needs or it is recommended to be weighted first. If you could explain to me this in very simple manner, I would appreciate it a lot.
I think you are not talking about weighing data but features.
A feature is a column in your table and as data I would understand rows.
The confusion comes now from the fact that weighing rows is also sometimes sensible, e.g., if you want to punish misclassification of positive class more.
Why do we need to weigh features?
I assume you are talking about a modle like
prediction = sigmoid(sum_i weight_i * feature_i) > base
Let's assume you want to predict whether a person is overweight based on Bodyweight, height, and age.
In R we can generate a sample dataset as
height = rnorm(100,1.80,0.1) #normal distributed mean 1.8,variance 0.1
weight = rnorm(100,70,10)
age = runif(100,0,100)
ow = weight / (height**2)>25 #overweight if BMI > 25
data = data.frame(height,weight,age,bc,ow)
if we now plot the data you can see that at least my sample of the data can be separated with a straight line in weight/height. However, age does not provide any value. If we weight it prior to the sum/sigmoid you can put all factors into relation.
Furthermore, as you can see from the following plot the weight/height have a very different domain. Hence, they need to be put into relation, such that the line in the following plot has the right slope, as the value of weight have are one order of magnitude larger