SVM Model does not support probability estimation? - svm

I am doing some classification task with Support Vector Machines (SVM).
I am using libSVM (with Matlab support) to predict probability estimates matrix. However, the libSVM displays message that;
Model does not support probabiliy estimates
Below is my sample code;
(train_label contains labels for training data and test_label contains label for test data)
model = svmtrain(train_label, train_data, '-t 2 -g .01 -c 0.7 -b 1);
[y,accuracy,prob_estimates]=svmpredict(test_label,test_data,model,'-b 1');
Can someone tell me if there is something wrong with the way I am doing it? Any help/suggestion will be appreciated.


Don't know about the Matlab implementation, but usually you have to set this option:
-b probability_estimates: whether to train a SVC or SVR model for probability estimates, 0 or 1 (default 0)


I am using libsvm in the same way without any problem.
In your code only a ' is missing in the following line
model = svmtrain(train_label, train_data, '-t 2 -g .01 -c 0.7 -b 1);
It should be
model = svmtrain(train_label, train_data, '-t 2 -g .01 -c 0.7 -b 1');


I had the same problem, model hasn't got ProbA and ProbB in it.
Before it was like this and giving error:
linear_model = svmtrain(trainClass, trainData, ['-t 0', cmd]);
Then I changed it to this, error dissappared:) - removed cmd and put exact values
linear_model = svmtrain(trainClass, trainData, ['-t 0 -c 1 -g 0.125 -b 1']);
if still gives error try to change c and g parameters.
Hope this helps.


It is because your model does not support probabiliy estimates.
You should use '-b 1' option both at training and testing process.
See also: https://stackoverflow.com/a/43509667/7893127


You may just train the model with default parameter.
Try to use '-b 1' when you are training and testing programe.


C:\setup\python36\Lib\site-packages\svm.py default value of self.probability is 0. You can set it 1.

Related

Clarifications on training job parameters with Tensorflow

Im using the new Tensorflow object detection API.
I need to replicate training parameters used on a paper but Im a bit confused.
In the paper is stated
When training neural network models, their base confguration is similar to that used to
train on the COCO 2017 dataset. For the unambiguous comparison of the selected models, the total number of
training steps was set to 100 equal to 100′000 iterations of learning.
Inside model_main_tf2.py, which is the script used to start the training, I can read the following:
"""Creates and runs TF2 object detection models.
For local training/evaluation run:
PIPELINE_CONFIG_PATH=path/to/pipeline.config
MODEL_DIR=/tmp/model_outputs
NUM_TRAIN_STEPS=10000
SAMPLE_1_OF_N_EVAL_EXAMPLES=1
python model_main_tf2.py -- \
--model_dir=$MODEL_DIR --num_train_steps=$NUM_TRAIN_STEPS \
--sample_1_of_n_eval_examples=$SAMPLE_1_OF_N_EVAL_EXAMPLES \
--pipeline_config_path=$PIPELINE_CONFIG_PATH \
--alsologtostderr
"""
Also, you can specify the num_steps and total_steps parameters in the pipeline.config file (used by the training script):
train_config: {
batch_size: 1
sync_replicas: true
startup_delay_steps: 0
replicas_to_aggregate: 8
num_steps: 50000
optimizer {
momentum_optimizer: {
learning_rate: {
cosine_decay_learning_rate {
learning_rate_base: .16
total_steps: 50000
warmup_learning_rate: 0
warmup_steps: 2500
}
}
momentum_optimizer_value: 0.9
}
use_moving_average: false
}
So, what Im not understanding is how should I map what is written in the paper with tensorflow parameters.
What is the num steps and total_steps inside the pipeline.config file?
What is the NUM_TRAIN_STEPS argument instead?
Does it overwrite config file steps or its a completely different thing?
If more details are needed feel free to ask.

Input 'input_image' of layer '63' not found in any of the outputs of the preceeding layers

Update #1 (original question and details below):
As per the suggestion of #MatthijsHollemans below I've tried to run this by removing dynamic_axes from the initial create_onnx step below. This removed both:
Description of image feature 'input_image' has missing or non-positive width 0.
and
Input 'input_image' of layer '63' not found in any of the outputs of the preceeding layers.
Unfortunately this opens up two sub-questions:
I still want to have a functional ONNX model. Is there a more appropriate way to make H and W dynamic? Or should I be saving two versions of the ONNX model, one without dynamic_axes for the CoreML conversion, and one with for use as a valid ONNX model?
Although this solves the compilation error in xcode (specified below) it introduces the following runtime issues:
Finalizing CVPixelBuffer 0x282f4c5a0 while lock count is 1.
[espresso] [Espresso::handle_ex_plan] exception=Invalid X-dimension 1/480 status=-7
[coreml] Error binding image input buffer input_image: -7
[coreml] Failure in bindInputsAndOutputs.
I am calling this the same way I was calling the fixed size model, which does still work fine. The image dimensions are 640 x 480.
As specified below the model should accept any image between 64x64 and higher.
For flexible shape models, do I need to provide an input differently in xcode?
Original Question (parts still relevant)
I have been slowly working on converting a style transfer model from pytorch > onnx > coreml. One of the issues that has been a struggle is flexible/dynamic input + output shape.
This method (besides i/o renaming) has worked well on iOS 12 & 13 when using a static input shape.
I am using the following code to do the onnx > coreml conversion:
def create_coreml(name):
mlmodel = convert(
model="onnx/" + name + ".onnx",
preprocessing_args={'is_bgr': True},
deprocessing_args={'is_bgr': True},
image_input_names=['input_image'],
image_output_names=['stylized_image'],
minimum_ios_deployment_target='13'
)
spec = mlmodel.get_spec()
img_size_ranges = flexible_shape_utils.NeuralNetworkImageSizeRange()
img_size_ranges.add_height_range((64, -1))
img_size_ranges.add_width_range((64, -1))
flexible_shape_utils.update_image_size_range(
spec,
feature_name='input_image',
size_range=img_size_ranges)
flexible_shape_utils.update_image_size_range(
spec,
feature_name='stylized_image',
size_range=img_size_ranges)
mlmodel = coremltools.models.MLModel(spec)
mlmodel.save("mlmodel/" + name + ".mlmodel")
Although the conversion 'succeeds' there are a couple of warnings (spaces added for readability):
Translation to CoreML spec completed. Now compiling the CoreML model.
/usr/local/lib/python3.7/site-packages/coremltools/models/model.py:111:
RuntimeWarning: You will not be able to run predict() on this Core ML model. Underlying exception message was:
Error compiling model:
"Error reading protobuf spec. validator error: Description of image feature 'input_image' has missing or non-positive width 0.".
RuntimeWarning)
Model Compilation done.
/usr/local/lib/python3.7/site-packages/coremltools/models/model.py:111:
RuntimeWarning: You will not be able to run predict() on this Core ML model. Underlying exception message was:
Error compiling model:
"compiler error: Input 'input_image' of layer '63' not found in any of the outputs of the preceeding layers.
".
RuntimeWarning)
If I ignore these warnings and try to compile the model for latest targets (13.0) I get the following error in xcode:
coremlc: Error: compiler error: Input 'input_image' of layer '63' not found in any of the outputs of the preceeding layers.
Here is what the problematic area appears to look like in netron:
My main question is how can I get these two warnings out of the way?
Happy to provide any other details.
Thanks for any advice!
Below is my pytorch > onnx conversion:
def create_onnx(name):
prior = torch.load("pth/" + name + ".pth")
model = transformer.TransformerNetwork()
model.load_state_dict(prior)
dummy_input = torch.zeros(1, 3, 64, 64) # I wasn't sure what I would set the H W to here?
torch.onnx.export(model, dummy_input, "onnx/" + name + ".onnx",
verbose=True,
opset_version=10,
input_names=["input_image"], # These are being renamed from garbled originals.
output_names=["stylized_image"], # ^
dynamic_axes={'input_image':
{2: 'height', 3: 'width'},
'stylized_image':
{2: 'height', 3: 'width'}}
)
onnx.save_model(original_model, "onnx/" + name + ".onnx")

How to convert scaled data into original data in SVM

I am new to SVM. I am using Support Vector Regression for forecasting in LibSVM. For LibSVM, I firstly scaled [0,1] my training and testing set in the same range and then select the optimal parameters. After I run svm-train and svm-predict, I got the predicted values for testing set in a scaled format.How do i get the original values back for output? Which conversion I need to use.
svm-scale -l 0 -u 1 -s range uni.train> unitrain.scale
svm-scale -r range uni.test > unitest.scale
svm-train -c 32768 -g 0.03125 unitrain.scale
svm-predict unitest.scale unitrain.scale.model uni.txt
Accuracy = 87.5% (266/304) (classification)
svm-train -s 3 -c 32768 -g 0.03125 unitrain.scale
svm-predict unitest.scale unitrain.scale.model uni.txt
Mean squared error = 0.558619 (regression)
Squared correlation coefficient = 0.514304 (regression)
After this, the output is uni.txt i.e. 0.619153 -1.10326 -0.94956 ........ -1.00426 -0.311161 -0.725936
How do i get the original values for output in uni.txt, these are in scaled format?

scikit-learn roc_curve: why does it return a threshold value = 2 some time?

Correct me if I'm wrong: the "thresholds" returned by scikit-learn's roc_curve should be an array of numbers that are in [0,1]. However, it sometimes gives me an array with the first number close to "2". Is it a bug or I did sth wrong? Thanks.
In [1]: import numpy as np
In [2]: from sklearn.metrics import roc_curve
In [3]: np.random.seed(11)
In [4]: aa = np.random.choice([True, False],100)
In [5]: bb = np.random.uniform(0,1,100)
In [6]: fpr,tpr,thresholds = roc_curve(aa,bb)
In [7]: thresholds
Out[7]:
array([ 1.97396826, 0.97396826, 0.9711752 , 0.95996265, 0.95744405,
0.94983331, 0.93290463, 0.93241372, 0.93214862, 0.93076592,
0.92960511, 0.92245024, 0.91179548, 0.91112166, 0.87529458,
0.84493853, 0.84068543, 0.83303741, 0.82565223, 0.81096657,
0.80656679, 0.79387241, 0.77054807, 0.76763223, 0.7644911 ,
0.75964947, 0.73995152, 0.73825262, 0.73466772, 0.73421299,
0.73282534, 0.72391126, 0.71296292, 0.70930102, 0.70116428,
0.69606617, 0.65869235, 0.65670881, 0.65261474, 0.6487222 ,
0.64805644, 0.64221486, 0.62699782, 0.62522484, 0.62283401,
0.61601839, 0.611632 , 0.59548669, 0.57555854, 0.56828967,
0.55652111, 0.55063947, 0.53885029, 0.53369398, 0.52157349,
0.51900774, 0.50547317, 0.49749635, 0.493913 , 0.46154029,
0.45275916, 0.44777116, 0.43822067, 0.43795921, 0.43624093,
0.42039077, 0.41866343, 0.41550367, 0.40032843, 0.36761763,
0.36642721, 0.36567017, 0.36148354, 0.35843793, 0.34371331,
0.33436415, 0.33408289, 0.33387442, 0.31887024, 0.31818719,
0.31367915, 0.30216469, 0.30097917, 0.29995201, 0.28604467,
0.26930354, 0.2383461 , 0.22803687, 0.21800338, 0.19301808,
0.16902881, 0.1688173 , 0.14491946, 0.13648451, 0.12704826,
0.09141459, 0.08569481, 0.07500199, 0.06288762, 0.02073298,
0.01934336])

Most of the time these thresholds are not used, for example in calculating the area under the curve, or plotting the False Positive Rate against the True Positive Rate.
Yet to plot what looks like a reasonable curve, one needs to have a threshold that incorporates 0 data points. Since Scikit-Learn's ROC curve function need not have normalised probabilities for thresholds (any score is fine), setting this point's threshold to 1 isn't sufficient; setting it to inf is sensible but coders often expect finite data (and it's possible the implementation also works for integer thresholds). Instead the implementation uses max(score) + epsilon where epsilon = 1. This may be cosmetically deficient, but you haven't given any reason why it's a problem!

From the documentation:
thresholds : array, shape = [n_thresholds]
Decreasing thresholds on the decision function used to compute
fpr and tpr. thresholds[0] represents no instances being predicted
and is arbitrarily set to max(y_score) + 1.
So the first element of thresholds is close to 2 because it is max(y_score) + 1, in your case thresholds[1] + 1.

this seems like a bug to me - in roc_curve(aa,bb), 1 is added to the first threshold. You should create an issue here https://github.com/scikit-learn/scikit-learn/issues

using SVM for binary classification

I am using sVM-light for binary classification.and I am using SVM in the learning mode.
I have my train.dat file ready.but when i run this command ,instead of creating file model ,it writes somethings in terminal:
my command:
./svm_learn example1/train.dat example1/model
output:
Scanning examples...done
Reading examples into memory...Feature numbers must be larger or equal to 1!!!
: Success
LINE: -1 0:1.0 6:1.0 16:1.0 18:1.0 28:1.0 29:1.0 31:1.0 48:1.0 58:1.0 73:1.0 82:1.0 93:1.0 95:1.0 106:1.0 108:1.0 118:1.0 121:1.0 122:1.0151:1.0 164:1.0 167:1.0 169:1.0 170:1.0 179:1.0 190:1.0 193:1.0 220:1.0 237:1.0250:1.0 252:1.0 267:1.0 268:1.0 269:1.0 278:1.0 283:1.0 291:1.0 300:1.0 305:1.0320:1.0 332:1.0 336:1.0 342:1.0 345:1.0 348:1.0 349:1.0 350:1.0 368:1.0 370:1.0384:1.0 390:1.0 394:1.0 395:1.0 396:1.0 397:1.0 400:1.0 401:1.0 408:1.0 416:1.0427:1.0 433:1.0 435:1.0 438:1.0 441:1.0 446:1.0 456:1.0 471:1.0 485:1.0 510:1.0523:1.0 525:1.0 526:1.0 532:1.0 540:1.0 553:1.0 567:1.0 568:1.0 581:1.0 583:1.0604:1.0 611:1.0 615:1.0 616:1.0 618:1.0 623:1.0 624:1.0 626:1.0 651:1.0 659:1.0677:1.0 678:1.0 683:1.0 690:1.0 694:1.0 699:1.0 713:1.0 714:1.0 720:1.0 722:1.0731:1.0 738:1.0 755:1.0 761:1.0 763:1.0 768:1.0 776:1.0 782:1.0 792:1.0 817:1.0823:1.0 827:1.0 833:1.0 834:1.0 838:1.0 842:1.0 848:1.0 851:1.0 863:1.0 867:1.0890:1.0 900:1.0 903:1.0 923:1.0 935:1.0 942:1.0 946:1.0 947:1.0 949:1.0 956:1.0962:1.0 965:1.0 968:1.0 983:1.0 986:1.0 987:1.0 990:1.0 998:1.0 1007:1.0 1014:1.0 1019:1.0 1022:1.0 1024:1.0 1029:1.0 1030:1.01032:1.0 1047:1.0 1054:1.0 1063:1.0 1069:1.0 1076:1.0 1085:1.0 1093:1.0 1098:1.0 1108:1.0 1109:1.01116:1.0 1120:1.0 1133:1.0 1134:1.0 1135:1.0 1138:1.0 1139:1.0 1144:1.0 1146:1.0 1148:1.0 1149:1.01161:1.0 1165:1.0 1169:1.0 1170:1.0 1177:1.0 1187:1.0 1194:1.0 1212:1.0 1214:1.0 1239:1.0 1243:1.01251:1.0 1257:1.0 1274:1.0 1278:1.0 1292:1.0 1297:1.0 1304:1.0 1319:1.0 1324:1.0 1325:1.0 1353:1.01357:1.0 1366:1.0 1374:1.0 1379:1.0 1392:1.0 1394:1.0 1407:1.0 1412:1.0 1414:1.0 1419:1.0 1433:1.01435:1.0 1437:1.0 1453:1.0 1463:1.0 1464:1.0 1469:1.0 1477:1.0 1481:1.0 1487:1.0 1506:1.0 1514:1.01519:1.0 1526:1.0 1536:1.0 1549:1.0 1551:1.0 1553:1.0 1561:1.0 1569:1.0 1578:1.0 1603:1.0 1610:1.01615:1.0 1617:1.0 1625:1.0 1638:1.0 1646:1.0 1663:1.0 1666:1.0 1672:1.0 1681:1.0 1690:1.0 1697:1.01699:1.0 1706:1.0 1708:1.0 1717:1.0 1719:1.0 1732:1.0 1737:1.0 1756:1.0 1766:1.0 1771:1.0 1789:1.01804:1.0 1805:1.0 1808:1.0 1814:1.0 1815:1.0 1820:1.0 1824:1.0 1832:1.0 1841:1.0 1844:1.0 1852:1.01861:1.0 1875:1.0 1899:1.0 1902:1.0 1904:1.0 1905:1.0 1917:1.0 1918:1.0 1919:1.0 1921:1.0 1926:1.01934:1.0 1937:1.0 1942:1.0 1956:1.0 1965:1.0 1966:1.0 1970:1.0 1971:1.0 1980:1.0 1995:1.0 2000:1.02009:1.0 2010:1.0 2012:1.0 2015:1.0 2018:1.0 2022:1.0 2047:1.0 2076:1.0 2082:1.0 2095:1.0 2108:1.02114:1.0 2123:1.0 2130:1.0 2133:1.0 2141:1.0 2142:1.0 2143:1.0 2148:1.0 2157:1.0 2160:1.0 2162:1.02170:1.0 2195:1.0 2199:1.0 2201:1.0 2202:1.0 2205:1.0 2211:1.0 2218:1.0
I dont know what to do.
p.s.when i make my train.dat very shorter ,everything works fine!!!
Thank you

From what I could interpret from the log, your training set has an issue.
The first few characters of the training row that has issue are
-1 0:1.0 6:1.0
The issue is not with the size but with feature indexing. You are starting your feature index at 0 (0:1) whereas svmlight requires that all feature index be equal or greater than 1.
Change the indexing to start at 1 and it should work fine.

Resources