The official document of AllenNLP suggests specifying "validation_data_path" in the configuration file, but what if one wants to construct a dataset from a single source and then randomly split it into train and validation datasets with a given ratio?
Does AllenNLP support this? I would greatly appreciate your comments.
AllenNLP does not have this functionality yet, but we are working on some stuff to get there.
In the meantime, here is how I did it for the VQAv2 reader: https://github.com/allenai/allennlp-models/blob/main/allennlp_models/vision/dataset_readers/vqav2.py#L354
This reader supports Python slicing syntax where you, for example, specify a data_path as "my_source_file[:1000]" to take the first 1000 instances from my_source_file. You can also supply multiple paths by setting data_path: ["file1", "file2[:1000]", "file3[1000-"]]. You can probably steal the top two blocks in that file (line 354 to 369) and put them into your own dataset reader to achieve the same result.
Related
I have an application that needs a configuration file with several inputs which depend on the project that is going to be delivered. Things that are included in this conf file are IP's of databases, activating certain functions depending on the customer's needs, changing the values of some title screens, etc... A short example of a file could be something like:
postgresdb=192.156.98.98
transactions.enabled=true
application.name="client-1-logistics"
historicaldb=196.125.125.16
....
This files can become large and it might be difficult to find which parameters must be changed, specially if the configuration process has to be done by an external department.
I was looking into some kind of tool or framework that allows you to create some sort of questionnaire by which the user answers yes or no questions and fills out boxes with specific IP's or messages and get as a result the configuration file needed. This would be much tidier as you could group the questions into sections and has the potential of customising the configuration process with more context on the different parameters.
Does anyone know of such a framework?. How do you handle this kind of complex configuration processes?
The approach I outline below is not exactly what you are looking for, but it might provide some food for thought.
Use a template engine (example, Velocity, or any of the
several dozen listed in Wikipedia) to create a templated
version of your configuration file, containing lots of boilerplate
configuration that won't change, with the occasional
${variable_name} placeholder (the syntax for a placeholder will
vary from one template engine to another).
Write a small metadata file containing variable_name=value
settings.
Write a trivial program that: (a) parses the metadata file and loads
the variable_name=value settings into a Map (the template engine
might refer to the Map as, say, a context object); (b) uses the
template engine to parse the template file; (c)
merges/evaluates/instantiates the parsed template file with the settings in
the Map; and (d) writes the result to the target
configuration file.
You might be able to use steps 1 and 3 above without change. It is only step 2 that you need to adapt to your questionnaire requirements. Instead of a questionnaire, perhaps you could give users a document that explains how to write the metadata file.
I’m new to deep learning and I was reading some state of art papers and I found that mask r-cnn is utterly used in segmentation and classification of images. I would like to apply it to my MSc project but I got some questions that you may be able to answer. I apologize if this isn’t the right place to do it.
First, I would like to know what are the best strategy to get the annotations. It seems kind of labor intensive and I’m not understanding if there is any easy way. Following that, I want to know if you know any annotation tool for mask r-cnn that generates the binary masks that are manually done by the user.
I hope this can turn into a productive and informative thread so any suggestion, experience would be highly appreciated.
Regards
You can use MASK-RCNN, I recommend it, is a two-stage framework, first you can scan the image and generate areas likely contain an object. And the second stage classifies the proposal drawing bounding boxes.
But the two-big question
how to train a model from scratch? And What happens when we want to
train our own dataset?
You can use annotations downloaded from the internet, or you can start creating your own annotations, this takes a lot of time!
You have tools like:
VIA GGC image annotator
http://www.robots.ox.ac.uk/~vgg/software/via/via_demo.html
it's online and you don't have to download any program. It is the one that I recommend you, save the images in a .json file, and so you can use the class of ballons that comes by default in SAMPLES in the framework MASK R-CNN, you would only have to put your json file and your images and to train your dataset.
But there are always more options, you have labellimg which is also used for annotation and is very well known but save the files in xml, you will have to make a few changes to your Class in python. You also have labelme, labelbox, etc.
I was consulting several references to discover how I may output trained Weka models into Java source code so that I may use the classifiers I am training in actual code for research applications I have been developing.
As I was playing with Weka 3.7, I noticed that while it does output Java code to its main text buffer when use simpler classification (supervised in my case this time) methods such as J48 decision tree, it removes the option (rather, it voids it by removing the ability to checkmark it and fades the text) to output Java code for RandomTree and RandomForest (which are the ones that give me the best performance in my situation).
Note: I am clicking on the "More Options" button and checking "Output source code:".
Does Weka not allow you to output RandomTree or RandomForest as Java code? If so, why? Or if it does and just doesn't put it in the output buffer (since RF is multiple decision trees which I imagine it doesn't want to waste buffer space), how does one go digging up where in the file system Weka outputs java code by default?
Are there any tricks to get Weka to give me my trained RandomForest as Java code? Or is Serialization of the output *.model files my only hope when it comes to RF and RandomTree?
Thanks in advance to those who provide help.
NOTE: (As an addendum to the answer provided below) If you run across a similar situation (requiring you to use your trained classifier/ML model in your code), I recommend following the links posted in the answer that was provided in response to my question. If you do not specifically need the Java code for the RandomForest, as an example, de-serializing the model works quite nicely and fits into Java application code, fulfilling its task as a trained model/hardened algorithm meant to predict future unlabelled instances.
RandomTree and RandomForest can't be output as Java code. I'm not sure for the reasoning why, but they don't implement the "Sourceable" interface.
This explains a little about outputting a classifier as Java code: Link 1
This shows which classifiers can be output as Java code: Link 2
Unfortunately I think the easiest route will be Serialization, although, you could maybe try implementing "Sourceable" for other classifiers on your own.
Another, but perhaps inconvenient solution, would be to use Weka to build the classifier every time you use it. You wouldn't need to load the ".model" file, but you would need to load your training data and relearn the model. Here is a starters guide to building classifiers in your own java code http://weka.wikispaces.com/Use+WEKA+in+your+Java+code.
Solved the problem for myself by turning the output of WEKA's -printTrees option of the RandomForest classifier into Java source code.
http://pielot.org/2015/06/exporting-randomforest-models-to-java-source-code/
Since I am using classifiers with Android, all of the existing options had disadvantages:
shipping Android apps with serialized models didn't reliably work across devices
computing the model on the phone took too much resources
The final code will consist of three classes only: the class with the generated model + two classes to make the classification work.
I have a parameter and, on F4, we can choose the directory. I'm trying to figure out how to choose a folder and read the content of all the files in it (the files are in .CSV) to an internal table. I think I have to use TMP_GUI_DIRECTORY_LIST_FILES function. Hope I'm explaining myself. Thank you.
You'll have to do this manually: first read the list of files, the go through each file and process its contents. There may be some odd function modules to read CSV files, but be aware that many of them are broken - for example, they just clip the lines that exceed a certain length. Therefore I won't recommend any of them - personally, I'd implement the CSV import part myself.
If you have access to the transaction KCLJ in your system you could analyze the coding behind it. This tool has an option to interpret CSV files so you might find interesting function modules that might help you with your tasks.
EDIT: I looked at it very quickly and the piece of coding you could reuse is reconvert_format from include RKCDFILEINCFOR. An example how to call it is located starting from line 128 in the same include.
I have a binary file. I don't know how it's formatted, I only know it comes from a delphi code.
Does it exist any way to analyze a binary file?
Does it exist any "pattern" to analyze and deserialize the binary content of a file with unknown format?
Try these:
Deserialize data: analyze how it's compiled your exe (try File Analyzer). Try to deserialize the binary data with the language discovered. Then serialize it in a xml format (language-indipendent) that every programming language can understand
Analyze the binary data: try to save various versions of the file with little variation and use a diff program to analyze the meaning of every bit with an hex editor. Use it in conjunction with binary hacking techniques (like How to crack a Binary File Format by Frans Faase)
Reverse Engineer the application: try getting code using reverse engineering tools for the programming language used for build the app (found with File Analyzer). Otherwise use disassembler analysis tool like IDA Pro Disassembler
For my hobby project I had to reverse engineer some old game files. My approaches were:
Have a good hex editor.
Look for readable words in the binary file. Note how their distribution is. If the distance between them is constant you know it is a listing.
Look for 2-3 consequent zeros. Might indicate an int32 value.
Some dwords might be pointers into the file.
Try to identify reoccurring patterns in the file.
Seeing lots of C0-CF might indicate RLE compressed data.
I've developed Hexinator (Window & Linux) and Synalyze It! (macOS) exactly for this purpose. These applications allow you to see the binary files like in other hex editors but additionally you can create a "grammar" with the specifics of a binary file format. The grammar contains all the building blocks and is used to parse the file automatically.
Thus you can keep the knowledge you gain in the analysis and apply it to multiple files simultaneously. You can also color-code the bits and pieces of file formats for a quick overview in the hex editor.
The parsing results are displayed in a tree view where you can also modify the files easily (applying endianness et cetera).
Reverse engineering a binary file when you have some idea of what it represents is a very time consuming process. If you have no idea what it is then it will be even harder.
It is possible though, but you have to have a pretty good reason for doing so.
The first step would be to open it up in a hex editor of your choice and see if you can find any English text to point you in the direction of what the file is even supposed to represent. From there, Google "Reverse Engineering binary files", there are much more knowledgeable people than me that have written guides about it.
The "strings" program from GNU binutils is very useful. It will print the strings of printable characters in a file, quite often giving a clue to what a file contains or a program does.
If the data represents serialized Delphi objects, you should start reading about the Delphi serialization process. If that's the case, I think your best bet would be to load it using Delphi and continue your analysis from the IDE. Some informations about Delphi serialization can be found here.
EDIT: if the file does contain serialized delphi objects, then you should write a small delphi program that loads it, and "convert" the data yourself to something neutral, like xml. If you manage to do this, you should check and see if delphi supports serializing to xml. Then, you could access those objects from any language.
The unix "file" command is really useful - I don't know if there is anything like it in windows. You run it like this:
file myfile.ext
And it spits out a text description based on the magic numbers and data contained therein.
Probably it is contained within cygwin.
If you have access to the application that creates the file, you can apply changes to the application, then save the file and see the effects (Keep in mind that numbers are probably stored in little endian):
First create the file repeatedly. If the files are not binary equal, the current date/time is probably stored in the area where hte differences occur.
Maybe you want to repeat that with the software running under different environments, to see if OS version etc are stored, but this is rather unusual.
Next you can try to change single variables and create several files that only differ in the value of this variable. This helps you identify where this variable is stored.
That way you can also exclude variables that are not stored in the file: If you change them, but the files created are identical, they are not stored.
In order to test the hypotheses you worked out with the steps above, edit one of the files and have the application read it.
If you don't have access to the application itself, I suggest that you forget about it and find another way to solve your problem. There is a very high probability that it will be faster...
If file does not give a meaningful answer, you may want to try TRiD by Marco Pontello to determine whether your data is stored in a known format.
Get the Delphi application and open it in IDA Pro freeware version, and find where it writes the file, and decode how it writes the file that way.
Unless it's plan text.
Do you know the program that uses it? If so you can hook that programs write to file function and get an idea of what data its writing, the size of the data and where.
More Info: http://www.codeproject.com/KB/DLL/Win32APIHooking_Trouble.aspx
Unlike traditional hex editors which only display the raw hex bytes of a file, 010 Editor can also parse a file into a hierarchical structure using a Binary Template. The results of running a Binary Template are much easier to understand and edit than using just the raw hex bytes.
http://www.sweetscape.com/010editor/
Try to open it in a hex editor and analyse.