Project Name: Server4recog Verion: 0.8 Git Repo.: https://github.com/AtsushiHashimoto/serv4recog

1. A web service which provides pattern recognition interface through http GET/POST requests.
2. JSON format input/output
3. Group based sample management. This enable to treat several recognition tasks at the same server process.
4. String format class ID.

OSX 10.9

% git clone https://github.com/AtsushiHashimoto/serv4recog.git

mongodb: 2.6.4 http://docs.mongodb.org/manual/tutorial/install-mongodb-on-os-x/

python: 2.7.8 http://docs.python-guide.org/en/latest/starting/install/osx/

python packages: execute following commands at the cloned directory.

% pip install -r packages_requirements.txt

rename 'myapp.conf.example' to 'myapp.conf' and edit the contents for your environment.

%cp myapp.conf.example myapp.conf

give write permission to ./classifiers directory (and files if exists).

%chmod 777 classifiers
%chmod 666 classifiers/*

• You can change the directory in myapp.conf setting.

At the serv4recog directory on a terminal, start up mongod.

% mongod --dbpath ./db

Then, start up the server on another terminal

% python app.py

execute the test script.

% python script/serv4recog_tester.py

If you got recognition result without error, all the instlation processes completed. Congratulation!!

• https://github.com/AtsushiHashimoto/serv4recog/blob/master/tutorial/

HTTP GET:

http://localhost:8080/ml/my_db/my_feature/add?json_data=${SAMPLE}

HTTP POST:

http://localhost:8080/ml/my_db/my_feature/add

• json_data: parameters dumped as a json-format string.

• ml : fixed path name (you can not change).

• my_db : name of database. You can use different name for each of your application.

• my_feature : name of feature_type. You can use any string for each type of feature vector. (e.g. rgb_histogram, hu_moment, SIFT, et al.)

• CAUTION: In your custome applications, '{' and ':' in URL string should be url-encoded!! Please check specification of the HTTP library used in your application.

• _ex) http://localhost:8080/ml/my_db/my_feature/add?json_data={"feature":[0.1,0.9],"ground_truth":"class001"}

sample has following parameters.

• feature: float array that contains feature vector
• ex) "feature":[0.1,0.9,0.3,0.7,0.5,0.5]
• id: sample ID
• ex) "id":"sample00001"
• ground_truth: teacher signal for this sample. (optional, but required as training sample.)
• ex) "ground_truth": "class001"
• likelihood: recognition results (only in output)
• ex) "likelihood:{"svc::${SELECTOR}":{"class001":0.9, "class002":0.1}}
• group: group tag that is used in ${SELECTOR}
• ex) "group":["group01","test_samples"]

Selector limits samples involved in the calculation.

• id: limit samples by its ID.
• ex) {"$or":[{"id":"sample00001"},{"id":"sample00002"}]}
• ground_truth: limit samples by its class
• ex) {"$or":[{"ground_truth":"class001"},{"ground_truth":"class002"}]}
• group: limit samples by its group
• ex) {"group":{"$all":["group01"]}}"

The format follows to pymongo. For more detail, please see online documentation http://docs.mongodb.org/manual/reference/operator/query/

http://localhost:8080/ml/my_db/my_feature/svc/train?json_data={"overwrite":${BOOL}, ${CLASSIFIER-PARAMS}}

• svc: name of classifier. currently, only svc is supported.
• overwrite: overwrite previously trained classifier if true (optional)
• ex) http://localhost:8080/ml/my_db/my_feature/svc/train?json_data={"overwrite":"True", "option":{"kernel":"rbf"}, "name":"my_classifier"}

${CLASSIFIER-PARAMS} has following parameters.

• selector: limits training samples. (optional)
• ex) "selector":${SELECTOR}
• option: argument used in classifier training. (optional)
• ex) "option":${OPTION}
• serv4recog imprimented options
• pca: apply principle component analysis and compress feature vectors to N dimentions.
• sparse: convert feature vector to sparse vector (scipy.sparse.csr_matrix).
• classifier depending options
• http://scikit-learn.org/stable/modules/generated/sklearn.svm.SVC.html
• class_remap: Class remapping by regular expression (for use with hierarchical ground truth labels.)
• ex) "class_remap":{"001":".*001","002":".*002"}
• name: Name of trained classifier data. This is used to identify the classifier at prediction. If name is set, "selector", "option", and "class_remap" can be abbreviated. (optional)
• ex) "name":"classifier20150622"

true or false

A hash identifying argment name and its value for training. For SVC, refer scikit-learn SVC page http://scikit-learn.org/stable/modules/generated/sklearn.svm.SVC.html

http://localhost:8080/ml/my_db/my_feature/svc/predict?json_data={${SAMPLE}, ${CLASSIFIER-PARAMS}}

http://localhost:8080/ml/my_db/my_feature/svc/evaluate?json_data=${CLASSIFIER-PARAMS}

HTTP GET: http://localhost:8080/ml/my_db/my_feature/get_samples?json_data=${Selector}

• return sample IDs

HTTP GET: http://localhost:8080/ml/my_db/my_feature/get_sample_id?json_data={"id":"sample_id"}

http://localhost:8080/ml/my_db/my_feature/clear_samples?json_data={"selector":${SELECTOR}}

http://localhost:8080/ml/my_db/my_feature/classifier/evaluate?json_data=$CLASSIFIER-PARAMS

http://localhost:8080/ml/my_db/my_feature/group?json_data={"selector":${SELECTOR}, "group":["selected_samples01"]}

We welocome new contributers. At first, please branch the project, edit it, and send us the editted branch!