sklearn-cmdline-wrapper

access sklearn (scikit-learn) machine learning library via command line.

Introduction

What is sklearn?

sklearn is a python machine learning library comprised of various machine learning algorithms. See here for detail.

Why command line?

Although you can write a python script every time you like to use functionalities provided by sklearn, it is still annoying to write duplicated script again and again. It is both error prone and inefficient, especially when you want to perform comparisons on different learning algorithms.

This off-the-shell command line tool is here to make your life easier

Prerequisite

Of course You have to install sklearn :) , move steps to the official website of sklearn

This script is only tested on version 0.14.1. For lower versions, something like AdaBoostClassifier will not work.

Features

• only ONE script, you can copy and paste it as you like.
• only supervised learning tasks are supported currently
• parameters(limited, but sufficient) of model can be passed via command-line, you can even make the model utilize mult-cores.
• automatically use sparse matrix if model supports
• libsvm format input

For detailed information: ./learner.py -h

Todo (let's make it a better script)

• compatibility between different versions of sklearn
• support for unsupervised learning tasks
• data visualization for unsupervised learning tasks
• more metrics
• more input file type (like csv, or just space separated columns, better if there's automatically detection).

Known Issue

Due to internal data structures transition, this script is not memory-efficient. It may eat up more memory that you expected. For this reason, it is not recommend to use this script on "Big Data"

Contact

Xinyu Zhou <zxytim[at]gmail[dot]com>

License

GPL v3

Help

output of ./learner.py -h

usage: learner.py [-h] -t {fit,predict,fitpredict,f,p,fp,doc}
				  [--training-file TRAINING_FILE] [--test-file TEST_FILE]
				  [--model-input MODEL_INPUT] [--model-output MODEL_OUTPUT]
				  [-m {logisticr,knnc,mnb,perceptron,lsvc,lasso,abc,ridge,abr,elasticnet,bnb,knnr,sgdc,etr,rfr,nusvr,gbc,dtc,linearr,svc,rfc,etc,gbr,dtr,svr}]
				  [--prediction-file PREDICTION_FILE]
				  [--model-format {pickle,joblib}] [--show-metrics]
				  [-v [VERBOSE]]
				  [model_options [model_options ...]]

command line wrapper for some models in scikit-learn

positional arguments:
  model_options         additional paramters for specific model of format
						"name:type:val", effective only when training is
						needed. type is either int, float or str, which
						abbreviates as i, f and s.

optional arguments:
  -h, --help            show this help message and exit
  -t {fit,predict,fitpredict,f,p,fp,doc}, --task {fit,predict,fitpredict,f,p,fp,doc}
						task to process, see help for detailed information
  --training-file TRAINING_FILE
						input: training file, svm format by default
  --test-file TEST_FILE
						input: test file, svm format by default
  --model-input MODEL_INPUT
						input: model input file, used in prediction
  --model-output MODEL_OUTPUT
						output: model output file, used in fitting
  -m {logisticr,knnc,mnb,perceptron,lsvc,lasso,abc,ridge,abr,elasticnet,bnb,knnr,sgdc,etr,rfr,nusvr,gbc,dtc,linearr,svc,rfc,etc,gbr,dtr,svr}, --model {logisticr,knnc,mnb,perceptron,lsvc,lasso,abc,ridge,abr,elasticnet,bnb,knnr,sgdc,etr,rfr,nusvr,gbc,dtc,linearr,svc,rfc,etc,gbr,dtr,svr}
						model, specified in fitting
  --prediction-file PREDICTION_FILE
						output: prediction file
  --model-format {pickle,joblib}
						model format, pickle(default) or joblib
  --show-metrics        show metric after prediction
  -v [VERBOSE], --verbose [VERBOSE]
						verbose level, -v <level> or multiple -v's or
						something like -vvv

task specification:
	task name: fit, f
		required arguments: training_file, model, model_output
		optional arguments: model_options
	task name: predict, p
		required arguments: test_file, model_input, prediction_file
		optional arguments:
	task name: fitpredict, fp
		required arguments: training_file, model, test_file, prediction_file
		optional arguments: model_options, model_output
	task name: doc
		required arguments: model
		optional arguments:
Notes:
	1. model abbreviation correspondence:
		Abbreviation Model
		abc          AdaBoostClassifier
		abr          AdaBoostRegressor
		bnb          BernoulliNB
		dtc          DecisionTreeClassifier
		dtr          DecisionTreeRegressor
		elasticnet   ElasticNet
		etc          ExtraTreesClassifier
		etr          ExtraTreesRegressor
		gbc          GradientBoostingClassifier
		gbr          GradientBoostingRegressor
		knnc         KNeighborsClassifier
		knnr         KNeighborsRegressor
		lasso        Lasso
		linearr      LinearRegression
		logisticr    LogisticRegression
		lsvc         LinearSVC
		mnb          MultinomialNB
		nusvr        NuSVR
		perceptron   Perceptron
		rfc          RandomForestClassifier
		rfr          RandomForestRegressor
		ridge        Ridge
		sgdc         SGDClassifier
		svc          SVC
		svr          SVR

	2. model compatible with sparse matrix:
		KNeighborsRegressor, SGDClassifier, LinearRegression, LogisticRegression, LinearSVC, KNeighborsClassifier, Ridge, Perceptron, NuSVR, SVR

Examples:
	1. fit(train) a SVR model with sigmoid kernel:
		./learner.py -t f --training-file training-data --model svr \
				--model-output model.svr kernel:s:sigmoid

	2. predict using precomputed model:
		./learner.py -t p --test-file test --model-input model.svr
			--prediction-file pred-result

	3. fit and predict, model saved, verbose output, and show metrics:
		./learner.py -t fp --training-file training-data --model svr \
			--model-output model.svr --test-file test-data \
			--prediction-file pred-result -v --show-metrics

	4. pass parameters for svc model, specify linear kernel:
		./learner.py --task fp --training-file training-data --model svc \
			--test-file test-data --prediction-file pred-result \
			--show-metrics kernel:s:linear

	5. show documents:
		./learner.py -t doc --model svc