Conditional Random Field - Platform For AI - Alibaba Cloud Documentation Center

This topic describes the Conditional Random Field component provided by Machine Learning Designer (formerly known as Machine Learning Studio).

A conditional random field (CRF) is a conditional probability distribution model of a group of output random variables based on a group of input random variables. This model presumes that the output random variables constitute a Markov random field (MRF). CRFs can be used in different prediction scenarios. The linear chain CRF is mostly used, especially in annotation scenarios. For more information, see Wikipedia.

Configure the component

You can use one of the following methods to configure the Conditional Random Field component.

Method 1: Configure the component on the pipeline page

You can configure the parameters of the Conditional Random Field component on the pipeline page of Machine Learning Designer of Machine Learning Platform for AI (PAI). Machine Learning Designer is formerly known as Machine Learning Studio. The following table describes the parameters.

Tab	Parameter	Description
Fields Setting	ID Columns	The column that contains the ID of each sample. Samples are stored in n-tuples.
	Feature Columns	The word to be annotated and its features if the word has features.
	Target Columns	The column that you want to select.
Parameters Setting	Feature Generation Template	Default value: `[-2:0],[-1:0],[0:0],[1:0],[2:0],[-1:0]/[0:0], [0:0]/[1:0],[-2:1],[-1:1],[0:1],[1:1],[2:1], [-2:1]/[-1:1],[-1:1]/[0:1],[0:1]/[1:1],[1:1]/[2:1], [-2:1]/[-1:1]/[0:1],[-1:1]/[0:1]/[1:1], [0:1]/[1:1]/[2:1]` .
	Infrequently Used Word Filtering Threshold	Default value: 1.
	L1 Regularization Coefficient	Default value: 1.
	L2 Regularization Coefficient	Default value: 0.
	Maximum Iterations	Default value: 100.
	Convergence Threshold	Default value: 0.00001.
Tuning	Cores	The number of cores. By default, the system determines the value
Tuning	Memory Size per Core	The memory size of each core. By default, the system determines the value

Method 2: Use PAI commands

Configure the component parameters by using PAI commands. You can use the SQL Script component to call PAI commands. For more information, see SQL Script.

PAI -name=linearcrf    
    -project=algo_public    
    -DinputTableName=crf_input_table    
    -DidColName=sentence_id    
    -DfeatureColNames=word,f1    
    -DlabelColName=label    
    -DoutputTableName=crf_model    
    -Dlifecycle=28    
    -DcoreNum=10

Parameter	Required	Description	Default value
inputTableName	Yes	The table that contains the input features.	No default value
inputTablePartitions	No	The partitions selected from the table that contains the input features.	All partitions
featureColNames	No	The feature columns selected from the input table.	All columns, excluding the label column
labelColName	Yes	The column that you want to select.	No default value
idColName	Yes	The column that contains sample labels.	No default value
outputTableName	Yes	The table that contains output models.	No default value
outputTablePartitions	No	The partitions selected from the output model table.	All partitions
template	No	The template that is used to generate features.	Definition `<template .=. <template_item,<template_item,...,<template_item <template_item .=. [row_offset:col_index]/[row_offset:col_index]/.../[row_offset:col_index] row_offset .=. integer col_index .=. integer>` Default value `[-2:0],[-1:0],[0:0],[1:0],[2:0],[-1:0]/[0:0],[0:0]/[1:0],[-2:1],[-1:1],[0:1],[1:1],[2:1],[-2:1]/[-1:1],[-1:1]/[0:1],[0:1]/[1:1],[1:1]/[2:1],[-2:1]/[-1:1]/[0:1],[-1:1]/[0:1]/[1:1],[0:1]/[1:1]/[2:1]`
freq	No	The parameter for filtering features. Only feature values greater than or equal to the freq value are retained.	1
iterations	No	The maximum number of iterations of optimizations.	100
l1Weight	No	The parameter weight of L1 regularization.	1.0
l2Weight	No	The parameter weight of L2 regularization.	1.0
epsilon	No	The convergence deviation. This parameter specifies the requirement to finish the Limited-memory Broyden Fletcher Goldfarb Shanno (L-BFGS) process, which is the deviation between the log-likelihood values in two iterations.	0.0001
lbfgsStep	No	The historical step size for optimization that is performed by using the L-BFGS algorithm. Only the L-BFGS algorithm supports this parameter.	10
threadNum	No	The number of parallel threads used for model training.	3
lifecycle	No	The lifecycle of the output table.	No default value
coreNum	No	The number of cores.	Determined by the system
memSizePerCore	No	The memory size of each core.	Determined by the system

Example

Input data
sentence_id
word
f1
label
1
Rockwell
NNP
B-NP
1
International
NNP
I-NP
1
Corp
NNP
I-NP
1
's
POS
B-NP
...
...
...
...
823
Ohio
NNP
B-NP
823
grew
VBD
B-VP
823
3.8
CD
B-NP
823
%
NN
I-NP
823
.
.
O

Prediction algorithm PAI command

PAI -name=crf_predict    
    -project=algo_public    
    -DinputTableName=crf_test_input_table    
    -DmodelTableName=crf_model    
    -DidColName=sentence_id    
    -DfeatureColNames=word,f1    
    -DlabelColName=label    
    -DoutputTableName=crf_predict_result    
    -DdetailColName=prediction_detail   
    -Dlifecycle=28    
    -DcoreNum=10

Parameter	Required	Description	Default value
inputTableName	Yes	The table that contains the input features.	No default value
inputTablePartitions	No	The partitions selected from the table that contains the input features.	All partitions
featureColNames	No	The feature columns selected from the input table.	All columns, excluding the label column
labelColName	No	The column that you want to select.	No default value
IdColName	Yes	The column that contains sample labels.	No default value
resultColName	No	The result column in the output table.	prediction_result
scoreColName	No	The score column in the output table.	prediction_score
detailColName	No	The detail column in the output table.	No default value
outputTableName	Yes	The output prediction result table.	No default value
outputTablePartitions	No	The partitions selected from the output prediction result table.	All partitions
modelTableName	Yes	The algorithm model table.	No default value
modelTablePartitions	No	The partitions selected from the algorithm model table.	All partitions
lifecycle	No	The lifecycle of the output table.	No default value
coreNum	No	The number of cores.	Determined by the system
memSizePerCore	No	The memory size of each core.	Determined by the system

Output data
sentence_id
word
f1
label
1
Confidence
NN
B-NP
1
in
IN
B-PP
1
the
DT
B-NP
1
pound
NN
I-NP
...
...
...
...
77
have
VBP
B-VP
77
announced
VBN
I-VP
77
similar
JJ
B-NP
77
increases
NNS
I-NP
77
.
.
O
Note
The label column is optional.

sentence_id	word	f1	label
1	Rockwell	NNP	B-NP
1	International	NNP	I-NP
1	Corp	NNP	I-NP
1	's	POS	B-NP
...	...	...	...
823	Ohio	NNP	B-NP
823	grew	VBD	B-VP
823	3.8	CD	B-NP
823	%	NN	I-NP
823	.	.	O