Documentation
¶
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type AnalyzeIR ¶
type AnalyzeIR struct {
// DataSource contains the connection information. For example, "hive://root:root@localhost:10000/churn"
DataSource string
// Select specifies the query for fetching the analysis data. For example, "select * from iris.test;".
Select string
// Attributes contain a list of parsed attribute in the WITH clause. For example, after parsing
// "select ... analyze ... with analyze.plot_type = "bar"",
// the Attributes will be {{"analyze.plot_type", "bar"}}
Attributes []Attribute
// TrainIR is the TrainIR used for generating the training job of the corresponding model
TrainIR TrainIR
}
AnalyzeIR is the intermediate representation for code generation of a analysis job
type Attribute ¶
type Attribute struct {
Key string
Value interface{}
}
Attribute represents an parsed entry in the WITH clause.
type FeatureColumn ¶
type FeatureColumn interface{}
FeatureColumn indicates the feature column to be applied on the field. Please refer to github.com/sql-machine-learning/sqlflow/sql/codegen/feature_column.go for detailed list of all feature columns.
type FieldMeta ¶
type FieldMeta struct {
Name string `json:"name"` // e.g. "spetal_length"
DType FieldType `json:"dtype"` // e.g. "float", "int32"
Delimiter string `json:"delimiter"` // e.g. ","
Shape []int `json:"shape"` // e.g. [1], [1 2 3]
IsSparse bool `json:"is_sparse"` // e.g. false
}
FieldMeta contains the meta information for decoding. A field is a selected column of a SQL result.
Name indicates the name for a field.
DType indicates the data type for a field. For example: Int, Float, String.
Delimiter indicates the decoding method of a field. For example, the field may contain a string like "1,23,42" which represent a 3-D tensor [1, 23, 42].
Shape indicates the shape of the tensor represented for a field. For exmaple, the field may contain a string like "1,23,42" which represent a 3-D tensor, the shape will be [3].
IsSparse indicates the type of tensor for a field. True means the tensor is a sparse tensor.
type NumericColumn ¶
type NumericColumn struct {
FieldMeta *FieldMeta
}
NumericColumn represents a dense tensor for the model input.
FieldMeta indicates the meta information for decoding the field. Please be aware that FieldMeta also contains information for dimension and data type.
type PredictIR ¶
type PredictIR struct {
// DataSource contains the connection information. For example, "hive://root:root@localhost:10000/churn"
DataSource string
// Select specifies the query for fetching the prediction data. For example, "select * from iris.test;".
Select string
// ResultTable specifies the table to store the prediction result.
ResultTable string
// Attributes contain a list of parsed attribute in the WITH clause. For example, after parsing
// "select ... predict ... with predict.batch_size = 32 into ...",
// the Attributes will be {{"predict.batch_size", 32}}
Attributes []Attribute
// TrainIR is the TrainIR used for generating the training job of the corresponding model
TrainIR TrainIR
}
PredictIR is the intermediate representation for code generation of a prediction job
Please be aware the PredictionIR contains the result table name, so the generated Python program is responsible to create and write the result table.
type TrainIR ¶
type TrainIR struct {
// DataSource contains the connection information. For example, "hive://root:root@localhost:10000/churn"
DataSource string
// Select specifies the query for fetching the training data. For example, "select * from iris.train;".
Select string
// ValidationSelect specifies the query for fetching the validation data. For example, "select * from iris.val;".
ValidationSelect string
// Estimator specifies the estimator type. For example, after parsing "select ... train DNNClassifier WITH ...",
// the Estimator will be "DNNClassifier".
Estimator string
// Attributes contain a list of parsed attribute in the WITH Clause. For example, after parsing
// "select ... train ... with train.epoch = 1000, model.hidden_units = [10, 10]",
// the Attributes will be {{"train.epoch", 1000}, {"model.hidden_units", [10 10]}}.
Attributes []Attribute
// Features contain a map of a list of feature columns in the COLUMN clause.
// For multiple COLUMN clauses like
// “`
// column ... for deep_feature
// column ... for wide_feature
// “`
// They will be parsed as {"deep_feature": {...}, "wide_feature": {...}}
// For single column clause like "column ...", "feature_columns" will be used as the default map key.
Features map[string][]FeatureColumn
// Label specifies the feature column in the LABEL clause.
Label FeatureColumn
}
TrainIR is the intermediate representation for code generation of a training job.
Please be aware that the TrainIR intentionally excludes the model table name in the INTO clause. The sql package will save the output files of a generated Python program. For prediction and analysis jobs, the sql will restore an identical working directly.
Source Files