func ExtractParamAndConvertToFloat(params data.Map, key string) (float64, error) {
v, ok := params[key]
if !ok {
return 0, fmt.Errorf("%s parameter is missing", key)
}
x, err := data.ToFloat(v)
if err != nil {
return 0, fmt.Errorf("%s parameter cannot be converted to float: %v", key, err)
}
return x, nil
}
func (pa *PassiveAggressiveState) Write(ctx *core.Context, t *core.Tuple) error {
vval, ok := t.Data[pa.valueField]
if !ok {
return fmt.Errorf("%s field is missing", pa.valueField)
}
val64, err := data.ToFloat(vval)
if err != nil {
return fmt.Errorf("%s cannot be converted to float: %v", pa.valueField, err)
}
val := float32(val64)
vfv, ok := t.Data[pa.featureVectorField]
if !ok {
return fmt.Errorf("%s field is missing", pa.featureVectorField)
}
fv, err := data.AsMap(vfv)
if err != nil {
return fmt.Errorf("%s value is not a map: %v", pa.featureVectorField, err)
}
err = pa.pa.Train(FeatureVector(fv), val)
return err
}
func flattenImpl(keyPrefix string, v data.Value, ap Appender) error {
switch v.Type() {
case data.TypeArray:
keyPrefix += "\x00"
a, _ := data.AsArray(v)
for i, v := range a {
err := flattenImpl(fmt.Sprintf(keyPrefix, i), v, ap)
if err != nil {
return err
}
}
case data.TypeMap:
if keyPrefix != "" {
keyPrefix += "\x00"
}
m, _ := data.AsMap(v)
for f, v := range m {
err := flattenImpl(keyPrefix+f, v, ap)
if err != nil {
return err
}
}
default:
xx, err := data.ToFloat(v)
if err != nil {
// TODO: return better error
return err
}
x := float32(xx)
ap(keyPrefix, x)
}
return nil
}
func (r *source) GenerateStream(ctx *core.Context, w core.Writer) error {
numFieldNames := []string{
"家賃(万円)",
"駅からの徒歩時間 (分)",
"専有面積 (m*m)",
"築年数 (年)",
"階数",
}
defer r.file.Close()
if r.training {
for {
line, err := r.readLine()
if err != nil {
if err == io.EOF {
return nil
}
return err
}
if line[0] == '#' {
continue
}
fields := strings.Split(line, ", ")
if len(fields) != 6 {
panic("hoge")
}
value, err := data.ToFloat(data.String(fields[0]))
if err != nil {
panic(err)
}
fv := make(data.Map)
for i := 1; i < len(numFieldNames); i++ {
x, err := data.ToFloat(data.String(fields[i]))
if err != nil {
panic(err)
}
fv[numFieldNames[i]] = data.Float(x)
}
fv[fields[len(fields)-1]] = data.Float(1)
now := time.Now()
w.Write(ctx, &core.Tuple{
Data: data.Map{
"value": data.Float(value),
"feature_vector": fv,
},
Timestamp: now,
ProcTimestamp: now,
})
}
} else {
fv := make(data.Map)
i := 1
for {
line, err := r.readLine()
if err != nil {
if err == io.EOF {
return nil
}
return err
}
if line == "" || line[0] == '#' {
continue
}
fields := strings.Split(line, ":")
if len(fields) != 2 {
panic("hoge")
}
for i := range fields {
fields[i] = strings.TrimSpace(fields[i])
}
if i < len(numFieldNames) {
x, err := data.ToFloat(data.String(fields[1]))
if err != nil {
panic(err)
}
fv[numFieldNames[i]] = data.Float(x)
i++
} else {
if fields[0] != "aspect" {
panic(fields)
}
aspect := strings.Trim(fields[1], "\"")
fv[aspect] = data.Float(1)
break
}
}
now := time.Now()
w.Write(ctx, &core.Tuple{
Data: data.Map{
"feature_vector": fv,
},
Timestamp: now,
ProcTimestamp: now,
})
}
return nil
}
