fyp/logic/models/run.go

package models

import (
	"bytes"
	"fmt"
	"io"
	"net/http"
	"os"
	"path"

	. "git.andr3h3nriqu3s.com/andr3/fyp/logic/models/utils"
	. "git.andr3h3nriqu3s.com/andr3/fyp/logic/utils"

	tf "github.com/galeone/tensorflow/tensorflow/go"
	"github.com/galeone/tensorflow/tensorflow/go/op"
	tg "github.com/galeone/tfgo"
	"github.com/galeone/tfgo/image"
)

func ReadPNG(scope *op.Scope, imagePath string, channels int64) *image.Image {
	scope = tg.NewScope(scope)
	contents := op.ReadFile(scope.SubScope("ReadFile"), op.Const(scope.SubScope("filename"), imagePath))
	output := op.DecodePng(scope.SubScope("DecodePng"), contents, op.DecodePngChannels(channels))
	output = op.ExpandDims(scope.SubScope("ExpandDims"), output, op.Const(scope.SubScope("axis"), []int32{0}))
	image := &image.Image{
		Tensor: tg.NewTensor(scope, output)}
	return image.Scale(0, 255)
}

func ReadJPG(scope *op.Scope, imagePath string, channels int64) *image.Image {
	scope = tg.NewScope(scope)
	contents := op.ReadFile(scope.SubScope("ReadFile"), op.Const(scope.SubScope("filename"), imagePath))
	output := op.DecodePng(scope.SubScope("DecodeJpeg"), contents, op.DecodePngChannels(channels))
	output = op.ExpandDims(scope.SubScope("ExpandDims"), output, op.Const(scope.SubScope("axis"), []int32{0}))
	image := &image.Image{
		Tensor: tg.NewTensor(scope, output)}
	return image.Scale(0, 255)
}

func runModelNormal(c *Context, model *BaseModel, def_id string, inputImage *tf.Tensor) (order int, err error) {
	order = 0
	err = nil

	tf_model := tg.LoadModel(path.Join("savedData", model.Id, "defs", def_id, "model"), []string{"serve"}, nil)

	results := tf_model.Exec([]tf.Output{
		tf_model.Op("StatefulPartitionedCall", 0),
	}, map[tf.Output]*tf.Tensor{
		tf_model.Op("serving_default_rescaling_input", 0): inputImage,
	})

	var vmax float32 = 0.0
	var predictions = results[0].Value().([][]float32)[0]

	for i, v := range predictions {
		if v > vmax {
			order = i
			vmax = v
		}
	}

	return
}

func runModelExp(c *Context, model *BaseModel, def_id string, inputImage *tf.Tensor) (order int, err error) {

	err = nil
	order = 0

	base_model := tg.LoadModel(path.Join("savedData", model.Id, "defs", def_id, "base", "model"), []string{"serve"}, nil)

	//results := base_model.Exec([]tf.Output{
	base_results := base_model.Exec([]tf.Output{
		base_model.Op("StatefulPartitionedCall", 0),
	}, map[tf.Output]*tf.Tensor{
		//base_model.Op("serving_default_rescaling_input", 0): inputImage,
		base_model.Op("serving_default_input_1", 0): inputImage,
	})

	type head struct {
		Id          string
		Range_start int
	}

	heads, err := GetDbMultitple[head](c, "exp_model_head where def_id=$1;", def_id)
	if err != nil {
		return
	}

	var vmax float32 = 0.0

	for _, element := range heads {
		head_model := tg.LoadModel(path.Join("savedData", model.Id, "defs", def_id, "head", element.Id, "model"), []string{"serve"}, nil)

		results := head_model.Exec([]tf.Output{
			head_model.Op("StatefulPartitionedCall", 0),
		}, map[tf.Output]*tf.Tensor{
			head_model.Op("serving_default_input_2", 0): base_results[0],
		})

		var predictions = results[0].Value().([][]float32)[0]

		for i, v := range predictions {
			if v > vmax {
				order = element.Range_start + i
				vmax = v
			}
		}
	}

	// TODO runthe head model

	c.Logger.Info("Got", "heads", len(heads))
	return
}

func handleRun(handle *Handle) {
	handle.Post("/models/run", func(w http.ResponseWriter, r *http.Request, c *Context) *Error {
		if !CheckAuthLevel(1, w, r, c) {
			return nil
		}
		if c.Mode == JSON {

			read_form, err := r.MultipartReader()
			if err != nil {
				// TODO improve message
				return ErrorCode(nil, 400, nil)
			}

			var id string
			var file []byte

			for {
				part, err_part := read_form.NextPart()
				if err_part == io.EOF {
					break
				} else if err_part != nil {
                    return c.JsonBadRequest("Invalid multipart data")
				}
				if part.FormName() == "id" {
					buf := new(bytes.Buffer)
					buf.ReadFrom(part)
					id = buf.String()
				}
				if part.FormName() == "file" {
					buf := new(bytes.Buffer)
					buf.ReadFrom(part)
					file = buf.Bytes()
				}
			}

			model, err := GetBaseModel(handle.Db, id)
			if err == ModelNotFoundError {
                return c.JsonBadRequest("Models not found");
			} else if err != nil {
				return c.Error500(err)
			}

			if model.Status != READY {
                return c.JsonBadRequest("Model not ready to run images")
			}

			def := JustId{}
			err = GetDBOnce(c, &def, "model_definition where model_id=$1", model.Id)
			if err == NotFoundError {
				return c.JsonBadRequest("Could not find definition")
			} else if err != nil {
				return c.Error500(err)
			}

			def_id := def.Id

			// TODO create a database table with tasks
			run_path := path.Join("/tmp", model.Id, "runs")
			os.MkdirAll(run_path, os.ModePerm)
			img_path := path.Join(run_path, "img."+model.Format)

			img_file, err := os.Create(img_path)
			if err != nil {
				return c.Error500(err)
			}
			defer img_file.Close()
			img_file.Write(file)

			if !testImgForModel(c, model, img_path) {
                return c.JsonBadRequest("Provided image does not match the model")
			}

			root := tg.NewRoot()

			var tf_img *image.Image = nil

			switch model.Format {
			case "png":
				tf_img = ReadPNG(root, img_path, int64(model.ImageMode))
			case "jpeg":
				tf_img = ReadJPG(root, img_path, int64(model.ImageMode))
			default:
				panic("Not sure what to do with '" + model.Format + "'")
			}

			exec_results := tg.Exec(root, []tf.Output{tf_img.Value()}, nil, &tf.SessionOptions{})
			inputImage, err := tf.NewTensor(exec_results[0].Value())
			if err != nil {
				return c.Error500(err)
			}

			vi := -1

			if model.ModelType == 2 {
				c.Logger.Info("Running model normal", "model", model.Id, "def", def_id)
				vi, err = runModelExp(c, model, def_id, inputImage)
				if err != nil {
					return c.Error500(err)
				}
			} else {
				c.Logger.Info("Running model normal", "model", model.Id, "def", def_id)
				vi, err = runModelNormal(c, model, def_id, inputImage)
				if err != nil {
					return c.Error500(err)
				}
			}

			os.RemoveAll(run_path)

			rows, err := handle.Db.Query("select name from model_classes where model_id=$1 and class_order=$2;", model.Id, vi)
			if err != nil {
				return c.Error500(err)
			}
			if !rows.Next() {
                return c.SendJSON(nil)
			}

			var name string
			if err = rows.Scan(&name); err != nil {
				return c.Error500(err)
			}

            return c.SendJSON(name)
		}

		read_form, err := r.MultipartReader()
		if err != nil {
			// TODO improve message
			return ErrorCode(nil, 400, nil)
		}

		var id string
		var file []byte

		for {
			part, err_part := read_form.NextPart()
			if err_part == io.EOF {
				break
			} else if err_part != nil {
				return &Error{Code: http.StatusBadRequest}
			}
			if part.FormName() == "id" {
				buf := new(bytes.Buffer)
				buf.ReadFrom(part)
				id = buf.String()
			}
			if part.FormName() == "file" {
				buf := new(bytes.Buffer)
				buf.ReadFrom(part)
				file = buf.Bytes()
			}
		}

		model, err := GetBaseModel(handle.Db, id)
		if err == ModelNotFoundError {
			return ErrorCode(nil, http.StatusNotFound, AnyMap{
				"NotFoundMessage": "Model not found",
				"GoBackLink":      "/models",
			})
		} else if err != nil {
			return Error500(err)
		}

		if model.Status != READY {
			// TODO improve this
			return ErrorCode(nil, 400, c.AddMap(nil))
		}

		def := JustId{}
		err = GetDBOnce(c, &def, "model_definition where model_id=$1", model.Id)
		if err == NotFoundError {
			// TODO improve this
			fmt.Printf("Could not find definition\n")
			return ErrorCode(nil, 400, c.AddMap(nil))
		} else if err != nil {
			return Error500(err)
		}

		def_id := def.Id

		// TODO create a database table with tasks
		run_path := path.Join("/tmp", model.Id, "runs")
		os.MkdirAll(run_path, os.ModePerm)
		img_path := path.Join(run_path, "img."+model.Format)

		img_file, err := os.Create(img_path)
		if err != nil {
			return Error500(err)
		}
		defer img_file.Close()
		img_file.Write(file)

		if !testImgForModel(c, model, img_path) {
			LoadDefineTemplate(w, "/models/edit.html", "run-model-card", c.AddMap(AnyMap{
				"Model":      model,
				"NotFound":   false,
				"Result":     nil,
				"ImageError": true,
			}))
			return nil
		}

		root := tg.NewRoot()

		var tf_img *image.Image = nil

		switch model.Format {
		case "png":
			tf_img = ReadPNG(root, img_path, int64(model.ImageMode))
		case "jpeg":
			tf_img = ReadJPG(root, img_path, int64(model.ImageMode))
		default:
			panic("Not sure what to do with '" + model.Format + "'")
		}

		exec_results := tg.Exec(root, []tf.Output{tf_img.Value()}, nil, &tf.SessionOptions{})
		inputImage, err := tf.NewTensor(exec_results[0].Value())
		if err != nil {
			return Error500(err)
		}

		vi := -1

		if model.ModelType == 2 {
			c.Logger.Info("Running model normal", "model", model.Id, "def", def_id)
			vi, err = runModelExp(c, model, def_id, inputImage)
			if err != nil {
				return c.Error500(err)
			}
		} else {
			c.Logger.Info("Running model normal", "model", model.Id, "def", def_id)
			vi, err = runModelNormal(c, model, def_id, inputImage)
			if err != nil {
				return c.Error500(err)
			}
		}

		os.RemoveAll(run_path)

		rows, err := handle.Db.Query("select name from model_classes where model_id=$1 and class_order=$2;", model.Id, vi)
		if err != nil {
			return Error500(err)
		}
		if !rows.Next() {
			LoadDefineTemplate(w, "/models/edit.html", "run-model-card", c.AddMap(AnyMap{
				"Model":    model,
				"NotFound": true,
				"Result":   nil,
			}))
			return nil
		}

		var name string
		if err = rows.Scan(&name); err != nil {
			return nil
		}

		LoadDefineTemplate(w, "/models/edit.html", "run-model-card", c.AddMap(AnyMap{
			"Model":  model,
			"Result": name,
		}))
		return nil
	})
}