started working on runner

This commit is contained in:
Andre Henriques 2024-05-06 01:10:58 +01:00
parent 7d346ba2ce
commit e22df8adc9
38 changed files with 4134 additions and 137 deletions

37
DockerfileServer Normal file
View File

@ -0,0 +1,37 @@
FROM docker.io/nvidia/cuda:12.3.2-devel-ubuntu22.04
ENV DEBIAN_FRONTEND=noninteractive
# Sometimes you have to get update twice because ?
RUN apt-get update
RUN apt-get update
RUN apt-get install -y wget unzip python3-pip vim python3 python3-pip curl
RUN wget https://go.dev/dl/go1.22.2.linux-amd64.tar.gz
RUN tar -xvf go1.22.2.linux-amd64.tar.gz -C /usr/local
ENV PATH=$PATH:/usr/local/go/bin
ENV GOPATH=/go
RUN bash -c 'curl -L "https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-cpu-linux-x86_64-2.9.1.tar.gz" | tar -C /usr/local -xz'
# RUN bash -c 'curl -L "https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-cpu-linux-x86_64-2.13.1.tar.gz" | tar -C /usr/local -xz'
RUN ldconfig
RUN ln -s /usr/bin/python3 /usr/bin/python
RUN python -m pip install nvidia-pyindex
ADD requirements.txt .
RUN python -m pip install -r requirements.txt
ENV CUDNN_PATH=/usr/local/lib/python3.10/dist-packages/nvidia/cudnn
ENV LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib/python3.10/dist-packages/nvidia/cudnn/lib
WORKDIR /app
ADD go.mod .
ADD go.sum .
ADD main.go .
ADD logic logic
RUN go install || true
CMD ["go", "run", "."]

View File

@ -6,3 +6,19 @@ const (
DATA_POINT_MODE_TRAINING DATA_POINT_MODE = 1
DATA_POINT_MODE_TESTING = 2
)
type ModelClassStatus int
const (
CLASS_STATUS_TO_TRAIN ModelClassStatus = iota + 1
CLASS_STATUS_TRAINING
CLASS_STATUS_TRAINED
)
type ModelClass struct {
Id string `db:"mc.id" json:"id"`
ModelId string `db:"mc.model_id" json:"model_id"`
Name string `db:"mc.name" json:"name"`
ClassOrder int `db:"mc.class_order" json:"class_order"`
Status int `db:"mc.status" json:"status"`
}

View File

@ -0,0 +1,95 @@
package dbtypes
import (
"time"
"git.andr3h3nriqu3s.com/andr3/fyp/logic/db"
)
type DefinitionStatus int
const (
DEFINITION_STATUS_CANCELD_TRAINING DefinitionStatus = -4
DEFINITION_STATUS_FAILED_TRAINING = -3
DEFINITION_STATUS_PRE_INIT = 1
DEFINITION_STATUS_INIT = 2
DEFINITION_STATUS_TRAINING = 3
DEFINITION_STATUS_PAUSED_TRAINING = 6
DEFINITION_STATUS_TRANIED = 4
DEFINITION_STATUS_READY = 5
)
type Definition struct {
Id string `db:"md.id" json:"id"`
ModelId string `db:"md.model_id" json:"model_id"`
Accuracy float64 `db:"md.accuracy" json:"accuracy"`
TargetAccuracy int `db:"md.target_accuracy" json:"target_accuracy"`
Epoch int `db:"md.epoch" json:"epoch"`
Status int `db:"md.status" json:"status"`
CreatedOn time.Time `db:"md.created_on" json:"created"`
EpochProgress int `db:"md.epoch_progress" json:"epoch_progress"`
}
type SortByAccuracyDefinitions []*Definition
func (nf SortByAccuracyDefinitions) Len() int { return len(nf) }
func (nf SortByAccuracyDefinitions) Swap(i, j int) { nf[i], nf[j] = nf[j], nf[i] }
func (nf SortByAccuracyDefinitions) Less(i, j int) bool {
return nf[i].Accuracy < nf[j].Accuracy
}
func GetDefinition(db db.Db, definition_id string) (definition Definition, err error) {
err = GetDBOnce(db, &definition, "model_definition as md where id=$1;", definition_id)
return
}
func MakeDefenition(db db.Db, model_id string, target_accuracy int) (definition Definition, err error) {
var NewDefinition = struct {
ModelId string `db:"model_id"`
TargetAccuracy int `db:"target_accuracy"`
}{ModelId: model_id, TargetAccuracy: target_accuracy}
id, err := InsertReturnId(db, &NewDefinition, "model_definition", "id")
if err != nil {
return
}
return GetDefinition(db, id)
}
func (d Definition) UpdateStatus(db db.Db, status DefinitionStatus) (err error) {
_, err = db.Exec("update model_definition set status=$1 where id=$2", status, d.Id)
return
}
func (d Definition) MakeLayer(db db.Db, layer_order int, layer_type LayerType, shape string) (layer Layer, err error) {
var NewLayer = struct {
DefinitionId string `db:"def_id"`
LayerOrder int `db:"layer_order"`
LayerType LayerType `db:"layer_type"`
Shape string `db:"shape"`
}{
DefinitionId: d.Id,
LayerOrder: layer_order,
LayerType: layer_type,
Shape: shape,
}
id, err := InsertReturnId(db, &NewLayer, "model_definition_layer", "id")
if err != nil {
return
}
return GetLayer(db, id)
}
func (d Definition) GetLayers(db db.Db, filter string, args ...any) (layer []*Layer, err error) {
args = append(args, d.Id)
return GetDbMultitple[Layer](db, "model_definition_layer as mdl where mdl.def_id=$1 "+filter, args...)
}
func (d *Definition) UpdateAfterEpoch(db db.Db, accuracy float64) (err error) {
d.Accuracy = accuracy
d.Epoch += 1
_, err = db.Exec("update model_definition set epoch=$1, accuracy=$2 where id=$3", d.Epoch, d.Accuracy, d.Id)
return
}

50
logic/db_types/layer.go Normal file
View File

@ -0,0 +1,50 @@
package dbtypes
import (
"encoding/json"
"git.andr3h3nriqu3s.com/andr3/fyp/logic/db"
)
type LayerType int
const (
LAYER_INPUT LayerType = 1
LAYER_DENSE = 2
LAYER_FLATTEN = 3
LAYER_SIMPLE_BLOCK = 4
)
type Layer struct {
Id string `db:"mdl.id" json:"id"`
DefinitionId string `db:"mdl.def_id" json:"definition_id"`
LayerOrder string `db:"mdl.layer_order" json:"layer_order"`
LayerType LayerType `db:"mdl.layer_type" json:"layer_type"`
Shape string `db:"mdl.shape" json:"shape"`
ExpType string `db:"mdl.exp_type" json:"exp_type"`
}
func ShapeToString(args ...int) string {
text, err := json.Marshal(args)
if err != nil {
panic("Could not generate Shape")
}
return string(text)
}
func StringToShape(str string) (shape []int64) {
err := json.Unmarshal([]byte(str), &shape)
if err != nil {
panic("Could not parse Shape")
}
return
}
func (l Layer) GetShape() []int64 {
return StringToShape(l.Shape)
}
func GetLayer(db db.Db, layer_id string) (layer Layer, err error) {
err = GetDBOnce(db, &layer, "model_definition_layer as mdl where mdl.id=$1", layer_id)
return
}

View File

@ -2,23 +2,26 @@ package dbtypes
import (
"errors"
"fmt"
"path"
"git.andr3h3nriqu3s.com/andr3/fyp/logic/db"
)
const (
FAILED_TRAINING = -4
FAILED_PREPARING_TRAINING = -3
FAILED_PREPARING_ZIP_FILE = -2
FAILED_PREPARING = -1
type ModelStatus int
PREPARING = 1
CONFIRM_PRE_TRAINING = 2
PREPARING_ZIP_FILE = 3
TRAINING = 4
READY = 5
READY_ALTERATION = 6
READY_ALTERATION_FAILED = -6
const (
FAILED_TRAINING ModelStatus = -4
FAILED_PREPARING_TRAINING = -3
FAILED_PREPARING_ZIP_FILE = -2
FAILED_PREPARING = -1
PREPARING = 1
CONFIRM_PRE_TRAINING = 2
PREPARING_ZIP_FILE = 3
TRAINING = 4
READY = 5
READY_ALTERATION = 6
READY_ALTERATION_FAILED = -6
READY_RETRAIN = 7
READY_RETRAIN_FAILED = -7
@ -26,15 +29,6 @@ const (
type ModelDefinitionStatus int
type LayerType int
const (
LAYER_INPUT LayerType = 1
LAYER_DENSE = 2
LAYER_FLATTEN = 3
LAYER_SIMPLE_BLOCK = 4
)
const (
MODEL_DEFINITION_STATUS_CANCELD_TRAINING ModelDefinitionStatus = -4
MODEL_DEFINITION_STATUS_FAILED_TRAINING = -3
@ -46,14 +40,6 @@ const (
MODEL_DEFINITION_STATUS_READY = 5
)
type ModelClassStatus int
const (
MODEL_CLASS_STATUS_TO_TRAIN ModelClassStatus = 1
MODEL_CLASS_STATUS_TRAINING = 2
MODEL_CLASS_STATUS_TRAINED = 3
)
type ModelHeadStatus int
const (
@ -97,6 +83,61 @@ func (m BaseModel) CanEval() bool {
return true
}
// DO NOT Pass un filtered data on filters
func (m BaseModel) GetDefinitions(db db.Db, filters string, args ...any) ([]*Definition, error) {
n_args := []any{m.Id}
n_args = append(n_args, args...)
return GetDbMultitple[Definition](db, fmt.Sprintf("model_definition as md where md.model_id=$1 %s", filters), n_args...)
}
func (m BaseModel) GetClasses(db db.Db, filters string, args ...any) ([]*ModelClass, error) {
n_args := []any{m.Id}
n_args = append(n_args, args...)
return GetDbMultitple[ModelClass](db, fmt.Sprintf("model_classes as mc where mc.model_id=$1 %s", filters), n_args...)
}
func (m *BaseModel) UpdateStatus(db db.Db, status ModelStatus) (err error) {
_, err = db.Exec("update models set status=$1 where id=$2", status, m.Id)
return
}
type DataPoint struct {
Class int `json:"class"`
Path string `json:"path"`
}
func (m BaseModel) DataPoints(db db.Db, mode DATA_POINT_MODE) (data []DataPoint, err error) {
rows, err := db.Query(
"select mdp.id, mc.class_order, mdp.file_path from model_data_point as mdp inner "+
"join model_classes as mc on mc.id = mdp.class_id "+
"where mc.model_id = $1 and mdp.model_mode=$2;",
m.Id, mode)
if err != nil {
return
}
defer rows.Close()
data = []DataPoint{}
for rows.Next() {
var id string
var class_order int
var file_path string
if err = rows.Scan(&id, &class_order, &file_path); err != nil {
return
}
if file_path == "id://" {
data = append(data, DataPoint{
Path: path.Join("./savedData", m.Id, "data", id+"."+m.Format),
Class: class_order,
})
} else {
panic("TODO remote file path")
}
}
return
}
func StringToImageMode(colorMode string) int {
switch colorMode {
case "greyscale":

View File

@ -7,15 +7,15 @@ import (
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/db_types"
)
type ModelClass struct {
type ModelClassJSON struct {
Id string `json:"id"`
ModelId string `json:"model_id" db:"model_id"`
Name string `json:"name"`
Status int `json:"status"`
}
func ListClasses(c BasePack, model_id string) (cls []*ModelClass, err error) {
return GetDbMultitple[ModelClass](c.GetDb(), "model_classes where model_id=$1", model_id)
func ListClasses(c BasePack, model_id string) (cls []*ModelClassJSON, err error) {
return GetDbMultitple[ModelClassJSON](c.GetDb(), "model_classes where model_id=$1", model_id)
}
func ModelHasDataPoints(db db.Db, model_id string) (result bool, err error) {

View File

@ -495,7 +495,7 @@ func handleDataUpload(handle *Handle) {
return c.E500M("Could not create class", err)
}
var modelClass model_classes.ModelClass
var modelClass model_classes.ModelClassJSON
err = GetDBOnce(c, &modelClass, "model_classes where id=$1;", id)
if err != nil {
return c.E500M("Failed to get class information but class was creted", err)
@ -704,7 +704,7 @@ func handleDataUpload(handle *Handle) {
return c.Error500(err)
}
} else {
_, err = handle.Db.Exec("delete from model_classes where model_id=$1 and status=$2;", model.Id, MODEL_CLASS_STATUS_TO_TRAIN)
_, err = handle.Db.Exec("delete from model_classes where model_id=$1 and status=$2;", model.Id, CLASS_STATUS_TO_TRAIN)
if err != nil {
return c.Error500(err)
}

View File

@ -51,7 +51,7 @@ func handleDelete(handle *Handle) {
return c.E500M("Faield to get model", err)
}
switch model.Status {
switch ModelStatus(model.Status) {
case FAILED_TRAINING:
fallthrough
case FAILED_PREPARING_ZIP_FILE:

View File

@ -35,9 +35,9 @@ func handleEdit(handle *Handle) {
}
type ReturnType struct {
Classes []*model_classes.ModelClass `json:"classes"`
HasData bool `json:"has_data"`
NumberOfInvalidImages int `json:"number_of_invalid_images"`
Classes []*model_classes.ModelClassJSON `json:"classes"`
HasData bool `json:"has_data"`
NumberOfInvalidImages int `json:"number_of_invalid_images"`
}
c.ShowMessage = false

View File

@ -0,0 +1,79 @@
package models_train
import (
"errors"
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/db_types"
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/tasks/utils"
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/utils"
"github.com/goccy/go-json"
)
func PrepareTraining(handler *Handle, b BasePack, task Task, runner_id string) (err error) {
l := b.GetLogger()
model, err := GetBaseModel(b.GetDb(), *task.ModelId)
if err != nil {
task.UpdateStatusLog(b, TASK_FAILED_RUNNING, "Failed to get model information")
l.Error("Failed to get model information", "err", err)
return err
}
if model.Status != TRAINING {
task.UpdateStatusLog(b, TASK_FAILED_RUNNING, "Model not in the correct status for training")
return errors.New("Model not in the right status")
}
// TODO do this when the runner says it's OK
//task.UpdateStatusLog(b, TASK_RUNNING, "Training model")
// TODO move this to the runner part as well
var dat struct {
NumberOfModels int
Accuracy int
}
err = json.Unmarshal([]byte(task.ExtraTaskInfo), &dat)
if err != nil {
task.UpdateStatusLog(b, TASK_FAILED_RUNNING, "Failed to get model extra information")
}
if model.ModelType == 2 {
panic("TODO")
full_error := generateExpandableDefinitions(b, model, dat.Accuracy, dat.NumberOfModels)
if full_error != nil {
l.Error("Failed to generate defintions", "err", full_error)
task.UpdateStatusLog(b, TASK_FAILED_RUNNING, "Failed generate model")
return errors.New("Failed to generate definitions")
}
} else {
error := generateDefinitions(b, model, dat.Accuracy, dat.NumberOfModels)
if error != nil {
task.UpdateStatusLog(b, TASK_FAILED_RUNNING, "Failed generate model")
return errors.New("Failed to generate definitions")
}
}
runners := handler.DataMap["runners"].(map[string]interface{})
runner := runners[runner_id].(map[string]interface{})
runner["task"] = &task
runners[runner_id] = runner
handler.DataMap["runners"] = runners
return
}
func CleanUpFailed(b BasePack, task *Task) {
db := b.GetDb()
l := b.GetLogger()
model, err := GetBaseModel(db, *task.ModelId)
if err != nil {
l.Error("Failed to get model", "err", err)
} else {
err = model.UpdateStatus(db, FAILED_TRAINING)
if err != nil {
l.Error("Failed to get status", err)
}
}
}

View File

@ -17,7 +17,7 @@ func handleRest(handle *Handle) {
return c.E500M("Failed to get model", err)
}
if model.Status != FAILED_PREPARING_TRAINING && model.Status != FAILED_TRAINING {
if model.Status != FAILED_PREPARING_TRAINING && model.Status != int(FAILED_TRAINING) {
return c.JsonBadRequest("Model is not in status that be reset")
}

View File

@ -39,16 +39,6 @@ func getDir() string {
return dir
}
// This function creates a new model_definition
func MakeDefenition(db db.Db, model_id string, target_accuracy int) (id string, err error) {
var NewDefinition = struct {
ModelId string `db:"model_id"`
TargetAccuracy int `db:"target_accuracy"`
}{ModelId: model_id, TargetAccuracy: target_accuracy}
return InsertReturnId(db, &NewDefinition, "model_definition", "id")
}
func ModelDefinitionUpdateStatus(c BasePack, id string, status ModelDefinitionStatus) (err error) {
_, err = c.GetDb().Exec("update model_definition set status = $1 where id = $2", status, id)
return
@ -118,14 +108,14 @@ func generateCvsExp(c BasePack, run_path string, model_id string, doPanic bool)
var co struct {
Count int `db:"count(*)"`
}
err = GetDBOnce(db, &co, "model_classes where model_id=$1 and status=$2;", model_id, MODEL_CLASS_STATUS_TRAINING)
err = GetDBOnce(db, &co, "model_classes where model_id=$1 and status=$2;", model_id, CLASS_STATUS_TRAINING)
if err != nil {
return
}
count = co.Count
if count == 0 {
err = setModelClassStatus(c, MODEL_CLASS_STATUS_TRAINING, "model_id=$1 and status=$2;", model_id, MODEL_CLASS_STATUS_TO_TRAIN)
err = setModelClassStatus(c, CLASS_STATUS_TRAINING, "model_id=$1 and status=$2;", model_id, CLASS_STATUS_TO_TRAIN)
if err != nil {
return
}
@ -137,7 +127,7 @@ func generateCvsExp(c BasePack, run_path string, model_id string, doPanic bool)
return generateCvsExp(c, run_path, model_id, true)
}
data, err := db.Query("select mdp.id, mc.class_order, mdp.file_path from model_data_point as mdp inner join model_classes as mc on mc.id = mdp.class_id where mc.model_id = $1 and mdp.model_mode=$2 and mc.status=$3;", model_id, DATA_POINT_MODE_TRAINING, MODEL_CLASS_STATUS_TRAINING)
data, err := db.Query("select mdp.id, mc.class_order, mdp.file_path from model_data_point as mdp inner join model_classes as mc on mc.id = mdp.class_id where mc.model_id = $1 and mdp.model_mode=$2 and mc.status=$3;", model_id, DATA_POINT_MODE_TRAINING, CLASS_STATUS_TRAINING)
if err != nil {
return
}
@ -287,7 +277,7 @@ func generateCvsExpandExp(c BasePack, run_path string, model_id string, offset i
var co struct {
Count int `db:"count(*)"`
}
err = GetDBOnce(db, &co, "model_classes where model_id=$1 and status=$2;", model_id, MODEL_CLASS_STATUS_TRAINING)
err = GetDBOnce(db, &co, "model_classes where model_id=$1 and status=$2;", model_id, CLASS_STATUS_TRAINING)
if err != nil {
return
}
@ -296,7 +286,7 @@ func generateCvsExpandExp(c BasePack, run_path string, model_id string, offset i
count := co.Count
if count == 0 {
err = setModelClassStatus(c, MODEL_CLASS_STATUS_TRAINING, "model_id=$1 and status=$2;", model_id, MODEL_CLASS_STATUS_TO_TRAIN)
err = setModelClassStatus(c, CLASS_STATUS_TRAINING, "model_id=$1 and status=$2;", model_id, CLASS_STATUS_TO_TRAIN)
if err != nil {
return
} else if doPanic {
@ -305,7 +295,7 @@ func generateCvsExpandExp(c BasePack, run_path string, model_id string, offset i
return generateCvsExpandExp(c, run_path, model_id, offset, true)
}
data, err := db.Query("select mdp.id, mc.class_order, mdp.file_path from model_data_point as mdp inner join model_classes as mc on mc.id = mdp.class_id where mc.model_id = $1 and mdp.model_mode=$2 and mc.status=$3;", model_id, DATA_POINT_MODE_TRAINING, MODEL_CLASS_STATUS_TRAINING)
data, err := db.Query("select mdp.id, mc.class_order, mdp.file_path from model_data_point as mdp inner join model_classes as mc on mc.id = mdp.class_id where mc.model_id = $1 and mdp.model_mode=$2 and mc.status=$3;", model_id, DATA_POINT_MODE_TRAINING, CLASS_STATUS_TRAINING)
if err != nil {
return
}
@ -339,7 +329,7 @@ func generateCvsExpandExp(c BasePack, run_path string, model_id string, offset i
// This is to load some extra data so that the model has more things to train on
//
data_other, err := db.Query("select mdp.id, mc.class_order, mdp.file_path from model_data_point as mdp inner join model_classes as mc on mc.id = mdp.class_id where mc.model_id = $1 and mdp.model_mode=$2 and mc.status=$3 limit $4;", model_id, DATA_POINT_MODE_TRAINING, MODEL_CLASS_STATUS_TRAINED, count*10)
data_other, err := db.Query("select mdp.id, mc.class_order, mdp.file_path from model_data_point as mdp inner join model_classes as mc on mc.id = mdp.class_id where mc.model_id = $1 and mdp.model_mode=$2 and mc.status=$3 limit $4;", model_id, DATA_POINT_MODE_TRAINING, CLASS_STATUS_TRAINED, count*10)
if err != nil {
return
}
@ -737,7 +727,7 @@ func trainModel(c BasePack, model *BaseModel) (err error) {
if err != nil {
l.Error("Failed to train Model! Err:")
l.Error(err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return
}
defer definitionsRows.Close()
@ -750,7 +740,7 @@ func trainModel(c BasePack, model *BaseModel) (err error) {
if err = definitionsRows.Scan(&rowv.id, &rowv.target_accuracy, &rowv.epoch); err != nil {
l.Error("Failed to train Model Could not read definition from db!Err:")
l.Error(err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return
}
definitions = append(definitions, rowv)
@ -758,7 +748,7 @@ func trainModel(c BasePack, model *BaseModel) (err error) {
if len(definitions) == 0 {
l.Error("No Definitions defined!")
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return
}
@ -788,14 +778,14 @@ func trainModel(c BasePack, model *BaseModel) (err error) {
_, err = db.Exec("update model_definition set accuracy=$1, status=$2, epoch=$3 where id=$4", accuracy, MODEL_DEFINITION_STATUS_TRANIED, def.epoch, def.id)
if err != nil {
l.Error("Failed to train definition!Err:\n", "err", err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return err
}
_, err = db.Exec("update model_definition set status=$1 where id!=$2 and model_id=$3 and status!=$4", MODEL_DEFINITION_STATUS_CANCELD_TRAINING, def.id, model.Id, MODEL_DEFINITION_STATUS_FAILED_TRAINING)
if err != nil {
l.Error("Failed to train definition!Err:\n", "err", err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return err
}
@ -813,7 +803,7 @@ func trainModel(c BasePack, model *BaseModel) (err error) {
_, err = db.Exec("update model_definition set accuracy=$1, epoch=$2, status=$3 where id=$4", accuracy, def.epoch, MODEL_DEFINITION_STATUS_PAUSED_TRAINING, def.id)
if err != nil {
l.Error("Failed to train definition!Err:\n", "err", err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return err
}
}
@ -868,7 +858,7 @@ func trainModel(c BasePack, model *BaseModel) (err error) {
if err != nil {
l.Error("DB: failed to read definition")
l.Error(err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return
}
defer rows.Close()
@ -876,7 +866,7 @@ func trainModel(c BasePack, model *BaseModel) (err error) {
if !rows.Next() {
// TODO Make the Model status have a message
l.Error("All definitions failed to train!")
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return
}
@ -884,14 +874,14 @@ func trainModel(c BasePack, model *BaseModel) (err error) {
if err = rows.Scan(&id); err != nil {
l.Error("Failed to read id:")
l.Error(err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return
}
if _, err = db.Exec("update model_definition set status=$1 where id=$2;", MODEL_DEFINITION_STATUS_READY, id); err != nil {
l.Error("Failed to update model definition")
l.Error(err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return
}
@ -899,7 +889,7 @@ func trainModel(c BasePack, model *BaseModel) (err error) {
if err != nil {
l.Error("Failed to select model_definition to delete")
l.Error(err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return
}
defer to_delete.Close()
@ -909,7 +899,7 @@ func trainModel(c BasePack, model *BaseModel) (err error) {
if err = to_delete.Scan(&id); err != nil {
l.Error("Failed to scan the id of a model_definition to delete")
l.Error(err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return
}
os.RemoveAll(path.Join("savedData", model.Id, "defs", id))
@ -919,7 +909,7 @@ func trainModel(c BasePack, model *BaseModel) (err error) {
if _, err = db.Exec("delete from model_definition where status!=$1 and model_id=$2;", MODEL_DEFINITION_STATUS_READY, model.Id); err != nil {
l.Error("Failed to delete model_definition")
l.Error(err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
return
}
@ -1066,7 +1056,7 @@ func trainModelExp(c BasePack, model *BaseModel) (err error) {
err = GetDBOnce(db, &dat, "model_definition where model_id=$1 and status=$2 order by accuracy desc limit 1;", model.Id, MODEL_DEFINITION_STATUS_TRANIED)
if err == NotFoundError {
// Set the class status to trained
err = setModelClassStatus(c, MODEL_CLASS_STATUS_TO_TRAIN, "model_id=$1 and status=$2;", model.Id, MODEL_CLASS_STATUS_TRAINING)
err = setModelClassStatus(c, CLASS_STATUS_TO_TRAIN, "model_id=$1 and status=$2;", model.Id, CLASS_STATUS_TRAINING)
if err != nil {
l.Error("All definitions failed to train! And Failed to set class status")
return err
@ -1101,7 +1091,7 @@ func trainModelExp(c BasePack, model *BaseModel) (err error) {
}
if err = splitModel(c, model); err != nil {
err = setModelClassStatus(c, MODEL_CLASS_STATUS_TO_TRAIN, "model_id=$1 and status=$2;", model.Id, MODEL_CLASS_STATUS_TRAINING)
err = setModelClassStatus(c, CLASS_STATUS_TO_TRAIN, "model_id=$1 and status=$2;", model.Id, CLASS_STATUS_TRAINING)
if err != nil {
l.Error("Failed to split the model! And Failed to set class status")
return err
@ -1112,7 +1102,7 @@ func trainModelExp(c BasePack, model *BaseModel) (err error) {
}
// Set the class status to trained
err = setModelClassStatus(c, MODEL_CLASS_STATUS_TRAINED, "model_id=$1 and status=$2;", model.Id, MODEL_CLASS_STATUS_TRAINING)
err = setModelClassStatus(c, CLASS_STATUS_TRAINED, "model_id=$1 and status=$2;", model.Id, CLASS_STATUS_TRAINING)
if err != nil {
l.Error("Failed to set class status")
return err
@ -1272,7 +1262,7 @@ func generateDefinition(c BasePack, model *BaseModel, target_accuracy int, numbe
db := c.GetDb()
l := c.GetLogger()
def_id, err := MakeDefenition(db, model.Id, target_accuracy)
def, err := MakeDefenition(db, model.Id, target_accuracy)
if err != nil {
failed()
return
@ -1281,60 +1271,68 @@ func generateDefinition(c BasePack, model *BaseModel, target_accuracy int, numbe
order := 1
// Note the shape of the first layer defines the import size
if complexity == 2 {
// Note the shape for now is no used
width := int(math.Pow(2, math.Floor(math.Log(float64(model.Width))/math.Log(2.0))))
height := int(math.Pow(2, math.Floor(math.Log(float64(model.Height))/math.Log(2.0))))
l.Warn("Complexity 2 creating model with smaller size", "width", width, "height", height)
err = MakeLayer(db, def_id, order, LAYER_INPUT, fmt.Sprintf("%d,%d,1", width, height))
if err != nil {
failed()
return
}
order++
} else {
err = MakeLayer(db, def_id, order, LAYER_INPUT, fmt.Sprintf("%d,%d,1", model.Width, model.Height))
if err != nil {
failed()
return
}
order++
}
loop := max(int((math.Log(float64(model.Width)) / math.Log(float64(10)))), 1)
for i := 0; i < loop; i++ {
err = MakeLayer(db, def_id, order, LAYER_SIMPLE_BLOCK, "")
order++
if err != nil {
failed()
return
}
}
err = MakeLayer(db, def_id, order, LAYER_FLATTEN, "")
//_, err = def.MakeLayer(db, order, LAYER_INPUT, ShapeToString(model.Width, model.Height, model.ImageMode))
_, err = def.MakeLayer(db, order, LAYER_INPUT, ShapeToString(3, model.Width, model.Height))
if err != nil {
failed()
return
}
order++
loop = max(int((math.Log(float64(number_of_classes))/math.Log(float64(10)))/2), 1)
for i := 0; i < loop; i++ {
err = MakeLayer(db, def_id, order, LAYER_DENSE, fmt.Sprintf("%d,1", number_of_classes*(loop-i)))
order++
if complexity == 0 {
_, err = def.MakeLayer(db, order, LAYER_FLATTEN, "")
if err != nil {
failed()
return
}
}
order++
err = ModelDefinitionUpdateStatus(c, def_id, MODEL_DEFINITION_STATUS_INIT)
if err != nil {
loop := int(math.Log2(float64(number_of_classes)))
for i := 0; i < loop; i++ {
_, err = def.MakeLayer(db, order, LAYER_DENSE, ShapeToString(number_of_classes*(loop-i)))
order++
if err != nil {
ModelUpdateStatus(c, model.Id, FAILED_PREPARING_TRAINING)
return
}
}
} else if complexity == 1 || complexity == 2 {
loop := max(1, int((math.Log(float64(model.Width)) / math.Log(float64(10)))))
for i := 0; i < loop; i++ {
_, err = def.MakeLayer(db, order, LAYER_SIMPLE_BLOCK, "")
order++
if err != nil {
failed()
return
}
}
_, err = def.MakeLayer(db, order, LAYER_FLATTEN, "")
if err != nil {
failed()
return
}
order++
loop = int((math.Log(float64(number_of_classes)) / math.Log(float64(10))) / 2)
if loop == 0 {
loop = 1
}
for i := 0; i < loop; i++ {
_, err = def.MakeLayer(db, order, LAYER_DENSE, ShapeToString(number_of_classes*(loop-i)))
order++
if err != nil {
failed()
return
}
}
} else {
l.Error("Unkown complexity", "complexity", complexity)
failed()
return
}
return nil
return def.UpdateStatus(db, DEFINITION_STATUS_INIT)
}
func generateDefinitions(c BasePack, model *BaseModel, target_accuracy int, number_of_models int) (err error) {
@ -1393,12 +1391,14 @@ func generateExpandableDefinition(c BasePack, model *BaseModel, target_accuracy
return
}
def_id, err := MakeDefenition(c.GetDb(), model.Id, target_accuracy)
def, err := MakeDefenition(c.GetDb(), model.Id, target_accuracy)
if err != nil {
failed()
return
}
def_id := def.Id
order := 1
width := model.Width
@ -1533,7 +1533,7 @@ func generateExpandableDefinitions(c BasePack, model *BaseModel, target_accuracy
}
func ResetClasses(c BasePack, model *BaseModel) {
_, err := c.GetDb().Exec("update model_classes set status=$1 where status=$2 and model_id=$3", MODEL_CLASS_STATUS_TO_TRAIN, MODEL_CLASS_STATUS_TRAINING, model.Id)
_, err := c.GetDb().Exec("update model_classes set status=$1 where status=$2 and model_id=$3", CLASS_STATUS_TO_TRAIN, CLASS_STATUS_TRAINING, model.Id)
if err != nil {
c.GetLogger().Error("Error while reseting the classes", "error", err)
}
@ -1544,7 +1544,7 @@ func trainExpandable(c *Context, model *BaseModel) {
failed := func(msg string) {
c.Logger.Error(msg, "err", err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
ModelUpdateStatus(c, model.Id, int(FAILED_TRAINING))
ResetClasses(c, model)
}
@ -1588,7 +1588,7 @@ func trainExpandable(c *Context, model *BaseModel) {
}
// Set the class status to trained
err = setModelClassStatus(c, MODEL_CLASS_STATUS_TRAINED, "model_id=$1 and status=$2;", model.Id, MODEL_CLASS_STATUS_TRAINING)
err = setModelClassStatus(c, CLASS_STATUS_TRAINED, "model_id=$1 and status=$2;", model.Id, CLASS_STATUS_TRAINING)
if err != nil {
failed("Failed to set class status")
return
@ -1648,7 +1648,7 @@ func RunTaskTrain(b BasePack, task Task) (err error) {
if err != nil {
l.Error("Failed to train model", "err", err)
task.UpdateStatusLog(b, TASK_FAILED_RUNNING, "Failed generate model")
ModelUpdateStatus(b, model.Id, FAILED_TRAINING)
ModelUpdateStatus(b, model.Id, int(FAILED_TRAINING))
return
}
@ -1731,7 +1731,7 @@ func RunTaskRetrain(b BasePack, task Task) (err error) {
l.Info("Model updaded")
_, err = db.Exec("update model_classes set status=$1 where status=$2 and model_id=$3", MODEL_CLASS_STATUS_TRAINED, MODEL_CLASS_STATUS_TRAINING, model.Id)
_, err = db.Exec("update model_classes set status=$1 where status=$2 and model_id=$3", CLASS_STATUS_TRAINED, CLASS_STATUS_TRAINING, model.Id)
if err != nil {
l.Error("Error while updating the classes", "error", err)
failed()
@ -1861,7 +1861,7 @@ func handleTrain(handle *Handle) {
c,
"model_classes where model_id=$1 and status=$2 order by class_order asc",
model.Id,
MODEL_CLASS_STATUS_TO_TRAIN,
CLASS_STATUS_TO_TRAIN,
)
if err != nil {
_err := c.RollbackTx()
@ -1882,7 +1882,7 @@ func handleTrain(handle *Handle) {
//Update the classes
{
_, err = c.Exec("update model_classes set status=$1 where status=$2 and model_id=$3", MODEL_CLASS_STATUS_TRAINING, MODEL_CLASS_STATUS_TO_TRAIN, model.Id)
_, err = c.Exec("update model_classes set status=$1 where status=$2 and model_id=$3", CLASS_STATUS_TRAINING, CLASS_STATUS_TO_TRAIN, model.Id)
if err != nil {
_err := c.RollbackTx()
if _err != nil {

7
logic/tasks/README.md Normal file
View File

@ -0,0 +1,7 @@
# Runner Protocol
```
/----\
\/ |
Register -> Init -> Active ---> Ready -> Info
```

View File

@ -8,4 +8,5 @@ func HandleTasks(handle *Handle) {
handleUpload(handle)
handleList(handle)
handleRequests(handle)
handleRemoteRunner(handle)
}

386
logic/tasks/runner.go Normal file
View File

@ -0,0 +1,386 @@
package tasks
import (
"sync"
"time"
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/db_types"
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/models/train"
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/tasks/utils"
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/utils"
)
func verifyRunner(c *Context, dat *JustId) (runner *Runner, e *Error) {
runner, err := GetRunner(c, dat.Id)
if err == NotFoundError {
e = c.JsonBadRequest("Could not find runner, please register runner first")
return
} else if err != nil {
e = c.E500M("Failed to get information about the runner", err)
return
}
if runner.Token != *c.Token {
return nil, c.SendJSONStatus(401, "Only runners can use this funcion")
}
return
}
type VerifyTask struct {
Id string `json:"id" validate:"required"`
TaskId string `json:"taskId" validate:"required"`
}
func verifyTask(x *Handle, c *Context, dat *VerifyTask) (task *Task, error *Error) {
mutex := x.DataMap["runners_mutex"].(*sync.Mutex)
mutex.Lock()
defer mutex.Unlock()
var runners map[string]interface{} = x.DataMap["runners"].(map[string]interface{})
if runners[dat.Id] == nil {
return nil, c.JsonBadRequest("Runner not active")
}
var runner_data map[string]interface{} = runners[dat.Id].(map[string]interface{})
if runner_data["task"] == nil {
return nil, c.SendJSONStatus(404, "No active task")
}
return runner_data["task"].(*Task), nil
}
func handleRemoteRunner(x *Handle) {
type RegisterRunner struct {
Token string `json:"token" validate:"required"`
Type RunnerType `json:"type" validate:"required"`
}
PostAuthJson(x, "/tasks/runner/register", User_Normal, func(c *Context, dat *RegisterRunner) *Error {
if *c.Token != dat.Token {
// TODO do admin
return c.E500M("Please make sure that the token is the same that is being registered", nil)
}
c.Logger.Info("test", "dat", dat)
var runner Runner
err := GetDBOnce(c, &runner, "remote_runner as ru where token=$1", dat.Token)
if err != NotFoundError && err != nil {
return c.E500M("Failed to get information remote runners", err)
}
if err != NotFoundError {
return c.JsonBadRequest("Token is already registered by a runner")
}
// TODO get id from token passed by when doing admin
var userId = c.User.Id
var new_runner = struct {
Type RunnerType
UserId string `db:"user_id"`
Token string
}{
Type: dat.Type,
Token: dat.Token,
UserId: userId,
}
id, err := InsertReturnId(c, &new_runner, "remote_runner", "id")
if err != nil {
return c.E500M("Failed to create remote runner", err)
}
return c.SendJSON(struct {
Id string `json:"id"`
}{
Id: id,
})
})
// TODO remove runner
PostAuthJson(x, "/tasks/runner/init", User_Normal, func(c *Context, dat *JustId) *Error {
runner, error := verifyRunner(c, dat)
if error != nil {
return error
}
mutex := x.DataMap["runners_mutex"].(*sync.Mutex)
mutex.Lock()
defer mutex.Unlock()
var runners map[string]interface{} = x.DataMap["runners"].(map[string]interface{})
if runners[dat.Id] != nil {
c.Logger.Info("Logger trying to register but already registerd")
c.ShowMessage = false
return c.SendJSON("Ok")
}
var new_runner = map[string]interface{}{}
new_runner["last_time_check"] = time.Now()
new_runner["runner_info"] = runner
runners[dat.Id] = new_runner
x.DataMap["runners"] = runners
return c.SendJSON("Ok")
})
PostAuthJson(x, "/tasks/runner/active", User_Normal, func(c *Context, dat *JustId) *Error {
_, error := verifyRunner(c, dat)
if error != nil {
return error
}
mutex := x.DataMap["runners_mutex"].(*sync.Mutex)
mutex.Lock()
defer mutex.Unlock()
var runners map[string]interface{} = x.DataMap["runners"].(map[string]interface{})
if runners[dat.Id] == nil {
return c.JsonBadRequest("Runner not active")
}
var runner_data map[string]interface{} = runners[dat.Id].(map[string]interface{})
if runner_data["task"] == nil {
c.ShowMessage = false
return c.SendJSONStatus(404, "No active task")
}
c.ShowMessage = false
// This should be a task obj
return c.SendJSON(runner_data["task"])
})
PostAuthJson(x, "/tasks/runner/ready", User_Normal, func(c *Context, dat *VerifyTask) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, dat)
if error != nil {
return error
}
err := task.UpdateStatus(c, TASK_RUNNING, "Task Running on Runner")
if err != nil {
return c.E500M("Failed to set task status", err)
}
return c.SendJSON("Ok")
})
type TaskFail struct {
Id string `json:"id" validate:"required"`
TaskId string `json:"taskId" validate:"required"`
Reason string `json:"reason" validate:"required"`
}
PostAuthJson(x, "/tasks/runner/fail", User_Normal, func(c *Context, dat *TaskFail) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, &VerifyTask{Id: dat.Id, TaskId: dat.TaskId})
if error != nil {
return error
}
err := task.UpdateStatus(c, TASK_FAILED_RUNNING, dat.Reason)
if err != nil {
return c.E500M("Failed to set task status", err)
}
// Do extra clean up on tasks
switch task.TaskType {
case int(TASK_TYPE_TRAINING):
CleanUpFailed(c, task)
default:
panic("Do not know how to handle this")
}
mutex := x.DataMap["runners_mutex"].(*sync.Mutex)
mutex.Lock()
defer mutex.Unlock()
var runners map[string]interface{} = x.DataMap["runners"].(map[string]interface{})
var runner_data map[string]interface{} = runners[dat.Id].(map[string]interface{})
runner_data["task"] = nil
runners[dat.Id] = runner_data
x.DataMap["runners"] = runners
return c.SendJSON("Ok")
})
PostAuthJson(x, "/tasks/runner/train/defs", User_Normal, func(c *Context, dat *VerifyTask) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, dat)
if error != nil {
return error
}
if task.TaskType != int(TASK_TYPE_TRAINING) {
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
model, err := GetBaseModel(c, *task.ModelId)
if err != nil {
return c.E500M("Failed to get model information", err)
}
defs, err := model.GetDefinitions(c, "and md.status=$2", DEFINITION_STATUS_INIT)
if err != nil {
return c.E500M("Failed to get the model definitions", err)
}
return c.SendJSON(defs)
})
PostAuthJson(x, "/tasks/runner/train/classes", User_Normal, func(c *Context, dat *VerifyTask) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, dat)
if error != nil {
return error
}
if task.TaskType != int(TASK_TYPE_TRAINING) {
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
model, err := GetBaseModel(c, *task.ModelId)
if err != nil {
return c.E500M("Failed to get model information", err)
}
classes, err := model.GetClasses(c, "and status=$2 order by mc.class_order asc", CLASS_STATUS_TO_TRAIN)
if err != nil {
return c.E500M("Failed to get the model classes", err)
}
return c.SendJSON(classes)
})
type RunnerTrainDefStatus struct {
Id string `json:"id" validate:"required"`
TaskId string `json:"taskId" validate:"required"`
DefId string `json:"defId" validate:"required"`
Status DefinitionStatus `json:"status" validate:"required"`
}
PostAuthJson(x, "/tasks/runner/train/def/status", User_Normal, func(c *Context, dat *RunnerTrainDefStatus) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, &VerifyTask{Id: dat.Id, TaskId: dat.TaskId})
if error != nil {
return error
}
if task.TaskType != int(TASK_TYPE_TRAINING) {
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
def, err := GetDefinition(c, dat.DefId)
if err != nil {
return c.E500M("Failed to get definition information", err)
}
err = def.UpdateStatus(c, dat.Status)
if err != nil {
return c.E500M("Failed to update model status", err)
}
return c.SendJSON("Ok")
})
type RunnerTrainDefLayers struct {
Id string `json:"id" validate:"required"`
TaskId string `json:"taskId" validate:"required"`
DefId string `json:"defId" validate:"required"`
}
PostAuthJson(x, "/tasks/runner/train/def/layers", User_Normal, func(c *Context, dat *RunnerTrainDefLayers) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, &VerifyTask{Id: dat.Id, TaskId: dat.TaskId})
if error != nil {
return error
}
if task.TaskType != int(TASK_TYPE_TRAINING) {
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
def, err := GetDefinition(c, dat.DefId)
if err != nil {
return c.E500M("Failed to get definition information", err)
}
layers, err := def.GetLayers(c, " order by layer_order asc")
if err != nil {
return c.E500M("Failed to get layers", err)
}
return c.SendJSON(layers)
})
PostAuthJson(x, "/tasks/runner/train/datapoints", User_Normal, func(c *Context, dat *VerifyTask) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, dat)
if error != nil {
return error
}
if task.TaskType != int(TASK_TYPE_TRAINING) {
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
model, err := GetBaseModel(c, *task.ModelId)
if err != nil {
return c.E500M("Failed to get model information", err)
}
training_points, err := model.DataPoints(c, DATA_POINT_MODE_TRAINING)
if err != nil {
return c.E500M("Failed to get the model classes", err)
}
testing_points, err := model.DataPoints(c, DATA_POINT_MODE_TRAINING)
if err != nil {
return c.E500M("Failed to get the model classes", err)
}
return c.SendJSON(struct {
Testing []DataPoint `json:"testing"`
Training []DataPoint `json:"training"`
}{
Testing: testing_points,
Training: training_points,
})
})
}

View File

@ -5,6 +5,7 @@ import (
"math"
"os"
"runtime/debug"
"sync"
"time"
"github.com/charmbracelet/log"
@ -90,6 +91,45 @@ func runner(config Config, db db.Db, task_channel chan Task, index int, back_cha
}
}
/**
* Handle remote runner
*/
func handleRemoteTask(handler *Handle, base BasePack, runner_id string, task Task) {
logger := log.NewWithOptions(os.Stdout, log.Options{
ReportCaller: true,
ReportTimestamp: true,
TimeFormat: time.Kitchen,
Prefix: fmt.Sprintf("Runner pre %s", runner_id),
})
defer func() {
if r := recover(); r != nil {
logger.Error("Runner failed to setup for runner", "due to", r, "stack", string(debug.Stack()))
// TODO maybe create better failed task
task.UpdateStatusLog(base, TASK_FAILED_RUNNING, "Failed to setup task for runner")
}
}()
err := task.UpdateStatus(base, TASK_PICKED_UP, "Failed to setup task for runner")
if err != nil {
logger.Error("Failed to mark task as PICK UP")
return
}
mutex := handler.DataMap["runners_mutex"].(*sync.Mutex)
mutex.Lock()
defer mutex.Unlock()
switch task.TaskType {
case int(TASK_TYPE_TRAINING):
if err := PrepareTraining(handler, base, task, runner_id); err != nil {
logger.Error("Failed to prepare for training", "err", err)
}
default:
logger.Error("Not sure what to do panicing", "taskType", task.TaskType)
panic("not sure what to do")
}
}
/**
* Tells the orcchestator to look at the task list from time to time
*/
@ -125,7 +165,7 @@ func attentionSeeker(config Config, back_channel chan int) {
/**
* Manages what worker should to Work
*/
func RunnerOrchestrator(db db.Db, config Config) {
func RunnerOrchestrator(db db.Db, config Config, handler *Handle) {
logger := log.NewWithOptions(os.Stdout, log.Options{
ReportCaller: true,
ReportTimestamp: true,
@ -133,6 +173,16 @@ func RunnerOrchestrator(db db.Db, config Config) {
Prefix: "Runner Orchestrator Logger",
})
// Setup vars
handler.DataMap["runners"] = map[string]interface{}{}
handler.DataMap["runners_mutex"] = &sync.Mutex{}
base := BasePackStruct{
Db: db,
Logger: logger,
Host: config.Hostname,
}
gpu_workers := config.GpuWorker.NumberOfWorkers
logger.Info("Starting runners")
@ -149,7 +199,7 @@ func RunnerOrchestrator(db db.Db, config Config) {
close(task_runners[x])
}
close(back_channel)
go RunnerOrchestrator(db, config)
go RunnerOrchestrator(db, config, handler)
}
}()
@ -198,19 +248,45 @@ func RunnerOrchestrator(db db.Db, config Config) {
}
if task_to_dispatch != nil {
for i := 0; i < len(task_runners_used); i += 1 {
if !task_runners_used[i] {
task_runners[i] <- *task_to_dispatch
task_runners_used[i] = true
// Only let CPU tasks be done by the local users
if task_to_dispatch.TaskType == int(TASK_TYPE_DELETE_USER) {
for i := 0; i < len(task_runners_used); i += 1 {
if !task_runners_used[i] {
task_runners[i] <- *task_to_dispatch
task_runners_used[i] = true
task_to_dispatch = nil
break
}
}
continue
}
mutex := handler.DataMap["runners_mutex"].(*sync.Mutex)
mutex.Lock()
remote_runners := handler.DataMap["runners"].(map[string]interface{})
for k, v := range remote_runners {
runner_data := v.(map[string]interface{})
runner_info := runner_data["runner_info"].(*Runner)
if runner_data["task"] != nil {
continue
}
if runner_info.UserId == task_to_dispatch.UserId {
go handleRemoteTask(handler, base, k, *task_to_dispatch)
task_to_dispatch = nil
break
}
}
mutex.Unlock()
}
}
}
func StartRunners(db db.Db, config Config) {
go RunnerOrchestrator(db, config)
func StartRunners(db db.Db, config Config, handler *Handle) {
go RunnerOrchestrator(db, config, handler)
}

View File

@ -0,0 +1,29 @@
package tasks_utils
import (
"time"
"git.andr3h3nriqu3s.com/andr3/fyp/logic/db"
dbtypes "git.andr3h3nriqu3s.com/andr3/fyp/logic/db_types"
)
type RunnerType int64
const (
RUNNER_TYPE_GPU RunnerType = iota + 1
)
type Runner struct {
Id string `json:"id" db:"ru.id"`
UserId string `json:"user_id" db:"ru.user_id"`
Token string `json:"token" db:"ru.token"`
Type RunnerType `json:"type" db:"ru.type"`
CreateOn time.Time `json:"createOn" db:"ru.created_on"`
}
func GetRunner(db db.Db, id string) (ru *Runner, err error) {
var runner Runner
err = dbtypes.GetDBOnce(db, &runner, "remote_runner as ru where ru.id=$1", id)
ru = &runner
return
}

View File

@ -374,7 +374,7 @@ func (c Context) JsonBadRequest(dat any) *Error {
c.SetReportCaller(true)
c.Logger.Warn("Request failed with a bad request", "dat", dat)
c.SetReportCaller(false)
return c.ErrorCode(nil, 404, dat)
return c.SendJSONStatus(http.StatusBadRequest, dat)
}
func (c Context) JsonErrorBadRequest(err error, dat any) *Error {

View File

@ -36,11 +36,11 @@ func main() {
log.Info("Config loaded!", "config", config)
config.GenerateToken(db)
StartRunners(db, config)
//TODO check if file structure exists to save data
handle := NewHandler(db, config)
StartRunners(db, config, handle)
config.Cleanup(db)
// TODO Handle this in other way

View File

@ -13,7 +13,7 @@ http {
server {
listen 8000;
client_max_body_size 1G;
client_max_body_size 5G;
location / {
proxy_http_version 1.1;

6
requirements.txt Normal file
View File

@ -0,0 +1,6 @@
# tensorflow[and-cuda] == 2.15.1
tensorflow[and-cuda] == 2.9.1
pandas
# Make sure to install the nvidia pyindex first
# nvidia-pyindex
nvidia-cudnn

2
run.sh Executable file
View File

@ -0,0 +1,2 @@
#!/bin/bash
podman run --rm --network host --gpus all -ti -v $(pwd):/app -e "TERM=xterm-256color" fyp-server bash

1
runner/.gitignore vendored Normal file
View File

@ -0,0 +1 @@
target/

1935
runner/Cargo.lock generated Normal file

File diff suppressed because it is too large Load Diff

16
runner/Cargo.toml Normal file
View File

@ -0,0 +1,16 @@
[package]
name = "runner"
version = "0.1.0"
edition = "2021"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
anyhow = "1.0.82"
serde = { version = "1.0.200", features = ["derive"] }
toml = "0.8.12"
reqwest = { version = "0.12", features = ["json"] }
tokio = { version = "1", features = ["full"] }
serde_json = "1.0.116"
serde_repr = "0.1"
tch = { version = "0.16.0", features = ["download-libtorch"] }

12
runner/Dockerfile Normal file
View File

@ -0,0 +1,12 @@
FROM docker.io/nvidia/cuda:11.7.1-devel-ubuntu22.04
RUN apt-get update
RUN apt-get install -y curl
RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
ENV PATH="$PATH:/root/.cargo/bin"
RUN rustup toolchain install stable
RUN apt-get install -y pkg-config libssl-dev
WORKDIR /app

3
runner/config.toml Normal file
View File

@ -0,0 +1,3 @@
hostname = "https://testing.andr3h3nriqu3s.com/api"
token = "d2bc41e8293937bcd9397870c98f97acc9603f742924b518e193cd1013e45d57897aa302b364001c72b458afcfb34239dfaf38a66b318e5cbc973eea"
data_path = "/home/andr3/Documents/my-repos/fyp"

1
runner/data.toml Normal file
View File

@ -0,0 +1 @@
id = "a7cec9e9-1d05-4633-8bc5-6faabe4fd5a3"

2
runner/run.sh Executable file
View File

@ -0,0 +1,2 @@
#!/bin/bash
podman run --rm --network host --gpus all -ti -v $(pwd):/app -e "TERM=xterm-256color" fyp-runner bash

106
runner/src/dataloader.rs Normal file
View File

@ -0,0 +1,106 @@
use crate::{model::DataPoint, settings::ConfigFile};
use std::{path::Path, sync::Arc};
use tch::Tensor;
pub struct DataLoader {
pub batch_size: i64,
pub len: usize,
pub inputs: Vec<Tensor>,
pub labels: Vec<Tensor>,
pub pos: usize,
}
impl DataLoader {
pub fn new(
config: Arc<ConfigFile>,
data: Vec<DataPoint>,
classes_len: i64,
batch_size: i64,
) -> DataLoader {
let len: f64 = (data.len() as f64) / (batch_size as f64);
let min_len: i64 = len.floor() as i64;
let max_len: i64 = len.ceil() as i64;
let base_path = Path::new(&config.data_path);
let mut inputs: Vec<Tensor> = Vec::new();
let mut all_labels: Vec<Tensor> = Vec::new();
for batch in 0..min_len {
let mut batch_acc: Vec<Tensor> = Vec::new();
let mut labels: Vec<Tensor> = Vec::new();
for image in 0..batch_size {
let i: usize = (batch * batch_size + image).try_into().unwrap();
let item = &data[i];
batch_acc.push(
tch::vision::image::load(base_path.join(&item.path))
.ok()
.unwrap(),
);
if item.class >= 0 {
let t = tch::Tensor::from_slice(&[item.class])
.onehot(classes_len.try_into().unwrap());
labels.push(t);
} else {
labels.push(tch::Tensor::zeros(
(classes_len),
(tch::Kind::Float, tch::Device::Cpu),
))
}
}
inputs.push(tch::Tensor::cat(&batch_acc[0..], 0));
all_labels.push(tch::Tensor::cat(&labels[0..], 0));
}
if min_len != max_len {
let mut batch_acc: Vec<Tensor> = Vec::new();
let mut labels: Vec<Tensor> = Vec::new();
for image in 0..(data.len() - (batch_size * min_len) as usize) {
let i: usize = (min_len * batch_size + (image as i64)) as usize;
let item = &data[i];
batch_acc.push(
tch::vision::image::load(base_path.join(&item.path))
.ok()
.unwrap(),
);
if item.class >= 0 {
let t = tch::Tensor::from_slice(&[item.class]).onehot(classes_len);
labels.push(t);
} else {
labels.push(tch::Tensor::zeros(
classes_len,
(tch::Kind::Float, tch::Device::Cpu),
))
}
}
inputs.push(tch::Tensor::cat(&batch_acc[0..], 0));
all_labels.push(tch::Tensor::cat(&labels[0..], 0));
}
return DataLoader {
batch_size,
inputs,
labels: all_labels,
len: max_len as usize,
pos: 0,
};
}
pub fn restart(self: &mut DataLoader) {
self.pos = 0;
}
pub fn next(self: &mut DataLoader) -> Option<(Tensor, Tensor)> {
if self.pos >= self.len {
return None;
}
let input = self.inputs[self.pos].empty_like();
self.inputs[self.pos] = self.inputs[self.pos].clone(&input);
let label = self.labels[self.pos].empty_like();
self.labels[self.pos] = self.labels[self.pos].clone(&label);
return Some((input, label));
}
}

206
runner/src/main.rs Normal file
View File

@ -0,0 +1,206 @@
mod dataloader;
mod model;
mod settings;
mod tasks;
mod training;
mod types;
use crate::settings::*;
use crate::tasks::{fail_task, Task, TaskType};
use crate::training::handle_train;
use anyhow::{bail, Result};
use reqwest::StatusCode;
use serde_json::json;
use std::{fs, process::exit, sync::Arc, time::Duration};
enum ResultAlive {
Ok,
Error,
NotInit,
}
async fn send_keep_alive_message(
config: Arc<ConfigFile>,
runner_data: Arc<RunnerData>,
) -> ResultAlive {
let client = reqwest::Client::new();
let to_send = json!({
"id": runner_data.id,
});
let resp = client
.post(format!("{}/tasks/runner/beat", config.hostname))
.header("token", &config.token)
.body(to_send.to_string())
.send()
.await;
if resp.is_err() {
return ResultAlive::Error;
}
let resp = resp.ok();
if resp.is_none() {
return ResultAlive::Error;
}
let resp = resp.unwrap();
// TODO see if the message is related to not being inited
if resp.status() != 200 {
println!("Could not connect with the status");
return ResultAlive::Error;
}
ResultAlive::Ok
}
async fn keep_alive(config: Arc<ConfigFile>, runner_data: Arc<RunnerData>) -> Result<()> {
let mut failed = 0;
loop {
match send_keep_alive_message(config.clone(), runner_data.clone()).await {
ResultAlive::Error => failed += 1,
ResultAlive::NotInit => {
println!("Runner not inited! Restarting!");
exit(1)
}
ResultAlive::Ok => failed = 0,
}
// TODO move to config
if failed > 20 {
println!("Failed to connect to API! More than 20 times in a row stoping");
exit(1)
}
tokio::time::sleep(Duration::from_secs(1)).await;
}
}
async fn handle_task(
task: Task,
config: Arc<ConfigFile>,
runner_data: Arc<RunnerData>,
) -> Result<()> {
let res = match task.task_type {
TaskType::Training => handle_train(&task, config.clone(), runner_data.clone()).await,
_ => {
println!("Do not know how to handle this task #{:?}", task);
bail!("Failed")
}
};
if res.is_err() {
println!("task failed #{:?}", res);
fail_task(
&task,
config,
runner_data,
"Do not know how to handle this kind of task",
)
.await?
}
Ok(())
}
#[tokio::main]
async fn main() -> Result<()> {
// Load config file
let config_data = fs::read_to_string("./config.toml")?;
let mut config: ConfigFile = toml::from_str(&config_data)?;
let client = reqwest::Client::new();
if config.config_path == None {
config.config_path = Some(String::from("./data.toml"))
}
let runner_data: RunnerData = load_runner_data(&config).await?;
let to_send = json!({
"id": runner_data.id,
});
// Inform the server that the runner is available
let resp = client
.post(format!("{}/tasks/runner/init", config.hostname))
.header("token", &config.token)
.body(to_send.to_string())
.send()
.await?;
if resp.status() != 200 {
println!(
"Could not connect with the api: status {} body {}",
resp.status(),
resp.text().await?
);
return Ok(());
}
let res = resp.json::<String>().await?;
if res != "Ok" {
print!("Unexpected problem: {}", res);
return Ok(());
}
let config = Arc::new(config);
let runner_data = Arc::new(runner_data);
let config_alive = config.clone();
let runner_data_alive = runner_data.clone();
std::thread::spawn(move || keep_alive(config_alive.clone(), runner_data_alive.clone()));
println!("Started main loop");
loop {
//TODO move time to config
tokio::time::sleep(Duration::from_secs(1)).await;
let to_send = json!({ "id": runner_data.id });
let resp = client
.post(format!("{}/tasks/runner/active", config.hostname))
.header("token", &config.token)
.body(to_send.to_string())
.send()
.await;
if resp.is_err() || resp.as_ref().ok().is_none() {
println!("Failed to get info from server {:?}", resp);
continue;
}
let resp = resp?;
match resp.status() {
// No active task
StatusCode::NOT_FOUND => (),
StatusCode::OK => {
println!("Found task!");
let task: Result<Task, reqwest::Error> = resp.json().await;
if task.is_err() || task.as_ref().ok().is_none() {
println!("Failed to resolve the json {:?}", task);
continue;
}
let task = task?;
let res = handle_task(task, config.clone(), runner_data.clone()).await;
if res.is_err() || res.as_ref().ok().is_none() {
println!("Failed to run the task");
}
_ = res;
()
}
_ => {
println!("Unexpected error #{:?}", resp);
exit(1)
}
}
}
}

99
runner/src/model/mod.rs Normal file
View File

@ -0,0 +1,99 @@
use anyhow::bail;
use serde::{Deserialize, Serialize};
use serde_repr::{Deserialize_repr, Serialize_repr};
use tch::{
nn::{self, Module},
Device,
};
#[derive(Debug)]
pub struct Model {
pub vs: nn::VarStore,
pub seq: nn::Sequential,
pub layers: Vec<Layer>,
}
#[derive(Debug, Clone, Copy, Serialize_repr, Deserialize_repr)]
#[repr(i8)]
pub enum LayerType {
Input = 1,
Dense = 2,
Flatten = 3,
SimpleBlock = 4,
}
#[derive(Debug, Serialize, Deserialize)]
pub struct Layer {
pub id: String,
pub definition_id: String,
pub layer_order: String,
pub layer_type: LayerType,
pub shape: String,
pub exp_type: String,
}
#[derive(Debug, Serialize, Deserialize)]
pub struct DataPoint {
pub class: i64,
pub path: String,
}
pub fn build_model(layers: Vec<Layer>, last_linear_size: i64, add_sigmoid: bool) -> Model {
let vs = nn::VarStore::new(Device::Cpu);
let mut seq = nn::seq();
let mut last_linear_size = last_linear_size;
let mut last_linear_conv: Vec<i64> = Vec::new();
for layer in layers.iter() {
match layer.layer_type {
LayerType::Input => {
last_linear_conv = serde_json::from_str(&layer.shape).ok().unwrap();
println!("Layer: Input, In: {:?}", last_linear_conv);
}
LayerType::Dense => {
let shape: Vec<i64> = serde_json::from_str(&layer.shape).ok().unwrap();
println!("Layer: Dense, In: {}, Out: {}", last_linear_size, shape[0]);
seq = seq
.add(nn::linear(
&vs.root(),
last_linear_size,
shape[0],
Default::default(),
))
.add_fn(|xs| xs.relu());
last_linear_size = shape[0];
}
LayerType::Flatten => {
seq = seq.add_fn(|xs| xs.flatten(1, -1));
last_linear_size = 1;
for i in &last_linear_conv {
last_linear_size *= i;
}
println!(
"Layer: flatten, In: {:?}, Out: {}",
last_linear_conv, last_linear_size
)
}
LayerType::SimpleBlock => {
panic!("DO not create Simple blocks yet");
let new_last_linear_conv =
vec![128, last_linear_conv[1] / 2, last_linear_conv[2] / 2];
println!(
"Layer: block, In: {:?}, Put: {:?}",
last_linear_conv, new_last_linear_conv,
);
//TODO
//m_layers = append(m_layers, NewSimpleBlock(vs, lastLinearConv[0]))
last_linear_conv = new_last_linear_conv;
}
}
}
if add_sigmoid {
seq = seq.add_fn(|xs| xs.sigmoid());
}
return Model { vs, layers, seq };
}

57
runner/src/settings.rs Normal file
View File

@ -0,0 +1,57 @@
use anyhow::{bail, Result};
use serde::{Deserialize, Serialize};
use serde_json::json;
use std::{fs, path};
#[derive(Deserialize)]
pub struct ConfigFile {
// Hostname to connect with the api
pub hostname: String,
// Token used in the api to authenticate
pub token: String,
// Path to where to store some generated configuration values
// defaults to ./data.toml
pub config_path: Option<String>,
// Data Path
// Path to where the data is mounted
pub data_path: String,
}
#[derive(Deserialize, Serialize)]
pub struct RunnerData {
pub id: String,
}
pub async fn load_runner_data(config: &ConfigFile) -> Result<RunnerData> {
let data_path = config.config_path.as_ref().unwrap();
let data_path = path::Path::new(&*data_path);
if data_path.exists() {
let runner_data = fs::read_to_string(data_path)?;
Ok(toml::from_str(&runner_data)?)
} else {
let client = reqwest::Client::new();
let to_send = json!({
"token": config.token,
"type": 1,
});
let register_resp = client
.post(format!("{}/tasks/runner/register", config.hostname))
.header("token", &config.token)
.body(to_send.to_string())
.send()
.await?;
if register_resp.status() != 200 {
bail!(format!("Could not create runner {:#?}", register_resp));
}
let runner_data: RunnerData = register_resp.json().await?;
fs::write(data_path, toml::to_string(&runner_data)?)
.expect("Faield to save data for runner");
Ok(runner_data)
}
}

90
runner/src/tasks.rs Normal file
View File

@ -0,0 +1,90 @@
use std::sync::Arc;
use anyhow::{bail, Result};
use serde::Deserialize;
use serde_json::json;
use serde_repr::Deserialize_repr;
use crate::{ConfigFile, RunnerData};
#[derive(Clone, Copy, Deserialize_repr, Debug)]
#[repr(i8)]
pub enum TaskStatus {
FailedRunning = -2,
FailedCreation = -1,
Preparing = 0,
Todo = 1,
PickedUp = 2,
Running = 3,
Done = 4,
}
#[derive(Clone, Copy, Deserialize_repr, Debug)]
#[repr(i8)]
pub enum TaskType {
Classification = 1,
Training = 2,
Retraining = 3,
DeleteUser = 4,
}
#[derive(Deserialize, Debug)]
pub struct Task {
pub id: String,
pub user_id: String,
pub model_id: String,
pub status: TaskStatus,
pub status_message: String,
pub user_confirmed: i8,
pub compacted: i8,
#[serde(alias = "type")]
pub task_type: TaskType,
pub extra_task_info: String,
pub result: String,
pub created: String,
}
pub async fn fail_task(
task: &Task,
config: Arc<ConfigFile>,
runner_data: Arc<RunnerData>,
reason: &str,
) -> Result<()> {
println!("Marking Task as faield");
let client = reqwest::Client::new();
let to_send = json!({
"id": runner_data.id,
"taskId": task.id,
"reason": reason
});
let resp = client
.post(format!("{}/tasks/runner/fail", config.hostname))
.header("token", &config.token)
.body(to_send.to_string())
.send()
.await?;
if resp.status() != 200 {
println!("Failed to update status of task");
bail!("Failed to update status of task");
}
Ok(())
}
impl Task {
pub async fn fail(
self: &mut Task,
config: Arc<ConfigFile>,
runner_data: Arc<RunnerData>,
reason: &str,
) -> Result<()> {
fail_task(self, config, runner_data, reason).await?;
self.status = TaskStatus::FailedRunning;
self.status_message = reason.to_string();
Ok(())
}
}

544
runner/src/training.rs Normal file
View File

@ -0,0 +1,544 @@
use crate::{
dataloader::DataLoader,
model::{self, build_model},
settings::{ConfigFile, RunnerData},
tasks::{fail_task, Task},
types::{DataPointRequest, Definition, ModelClass},
};
use std::sync::Arc;
use anyhow::Result;
use serde_json::json;
use tch::{
nn::{self, Module, OptimizerConfig},
Tensor,
};
pub async fn handle_train(
task: &Task,
config: Arc<ConfigFile>,
runner_data: Arc<RunnerData>,
) -> Result<()> {
let client = reqwest::Client::new();
println!("Preparing to train a model");
let to_send = json!({
"id": runner_data.id,
"taskId": task.id,
});
let mut defs: Vec<Definition> = client
.post(format!("{}/tasks/runner/train/defs", config.hostname))
.header("token", &config.token)
.body(to_send.to_string())
.send()
.await?
.json()
.await?;
let classes: Vec<ModelClass> = client
.post(format!("{}/tasks/runner/train/classes", config.hostname))
.header("token", &config.token)
.body(to_send.to_string())
.send()
.await?
.json()
.await?;
let data: DataPointRequest = client
.post(format!("{}/tasks/runner/train/datapoints", config.hostname))
.header("token", &config.token)
.body(to_send.to_string())
.send()
.await?
.json()
.await?;
let mut data_loader = DataLoader::new(config.clone(), data.testing, classes.len() as i64, 64);
// TODO make this a vec
let mut model: Option<model::Model> = None;
loop {
let config = config.clone();
let runner_data = runner_data.clone();
let mut to_remove: Vec<usize> = Vec::new();
let mut def_iter = defs.iter_mut();
let mut i: usize = 0;
while let Some(def) = def_iter.next() {
def.updateStatus(
task,
config.clone(),
runner_data.clone(),
crate::types::DefinitionStatus::Training,
)
.await?;
let model_err = train_definition(
def,
&mut data_loader,
model,
config.clone(),
runner_data.clone(),
&task,
)
.await;
if model_err.is_err() {
println!("Failed to create model {:?}", model_err);
model = None;
to_remove.push(i);
continue;
}
model = model_err?;
i += 1;
}
defs = defs
.into_iter()
.enumerate()
.filter(|&(i, _)| to_remove.iter().any(|b| *b == i))
.map(|(_, e)| e)
.collect();
break;
}
fail_task(task, config, runner_data, "TODO").await?;
Ok(())
/*
for {
// Keep track of definitions that did not train fast enough
var toRemove ToRemoveList = []int{}
for i, def := range definitions {
accuracy, ml_model, err := trainDefinition(c, model, def, models[def.Id], classes)
if err != nil {
log.Error("Failed to train definition!Err:", "err", err)
def.UpdateStatus(c, DEFINITION_STATUS_FAILED_TRAINING)
toRemove = append(toRemove, i)
continue
}
models[def.Id] = ml_model
if accuracy >= float64(def.TargetAccuracy) {
log.Info("Found a definition that reaches target_accuracy!")
_, err = db.Exec("update model_definition set accuracy=$1, status=$2, epoch=$3 where id=$4", accuracy, DEFINITION_STATUS_TRANIED, def.Epoch, def.Id)
if err != nil {
log.Error("Failed to train definition!Err:\n", "err", err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
return err
}
_, err = db.Exec("update model_definition set status=$1 where id!=$2 and model_id=$3 and status!=$4", DEFINITION_STATUS_CANCELD_TRAINING, def.Id, model.Id, DEFINITION_STATUS_FAILED_TRAINING)
if err != nil {
log.Error("Failed to train definition!Err:\n", "err", err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
return err
}
finished = true
break
}
if def.Epoch > MAX_EPOCH {
fmt.Printf("Failed to train definition! Accuracy less %f < %d\n", accuracy, def.TargetAccuracy)
def.UpdateStatus(c, DEFINITION_STATUS_FAILED_TRAINING)
toRemove = append(toRemove, i)
continue
}
_, err = db.Exec("update model_definition set accuracy=$1, epoch=$2, status=$3 where id=$4", accuracy, def.Epoch, DEFINITION_STATUS_PAUSED_TRAINING, def.Id)
if err != nil {
log.Error("Failed to train definition!Err:\n", "err", err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
return err
}
}
if finished {
break
}
sort.Sort(sort.Reverse(toRemove))
log.Info("Round done", "toRemove", toRemove)
for _, n := range toRemove {
// Clean up unsed models
models[definitions[n].Id] = nil
definitions = remove(definitions, n)
}
len_def := len(definitions)
if len_def == 0 {
break
}
if len_def == 1 {
continue
}
sort.Sort(sort.Reverse(definitions))
acc := definitions[0].Accuracy - 20.0
log.Info("Training models, Highest acc", "acc", definitions[0].Accuracy, "mod_acc", acc)
toRemove = []int{}
for i, def := range definitions {
if def.Accuracy < acc {
toRemove = append(toRemove, i)
}
}
log.Info("Removing due to accuracy", "toRemove", toRemove)
sort.Sort(sort.Reverse(toRemove))
for _, n := range toRemove {
log.Warn("Removing definition not fast enough learning", "n", n)
definitions[n].UpdateStatus(c, DEFINITION_STATUS_FAILED_TRAINING)
models[definitions[n].Id] = nil
definitions = remove(definitions, n)
}
}
var def Definition
err = GetDBOnce(c, &def, "model_definition as md where md.model_id=$1 and md.status=$2 order by md.accuracy desc limit 1;", model.Id, DEFINITION_STATUS_TRANIED)
if err != nil {
if err == NotFoundError {
log.Error("All definitions failed to train!")
} else {
log.Error("DB: failed to read definition", "err", err)
}
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
return
}
if err = def.UpdateStatus(c, DEFINITION_STATUS_READY); err != nil {
log.Error("Failed to update model definition", "err", err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
return
}
to_delete, err := db.Query("select id from model_definition where status != $1 and model_id=$2", DEFINITION_STATUS_READY, model.Id)
if err != nil {
log.Error("Failed to select model_definition to delete")
log.Error(err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
return
}
defer to_delete.Close()
for to_delete.Next() {
var id string
if err = to_delete.Scan(&id); err != nil {
log.Error("Failed to scan the id of a model_definition to delete", "err", err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
return
}
os.RemoveAll(path.Join("savedData", model.Id, "defs", id))
}
// TODO Check if returning also works here
if _, err = db.Exec("delete from model_definition where status!=$1 and model_id=$2;", DEFINITION_STATUS_READY, model.Id); err != nil {
log.Error("Failed to delete model_definition")
log.Error(err)
ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
return
}
ModelUpdateStatus(c, model.Id, READY)
return
*/
}
async fn train_definition(
def: &Definition,
data_loader: &mut DataLoader,
model: Option<model::Model>,
config: Arc<ConfigFile>,
runner_data: Arc<RunnerData>,
task: &Task,
) -> Result<Option<model::Model>> {
let client = reqwest::Client::new();
println!("About to start training definition");
let mut accuracy = 0;
let model = model.unwrap_or({
let layers: Vec<model::Layer> = client
.post(format!("{}/tasks/runner/train/def/layers", config.hostname))
.header("token", &config.token)
.body(
json!({
"id": runner_data.id,
"taskId": task.id,
"defId": def.id,
})
.to_string(),
)
.send()
.await?
.json()
.await?;
build_model(layers, 0, true)
});
println!("here1!");
// TODO CUDA
// get device
// Move model to cuda
let mut opt = nn::Adam::default().build(&model.vs, 1e-5)?;
println!("here2!");
for epoch in 1..20 {
data_loader.restart();
while let Some((inputs, labels)) = data_loader.next() {
let inputs = inputs.to_kind(tch::Kind::Float);
let labels = labels.to_kind(tch::Kind::Float);
println!("ins: {:?} labels: {:?}", inputs.size(), labels.size());
let out = model.seq.forward(&inputs);
let weight: Option<Tensor> = None;
let loss = out.binary_cross_entropy(&labels, weight, tch::Reduction::Mean);
opt.backward_step(&loss);
println!("out: {:?}", out);
}
}
return Ok(Some(model));
/*
opt, err := my_nn.DefaultAdamConfig().Build(model.Vs, 0.001)
if err != nil {
return
}
for epoch := 0; epoch < EPOCH_PER_RUN; epoch++ {
var trainIter *torch.Iter2
trainIter, err = ds.TrainIter(32)
if err != nil {
return
}
// trainIter.ToDevice(device)
log.Info("epoch", "epoch", epoch)
var trainLoss float64 = 0
var trainCorrect float64 = 0
ok := true
for ok {
var item torch.Iter2Item
var loss *torch.Tensor
item, ok = trainIter.Next()
if !ok {
continue
}
data := item.Data
data, err = data.ToDevice(device, gotch.Float, false, true, false)
if err != nil {
return
}
var size []int64
size, err = data.Size()
if err != nil {
return
}
var zeros *torch.Tensor
zeros, err = torch.Zeros(size, gotch.Float, device)
if err != nil {
return
}
data, err = zeros.Add(data, true)
if err != nil {
return
}
log.Info("\n\nhere 1, data\n\n", "retains", data.MustRetainsGrad(false), "requires", data.MustRequiresGrad())
data, err = data.SetRequiresGrad(true, false)
if err != nil {
return
}
log.Info("\n\nhere 2, data\n\n", "retains", data.MustRetainsGrad(false), "requires", data.MustRequiresGrad())
err = data.RetainGrad(false)
if err != nil {
return
}
log.Info("\n\nhere 2, data\n\n", "retains", data.MustRetainsGrad(false), "requires", data.MustRequiresGrad())
pred := model.ForwardT(data, true)
pred, err = pred.SetRequiresGrad(true, true)
if err != nil {
return
}
err = pred.RetainGrad(false)
if err != nil {
return
}
label := item.Label
label, err = label.ToDevice(device, gotch.Float, false, true, false)
if err != nil {
return
}
label, err = label.SetRequiresGrad(true, true)
if err != nil {
return
}
err = label.RetainGrad(false)
if err != nil {
return
}
// Calculate loss
loss, err = pred.BinaryCrossEntropyWithLogits(label, &torch.Tensor{}, &torch.Tensor{}, 2, false)
if err != nil {
return
}
loss, err = loss.SetRequiresGrad(true, false)
if err != nil {
return
}
err = loss.RetainGrad(false)
if err != nil {
return
}
err = opt.ZeroGrad()
if err != nil {
return
}
err = loss.Backward()
if err != nil {
return
}
log.Info("pred grad", "pred", pred.MustGrad(false).MustMax(false).Float64Values())
log.Info("pred grad", "outs", label.MustGrad(false).MustMax(false).Float64Values())
log.Info("pred grad", "data", data.MustGrad(false).MustMax(false).Float64Values(), "lol", data.MustRetainsGrad(false))
vars := model.Vs.Variables()
for k, v := range vars {
log.Info("[grad check]", "k", k, "grad", v.MustGrad(false).MustMax(false).Float64Values(), "lol", v.MustRetainsGrad(false))
}
model.Debug()
err = opt.Step()
if err != nil {
return
}
trainLoss = loss.Float64Values()[0]
// Calculate accuracy
/ *var p_pred, p_labels *torch.Tensor
p_pred, err = pred.Argmax([]int64{1}, true, false)
if err != nil {
return
}
p_labels, err = item.Label.Argmax([]int64{1}, true, false)
if err != nil {
return
}
floats := p_pred.Float64Values()
floats_labels := p_labels.Float64Values()
for i := range floats {
if floats[i] == floats_labels[i] {
trainCorrect += 1
}
} * /
panic("fornow")
}
//v := []float64{}
log.Info("model training epoch done loss", "loss", trainLoss, "correct", trainCorrect, "out", ds.TrainImagesSize, "accuracy", trainCorrect/float64(ds.TrainImagesSize))
/ *correct := int64(0)
//torch.NoGrad(func() {
ok = true
testIter := ds.TestIter(64)
for ok {
var item torch.Iter2Item
item, ok = testIter.Next()
if !ok {
continue
}
output := model.Forward(item.Data)
var pred, labels *torch.Tensor
pred, err = output.Argmax([]int64{1}, true, false)
if err != nil {
return
}
labels, err = item.Label.Argmax([]int64{1}, true, false)
if err != nil {
return
}
floats := pred.Float64Values()
floats_labels := labels.Float64Values()
for i := range floats {
if floats[i] == floats_labels[i] {
correct += 1
}
}
}
accuracy = float64(correct) / float64(ds.TestImagesSize)
log.Info("Eval accuracy", "accuracy", accuracy)
err = def.UpdateAfterEpoch(db, accuracy*100)
if err != nil {
return
}* /
//})
}
result_path := path.Join(getDir(), "savedData", m.Id, "defs", def.Id)
err = os.MkdirAll(result_path, os.ModePerm)
if err != nil {
return
}
err = my_torch.SaveModel(model, path.Join(result_path, "model.dat"))
if err != nil {
return
}
log.Info("Model finished training!", "accuracy", accuracy)
return
*/
}

89
runner/src/types.rs Normal file
View File

@ -0,0 +1,89 @@
use crate::{model, tasks::Task, ConfigFile, RunnerData};
use anyhow::{bail, Result};
use serde::Deserialize;
use serde_json::json;
use serde_repr::{Deserialize_repr, Serialize_repr};
use std::sync::Arc;
#[derive(Clone, Copy, Deserialize_repr, Serialize_repr, Debug)]
#[repr(i8)]
pub enum DefinitionStatus {
CanceldTraining = -4,
FailedTraining = -3,
PreInit = 1,
Init = 2,
Training = 3,
PausedTraining = 6,
Tranied = 4,
Ready = 5,
}
#[derive(Deserialize, Debug)]
pub struct Definition {
pub id: String,
pub model_id: String,
pub accuracy: f64,
pub target_accuracy: i64,
pub epoch: i64,
pub status: i64,
pub created: String,
pub epoch_progress: i64,
}
impl Definition {
pub async fn updateStatus(
self: &mut Definition,
task: &Task,
config: Arc<ConfigFile>,
runner_data: Arc<RunnerData>,
status: DefinitionStatus,
) -> Result<()> {
println!("Marking Task as faield");
let client = reqwest::Client::new();
let to_send = json!({
"id": runner_data.id,
"taskId": task.id,
"defId": self.id,
"status": status,
});
let resp = client
.post(format!("{}/tasks/runner/train/def/status", config.hostname))
.header("token", &config.token)
.body(to_send.to_string())
.send()
.await?;
if resp.status() != 200 {
println!("Failed to update status of task");
bail!("Failed to update status of task");
}
Ok(())
}
}
#[derive(Clone, Copy, Deserialize_repr, Debug)]
#[repr(i8)]
pub enum ModelClassStatus {
ToTrain = 1,
Training = 2,
Trained = 3,
}
#[derive(Deserialize, Debug)]
pub struct ModelClass {
pub id: String,
pub model_id: String,
pub name: String,
pub class_order: i64,
pub status: ModelClassStatus,
}
#[derive(Deserialize, Debug)]
pub struct DataPointRequest {
pub testing: Vec<model::DataPoint>,
pub training: Vec<model::DataPoint>,
}

View File

@ -38,3 +38,14 @@ create table if not exists tasks_dependencies (
main_id uuid references tasks (id) on delete cascade not null,
dependent_id uuid references tasks (id) on delete cascade not null
);
create table if not exists remote_runner (
id uuid primary key default gen_random_uuid(),
user_id uuid references users (id) on delete cascade not null,
token text not null,
-- 1: GPU
type integer,
created_on timestamp default current_timestamp
);

View File

@ -82,7 +82,7 @@ def prepare_dataset(ds: tf.data.Dataset, size: int) -> tf.data.Dataset:
def filterDataset(path):
path = tf.strings.regex_replace(path, DATA_DIR_PREPARE, "")
{{ if eq .Model.Format "png" }}
path = tf.strings.regex_replace(path, ".png", "")
{{ else if eq .Model.Format "jpeg" }}
@ -90,7 +90,7 @@ def filterDataset(path):
{{ else }}
ERROR
{{ end }}
return tf.reshape(table.lookup(tf.strings.as_string([path])), []) != -1
seed = random.randint(0, 100000000)
@ -135,9 +135,9 @@ def addBlock(
model.add(layers.ReLU())
if top:
if pooling_same:
model.add(pool_func(padding="same", strides=(1, 1)))
model.add(pool_func(pool_size=(2,2), padding="same", strides=(1, 1)))
else:
model.add(pool_func())
model.add(pool_func(pool_size=(2,2)))
model.add(layers.BatchNormalization())
model.add(layers.LeakyReLU())
model.add(layers.Dropout(0.4))
@ -172,7 +172,7 @@ model.compile(
his = model.fit(dataset, validation_data= dataset_validation, epochs={{.EPOCH_PER_RUN}}, callbacks=[
NotifyServerCallback(),
tf.keras.callbacks.EarlyStopping("loss", mode="min", patience=5)], use_multiprocessing = True)
tf.keras.callbacks.EarlyStopping("loss", mode="min", patience=5)])
acc = his.history["accuracy"]