worked on #32

logic/models/edit.go

@@ -101,7 +101,7 @@ func handleEdit(handle *Handle) {
 			type defrow struct {
 				Status        int
 				EpochProgress int
-				Accuracy      int
+				Accuracy      float64
 			}
 
 			def_rows, err := c.Db.Query("select status, epoch_progress, accuracy from model_definition where model_id=$1", model.Id)

logic/models/train/train.go

@@ -53,6 +53,7 @@ const (
 	LAYER_INPUT   LayerType = 1
 	LAYER_DENSE             = 2
 	LAYER_FLATTEN           = 3
+	LAYER_SIMPLE_BLOCK      = 4
 )
 
 func ModelDefinitionUpdateStatus(c *Context, id string, status ModelDefinitionStatus) (err error) {
@@ -207,13 +208,13 @@ func trainDefinition(c *Context, model *BaseModel, definition_id string, load_pr
 		return
 	}
 
-	c.Logger.Info("Model finished training!", "accuracy", accuracy)
-
 	accuracy, err = strconv.ParseFloat(string(accuracy_file_bytes), 64)
 	if err != nil {
 		return
 	}
 
+	c.Logger.Info("Model finished training!", "accuracy", accuracy)
+
 	os.RemoveAll(run_path)
 	return
 }
@@ -286,12 +287,11 @@ func trainModel(c *Context, model *BaseModel) {
 				continue
 			}
 			def.epoch += EPOCH_PER_RUN
+			accuracy = accuracy * 100
 
-			int_accuracy := int(accuracy * 100)
+			if accuracy >= float64(def.target_accuracy) {
-
-			if int_accuracy >= def.target_accuracy {
 				c.Logger.Info("Found a definition that reaches target_accuracy!")
-				_, err = c.Db.Exec("update model_definition set accuracy=$1, status=$2, epoch=$3 where id=$4", int_accuracy, MODEL_DEFINITION_STATUS_TRANIED, def.epoch, def.id)
+				_, err = c.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 {
 					c.Logger.Error("Failed to train definition!Err:\n", "err", err)
 					ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
@@ -310,13 +310,19 @@ func trainModel(c *Context, model *BaseModel) {
 			}
 
 			if def.epoch > MAX_EPOCH {
-				fmt.Printf("Failed to train definition! Accuracy less %d < %d\n", int_accuracy, def.target_accuracy)
+				fmt.Printf("Failed to train definition! Accuracy less %f < %d\n", accuracy, def.target_accuracy)
 				ModelDefinitionUpdateStatus(c, def.id, MODEL_DEFINITION_STATUS_FAILED_TRAINING)
 				toTrain = toTrain - 1
 				newDefinitions = remove(newDefinitions, i)
 				continue
 			}
 
+            _, err = c.Db.Exec("update model_definition set accuracy=$1, epoch=$2 where id=$3", accuracy, def.epoch, def.id)
+            if err != nil {
+                c.Logger.Error("Failed to train definition!Err:\n", "err", err)
+                ModelUpdateStatus(c, model.Id, FAILED_TRAINING)
+                return
+            }
 		}
 		copy(definitions, newDefinitions)
 		firstRound = false
@@ -403,7 +409,7 @@ func removeFailedDataPoints(c *Context, model *BaseModel) (err error) {
 			return
 		}
 
-        p := path.Join(base_path, dataPointId + "." + model.Format)
+		p := path.Join(base_path, dataPointId+"."+model.Format)
 
 		c.Logger.Warn("Removing image", "path", p)
 
@@ -418,7 +424,7 @@ func removeFailedDataPoints(c *Context, model *BaseModel) (err error) {
 }
 
 // This generates a definition
-func generateDefinition(c *Context, model *BaseModel, number_of_classes int,  complexity int) *Error {
+func generateDefinition(c *Context, model *BaseModel, target_accuracy int, number_of_classes int, complexity int) *Error {
 	var err error = nil
 	failed := func() *Error {
 		ModelUpdateStatus(c, model.Id, FAILED_PREPARING_TRAINING)
@@ -426,28 +432,32 @@ func generateDefinition(c *Context, model *BaseModel, number_of_classes int,  co
 		return c.Error500(err)
 	}
 
-
+	def_id, err := MakeDefenition(c.Db, model.Id, target_accuracy)
-	def_id, err := MakeDefenition(c.Db, model.Id, 0)
 	if err != nil {
 		return failed()
 	}
 
+    order := 1;
+
 	// Note the shape for now is no used
-    err = MakeLayer(c.Db, def_id, 1, LAYER_INPUT, fmt.Sprintf("%d,%d,1", model.Width, model.Height))
+	err = MakeLayer(c.Db, def_id, order, LAYER_INPUT, fmt.Sprintf("%d,%d,1", model.Width, model.Height))
 	if err != nil {
 		return failed()
 	}
+    order++;
 
 	if complexity == 0 {
 
-		err = MakeLayer(c.Db, def_id, 4, LAYER_FLATTEN, "")
+		err = MakeLayer(c.Db, def_id, order, LAYER_FLATTEN, "")
 		if err != nil {
 			return failed()
 		}
+        order++;
 
-        loop := int(math.Log2(float64(number_of_classes)))
+		loop := int(math.Log2(float64(number_of_classes))/2)
 		for i := 0; i < loop; i++ {
-            err = MakeLayer(c.Db, def_id, 5, LAYER_DENSE, fmt.Sprintf("%d,1", number_of_classes*(loop - i)))
+			err = MakeLayer(c.Db, def_id, order, LAYER_DENSE, fmt.Sprintf("%d,1", number_of_classes*(loop-i)))
+            order++;
 			if err != nil {
 				ModelUpdateStatus(c, model.Id, FAILED_PREPARING_TRAINING)
 				// TODO improve this response
@@ -455,6 +465,38 @@ func generateDefinition(c *Context, model *BaseModel, number_of_classes int,  co
 			}
 		}
 
+    } else if (complexity == 1) {
+
+		loop := int((math.Log(float64(model.Width))/math.Log(float64(10))))
+        if loop == 0 {
+            loop = 1;
+        }
+		for i := 0; i < loop; i++ {
+			err = MakeLayer(c.Db, def_id, order, LAYER_SIMPLE_BLOCK, "")
+            order++;
+			if err != nil {
+                return failed();
+			}
+		}
+
+		err = MakeLayer(c.Db, def_id, order, LAYER_FLATTEN, "")
+		if err != nil {
+			return failed()
+		}
+        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 = MakeLayer(c.Db, def_id, order, LAYER_DENSE, fmt.Sprintf("%d,1", number_of_classes*(loop-i)))
+            order++;
+			if err != nil {
+                return failed();
+			}
+		}
+
 	} else {
 		c.Logger.Error("Unkown complexity", "complexity", complexity)
 		return failed()
@@ -468,7 +510,7 @@ func generateDefinition(c *Context, model *BaseModel, number_of_classes int,  co
 	return nil
 }
 
-func generateDefinitions(c *Context, model *BaseModel, number_of_models int) *Error {
+func generateDefinitions(c *Context, model *BaseModel, target_accuracy int, number_of_models int) *Error {
 	cls, err := model_classes.ListClasses(c.Db, model.Id)
 	if err != nil {
 		ModelUpdateStatus(c, model.Id, FAILED_PREPARING_TRAINING)
@@ -481,10 +523,19 @@ func generateDefinitions(c *Context, model *BaseModel, number_of_models int) *Er
 		return c.Error500(err)
 	}
 
-	for i := 0; i < number_of_models; i++ {
+    if (number_of_models == 1) {
-        // TODO handle incrisea the complexity
+        if (model.Width < 100 && model.Height < 100 && len(cls) < 30) {
-        generateDefinition(c, model, len(cls), 0)
+            generateDefinition(c, model, target_accuracy, len(cls), 0)
+        } else {
+            generateDefinition(c, model, target_accuracy, len(cls), 1)
         }
+    } else {
+        // TODO handle incrisea the complexity
+        for i := 0; i < number_of_models; i++ {
+            generateDefinition(c, model, target_accuracy, len(cls), 0)
+        }
+    }
+
 
 	return nil
 }
@@ -551,7 +602,7 @@ func handleTrain(handle *Handle) {
 			return ErrorCode(nil, 400, c.AddMap(nil))
 		}
 
-        full_error := generateDefinitions(c, model, number_of_models)
+		full_error := generateDefinitions(c, model, accuracy, number_of_models)
 		if full_error != nil {
 			return full_error
 		}
@@ -573,11 +624,15 @@ func handleTrain(handle *Handle) {
 
 		f := r.URL.Query()
 
-		if !CheckId(f, "model_id") || !CheckId(f, "definition") || CheckEmpty(f, "epoch") {
+        accuracy := 0.0
-			c.Logger.Warn("Invalid: model_id or definition or epoch")
+
+		if !CheckId(f, "model_id") || !CheckId(f, "definition") || CheckEmpty(f, "epoch") || !CheckFloat64(f, "accuracy", &accuracy){
+			c.Logger.Warn("Invalid: model_id or definition or epoch or accuracy")
 			return c.UnsafeErrorCode(nil, 400, nil)
 		}
 
+        accuracy  = accuracy * 100
+
 		model_id := f.Get("model_id")
 		def_id := f.Get("definition")
 		epoch, err := strconv.Atoi(f.Get("epoch"))
@@ -610,7 +665,9 @@ func handleTrain(handle *Handle) {
 			return c.UnsafeErrorCode(nil, 400, nil)
 		}
 
-		_, err = c.Db.Exec("update model_definition set epoch_progress=$1 where id=$2", epoch, def_id)
+        c.Logger.Info("Updated model_definition!", "model", model_id, "progress", epoch, "accuracy", accuracy)
+
+		_, err = c.Db.Exec("update model_definition set epoch_progress=$1, accuracy=$2 where id=$3", epoch, accuracy, def_id)
 		if err != nil {
 			return c.Error500(err)
 		}

logic/utils/utils.go

@@ -31,6 +31,21 @@ func CheckNumber(f url.Values, path string, number *int) bool {
     return true
 }
 
+func CheckFloat64(f url.Values, path string, number *float64) bool {
+    if CheckEmpty(f, path) {
+        fmt.Println("here", path)
+        fmt.Println(f.Get(path))
+        return false
+    }
+    n, err := strconv.ParseFloat(f.Get(path), 64)
+    if err != nil {
+        fmt.Println(err)
+        return false
+    }
+    *number = n
+    return true
+}
+
 func CheckId(f url.Values, path string) bool {
     return !CheckEmpty(f, path) && IsValidUUID(f.Get(path))
 }

sql/models.sql

@@ -44,7 +44,7 @@ create table if not exists model_data_point (
 create table if not exists model_definition (
     id uuid primary key default gen_random_uuid(),
     model_id uuid references models (id) on delete cascade,
-    accuracy integer default 0,
+    accuracy real default 0,
     target_accuracy integer not null,
     epoch integer default 0,
     -- TODO add max epoch

views/py/python_model_template.py

@@ -7,10 +7,8 @@ from keras import layers, losses, optimizers
 import requests
 
 class NotifyServerCallback(tf.keras.callbacks.Callback):
-    def on_epoch_begin(self, epoch, *args, **kwargs):
+    def on_epoch_end(self, epoch, log, *args, **kwargs):
-        if (epoch % 5) == 0:
+        requests.get(f'http://localhost:8000/model/epoch/update?model_id={{.Model.Id}}&epoch={epoch}&accuracy={log["accuracy"]}&definition={{.DefId}}')
-            # TODO change this
-            requests.get(f'http://localhost:8000/model/epoch/update?model_id={{.Model.Id}}&epoch={epoch}&definition={{.DefId}}')
 
 
 DATA_DIR = "{{ .DataDir }}"
@@ -84,7 +82,7 @@ def filterDataset(path):
 
 seed = random.randint(0, 100000000)
 
-batch_size = 100
+batch_size = 64
 
 # Read all the files from the direcotry
 list_ds = tf.data.Dataset.list_files(str(f'{DATA_DIR}/*'), shuffle=False)
@@ -102,23 +100,55 @@ val_ds = list_ds.take(val_size)
 dataset = prepare_dataset(train_ds)
 dataset_validation   = prepare_dataset(val_ds)
 
+track = 0
+
+def addBlock(
+    b_size: int,
+    filter_size: int,
+    kernel_size: int = 3,
+    top: bool = True,
+    pooling_same: bool = False,
+    pool_func=layers.MaxPool2D
+):
+    global track
+    model = keras.Sequential(name=f"{track}-{b_size}-{filter_size}-{kernel_size}")
+    track += 1
+    for _ in range(b_size):
+        model.add(layers.Conv2D(
+            filter_size,
+            kernel_size,
+            padding="same"
+        ))
+        model.add(layers.ReLU())
+    if top:
+        if pooling_same:
+            model.add(pool_func(padding="same", strides=(1, 1)))
+        else:
+            model.add(pool_func())
+        model.add(layers.BatchNormalization())
+        model.add(layers.LeakyReLU())
+        model.add(layers.Dropout(0.4))
+    return model
+
 
 {{ if .LoadPrev }}
 model = tf.keras.saving.load_model('{{.LastModelRunPath}}')
 {{ else }}
-model = keras.Sequential([
+model = keras.Sequential()
-    {{- range .Layers }}
+
-    {{- if eq .LayerType 1}}
+{{- range .Layers }}
-    layers.Rescaling(1./255),
+{{- if eq .LayerType 1}}
-    {{- else if eq .LayerType 2 }}
+model.add(layers.Rescaling(1./255))
-    layers.Dense({{ .Shape }}, activation="sigmoid"),
+{{- else if eq .LayerType 2 }}
-    {{- else if eq .LayerType 3}}
+model.add(layers.Dense({{ .Shape }}, activation="sigmoid"))
-    layers.Flatten(),
+{{- else if eq .LayerType 3}}
-    {{- else }}
+model.add(layers.Flatten())
-    ERROR
+{{- else if eq .LayerType 4}}
-    {{- end }}
+model.add(addBlock(2, 128, 3, pool_func=layers.AveragePooling2D))
-    {{- end }}
+{{- else }}
-])
+ERROR
+{{- end }}
+{{- end }}
 {{ end }}
 
 model.compile(
@@ -126,7 +156,7 @@ model.compile(
     optimizer=tf.keras.optimizers.Adam(),
     metrics=['accuracy'])
 
-his = model.fit(dataset, validation_data= dataset_validation, epochs={{.EPOCH_PER_RUN}}, callbacks=[NotifyServerCallback()])
+his = model.fit(dataset, validation_data= dataset_validation, epochs={{.EPOCH_PER_RUN}}, callbacks=[NotifyServerCallback()], use_multiprocessing = True)
 
 acc = his.history["accuracy"]