Problem In Building The Model

Dear Experts,

I tried to build a model similar to (CIFAR-10), For details about My Project Please refer to the following link: https://www.kaggle.com/c/state-farm-distracted-driver-detection

I depict the similar Model in (CIFAR-10) but I got an error after building the model:

(TypeError : can only concatenate tuple (not “dict”) to tuple)

In addition, since I’m still a beginner in building deep learning model the (Wizard feature) is not working on the Desktop Version, attached screenshots about the errors I encounter it in building the model and using Wizard Feature.

The Model Code as the following:
data:
dataset: {name: train, samples: 22424, type: private}
datasetLoadOption: batch
kfold: 1
mapping:
Image:
options: {}
port: ‘’
shape: ‘’
type: Categorical
Rating:
options: {}
port: ‘’
shape: ‘’
type: Categorical
numPorts: 1
samples: {split: 3, test: 3363, training: 15696, validation: 3363}
shuffle: true
model:
connections:

  • {source: Input_1, target: BatchNormalization_1}
  • {source: Dropout_2, target: Flatten_1}
  • {source: Dropout_3, target: Dense_2}
  • {source: MaxPooling2D_1, target: Dropout_1}
  • {source: Convolution2D_3, target: MaxPooling2D_2}
  • {source: MaxPooling2D_2, target: Dropout_2}
  • {source: Convolution2D_2, target: Convolution2D_3}
  • {source: Convolution2D_4, target: MaxPooling2D_1}
  • {source: Dropout_1, target: Convolution2D_2}
  • {source: Dense_2, target: Output_1}
  • {source: Dense_1, target: Dropout_3}
  • {source: Flatten_1, target: Dense_1}
    layers:
  • args: {dim_ordering: th}
    class: Input
    name: Input_1
    x: 10
    y: 50
  • args: {}
    class: Output
    name: Output_1
    x: 824.4
    y: 615.88
  • args: {activation: relu, border_mode: same, nb_col: ‘3’, nb_filter: ‘32’, nb_row: ‘3’}
    class: Convolution2D
    name: Convolution2D_1
    x: 9
    y: 304
  • args: {axis: ‘1’}
    class: BatchNormalization
    name: BatchNormalization_1
    x: 12
    y: 183
  • args: {activation: relu, border_mode: same, nb_col: ‘3’, nb_filter: ‘64’, nb_row: ‘3’}
    class: Convolution2D
    name: Convolution2D_2
    x: 415.24
    y: 54.019999999999996
  • args: {activation: relu, nb_col: ‘3’, nb_filter: ‘64’, nb_row: ‘3’}
    class: Convolution2D
    name: Convolution2D_3
    x: 429.52
    y: 180
  • args: {activation: relu, nb_col: ‘3’, nb_filter: ‘32’, nb_row: ‘3’}
    class: Convolution2D
    name: Convolution2D_4
    x: 2
    y: 432
  • args: {}
    class: MaxPooling2D
    name: MaxPooling2D_1
    x: 11
    y: 559
  • args: {p: ‘0.25’}
    class: Dropout
    name: Dropout_1
    x: 13
    y: 662
  • args: {}
    class: MaxPooling2D
    name: MaxPooling2D_2
    x: 442.8
    y: 330
  • args: {p: ‘0.25’}
    class: Dropout
    name: Dropout_2
    x: 426.88
    y: 532.08
  • args: {}
    class: Flatten
    name: Flatten_1
    x: 846.6999999999999
    y: 91.12
  • args: {activation: relu, output_dim: ‘512’}
    class: Dense
    name: Dense_1
    x: 842.12
    y: 217.38000000000002
  • args: {p: ‘0.5’}
    class: Dropout
    name: Dropout_3
    x: 839.36
    y: 329.12
  • args: {activation: softmax, output_dim: ‘10’}
    class: Dense
    name: Dense_2
    x: 823.3
    y: 481.72
    params:
    advance_params: true
    batch_size: 32
    is_custom_loss: false
    loss_func: categorical_crossentropy
    num_epoch: 10
    optimizer: {decay: 1e-6, momentum: ‘0.9’, name: SGD, nesterov: true}
    project: Driver

As per the following your model configuration, you need to configure type and port information correctly on the data tab.

Image:
options: {}
port: ‘’
shape: ‘’
type: Categorical
Rating:
options: {}
port: ‘’
shape: ‘’
type: Categorical
  1. choose Image as the Data Type for the ‘Image’ column and select port as InputPort0

  2. choose OutputPort0 for the ‘Rating’ column.

That’s what I did already

Choose “Image” from the DataType dropdown for the “Image” column.

But I note it’s not save the steps automatically , I think need some documentation on how to save the steps

Ok Sorry , I did it works now

But what about the Wizard Feature ? Why It is not working on the Desktop model?

Now Everything ok but when we run it after a while the machine halted (stop responding) , I have 8 GHZ ram , do you think it is enough for the case of my project? If not do you recommends going to the cloud version?

You should be able to train on your system. You may want to reduce your batch size (under hyper parameters). You can also reduce your image size if it is big by choosing the resize option in data tab.

For resize option, click on the Image column, and see resize option in the right side panel (data tab).

Sorry Again how can we resize the Image In the data type?

If reduced shall we Increase the number of Convolution Layers I mean (revers fit) relationship?