Coverage for source/model/model_blue_prints/dnn_blue_print.py: 23%

1# model/model_blue_prints/dnn_blue_print.py 

2 

3# global imports 

4import math 

5from tensorflow.keras import layers, Model 

6from typing import Optional 

7 

8# local imports 

9from source.model import BluePrintBase, ModelAdapterBase, TFModelAdapter 

10 

11class DnnBluePrint(BluePrintBase): 

12 """ 

13 Blueprint for creating a DNN. 

14 

15 This class implements a model blueprint that constructs a neural network using 

16 fully connected (dense) layers. Expected input is a 1D vector of features. 

17 """ 

18 

19 def __init__(self, dense_squeezing_coeff: int = 2, dense_repetition_coeff: int = 1) -> None: 

20 """ 

21 Initializes the DnnBluePrint with the specified configuration parameters. 

22 

23 Parameters: 

24 dense_squeezing_coeff (int): Factor by which dense layer sizes are reduced. 

25 dense_repetition_coeff (int): Number of dense layers of the same size to use. 

26 """ 

27 

28 self.__dense_squeezing_coeff = dense_squeezing_coeff 

29 self.__dense_repetition_coeff = dense_repetition_coeff 

30 

31 def instantiate_model(self, input_shape: tuple[int, int], output_length: int, 

32 dense_squeezing_coeff: Optional[int] = None, dense_repetition_coeff: Optional[int] = None) -> ModelAdapterBase: 

33 """ 

34 Creates and returns a DNN model according to specified parameters. 

35 

36 The method constructs a neural network that: 

37 1. Processes a 1D input vector through multiple dense layers 

38 2. Produces a softmax output for classification 

39 

40 Parameters: 

41 input_shape (tuple[int, int]): Shape of the input tensor 

42 output_length (int): Number of output classes/actions 

43 dense_squeezing_coeff (int): Factor by which dense layer sizes are reduced 

44 dense_repetition_coeff (int): Number of dense layers of the same size to use 

45 

46 Returns: 

47 Model: Keras model implementing the DNN architecture to be compiled further. 

48 """ 

49 

50 if dense_squeezing_coeff is None: 

51 dense_squeezing_coeff = self.__dense_squeezing_coeff 

52 if dense_repetition_coeff is None: 

53 dense_repetition_coeff = self.__dense_repetition_coeff 

54 

55 input_vector = layers.Input((1, input_shape[0])) 

56 reshaped_input_vector = layers.Reshape((input_shape[0],))(input_vector) 

57 

58 closest_smaller_power_of_coeff = int(math.pow(dense_squeezing_coeff, 

59 int(math.log(reshaped_input_vector.shape[-1], 

60 dense_squeezing_coeff)))) 

61 dense = layers.Dense(closest_smaller_power_of_coeff, activation='relu')(reshaped_input_vector) 

62 dense = layers.BatchNormalization()(dense) 

63 

64 number_of_nodes = closest_smaller_power_of_coeff // dense_squeezing_coeff 

65 nr_of_dense_layers = int(math.log(closest_smaller_power_of_coeff, dense_squeezing_coeff)) 

66 for _ in range(nr_of_dense_layers): 

67 for _ in range(dense_repetition_coeff): 

68 dense = layers.Dense(number_of_nodes, activation='relu')(dense) 

69 dense = layers.BatchNormalization()(dense) 

70 number_of_nodes //= dense_squeezing_coeff 

71 if int(math.log(number_of_nodes, 10)) == int(math.log(output_length, 10)) + 1: 

72 dense = layers.Dropout(0.3)(dense) 

73 elif int(math.log(number_of_nodes, 10)) == int(math.log(output_length, 10)): 

74 break 

75 

76 output = layers.Dense(output_length, activation='softmax')(dense) 

77 

78 return TFModelAdapter(Model(inputs = input_vector, outputs = output))