hqm.regression.hmlp

  1import sys
  2sys.path += ['.', './layers/']
  3
  4from hqm.layers.basiclayer import BasicLayer
  5import torch
  6
  7
  8class BasicHybridMLPRegressor(torch.nn.Module):
  9    '''
 10        This class implements a basic hybrid multilayer perceptron for regression purposes.
 11        BasicHybridMLPRegressor is composed of quantum layers stacked between two fully connected layers.
 12        The size of fully connected layers is set by means of in_dim and ou_dim.
 13    '''
 14
 15    def __init__(self, qlayer : BasicLayer, in_dim : int, ou_dim : int) -> None:
 16        '''
 17            BasicHybridMLPRegressor constructor.  
 18
 19            Parameters:  
 20            -----------  
 21            - qlayer : hqm.layers.basiclayer.BasicLayer
 22                hqm quantum layer to be stacked between two fully connected layers  
 23            - in_dim : int  
 24                integer representing the input size for the first fully connected layer  
 25            - ou_dim : int  
 26                integer representing the output size of the hybrid model  
 27            
 28            Returns:  
 29            --------  
 30            Nothing, a BasicHybridMLPRegressor object will be created.    
 31        '''
 32
 33        super().__init__()
 34
 35        if in_dim < 1: raise Exception(f"The parameter in_dim must be greater than 1, found {in_dim}")
 36        if ou_dim < 1: raise Exception(f"The parameter ou_dim must be greater than 1, found {ou_dim}")
 37
 38        n_qubits  = qlayer.n_qubits
 39        self.fc_1 = torch.nn.Linear(in_dim, n_qubits)
 40        self.qc_1 = qlayer.qlayer
 41        self.fc_2 = torch.nn.Linear(n_qubits, ou_dim)
 42        self.tanh = torch.nn.Tanh()
 43
 44    def forward(self, x : torch.Tensor) -> torch.Tensor:
 45        ''' 
 46            Torch forward method  
 47
 48            Parameters:  
 49            -----------  
 50            - x : torch.Tensor  
 51                input for the torch model  
 52
 53            Returns:  
 54            --------  
 55            - out : torch.Tensor  
 56                output from the torch model  
 57        '''
 58
 59        x = self.fc_1(x)
 60        x = self.tanh(x)
 61        x = self.qc_1(x)
 62        x = self.tanh(x)
 63        x = self.fc_2(x)
 64        out = self.tanh(x)
 65        return out
 66    
 67class MultiHybridMLPRegressor(torch.nn.Module):
 68    '''
 69        This class implements a hybrid multilayer perceptron with multiple quantum circuits for regression purposes.
 70        MultiHybridMLPRegressor is composed of several quantum layers stacked between two fully connected layers.
 71        The size of fully connected layers is set by means of in_dim and ou_dim.
 72    '''
 73
 74    def __init__(self, qlayers : list, in_dim : int, ou_dim : int) -> None:
 75        '''
 76            MultiHybridMLPRegressor constructor.  
 77
 78            Parameters:  
 79            -----------  
 80            - qlayer : list  
 81                list of hqm quantum layers to be stacked between two fully connected layers  
 82            - in_dim : int  
 83                integer representing the input size for the first fully connected layer  
 84            - ou_dim : int  
 85                integer representing the output size of the hybrid model  
 86            
 87            Returns:  
 88            --------  
 89            Nothing, a MultiHybridMLPRegressor object will be created.    
 90        '''
 91
 92        super().__init__()
 93
 94        if in_dim < 1: raise Exception(f"The parameter in_dim must be greater than 1, found {in_dim}")
 95        if ou_dim < 1: raise Exception(f"The parameter ou_dim must be greater than 1, found {ou_dim}")
 96        if len(qlayers) < 1: raise Exception(f"Size of qlayers must be greater than 1, found {len(qlayers)}")
 97
 98        n_qubits_0 = qlayers[0].n_qubits
 99        n_qubits_1 = qlayers[-1].n_qubits
100        self.fc_1  = torch.nn.Linear(in_dim, n_qubits_0)
101        self.qcs   = [circ.qlayer for circ in qlayers]
102        self.fc_2  = torch.nn.Linear(n_qubits_1, ou_dim)
103        self.tanh  = torch.nn.Tanh()
104
105    def forward(self, x : torch.Tensor) -> torch.Tensor:
106        '''
107            Torch forward method  
108
109            Parameters:  
110            -----------  
111            - x : torch.Tensor  
112                input for the torch model  
113
114            Returns:  
115            --------  
116            - out : torch.Tensor  
117                output from the torch model  
118        '''
119
120        x = self.fc_1(x)
121        x = self.tanh(x)
122        for qc in self.qcs: 
123            x = qc(x)
124            x = self.tanh(x)
125        x = self.fc_2(x)
126        out = self.tanh(x)
127        return out
128
129class MultiHybridMultiMLPRegressor(torch.nn.Module):
130    '''
131        This class implements a hybrid multilayer perceptron with multiple quantum circuits for regression purposes.
132        MultiHybridMultiMLPRegressor is composed of several quantum layers stacked between alternating fully connected layers.
133        The size of fully connected layers is set by means of in_dim and ou_dim.
134    '''
135
136    def __init__(self, qlayers : list, in_dims : list, ou_dim : list) -> None:
137        '''
138            MultiHybridMultiMLPRegressor constructor.  
139
140            Parameters:  
141            -----------  
142            - qlayers : list  
143                list of hqm quantum qlayerss to be stacked between two fully connected layers  
144            - in_dims: list  
145                list of integers representing the input size for the i-th fully connected layer (first value should correspond to size of input data)  
146            - ou_dim : list  
147                list of integers representing the output size for the i-th fully connected layer (last value should correspond to desired output size)  
148            
149            Returns:  
150            --------  
151            Nothing, a MultiHybridMLPRegressor object will be created.    
152        '''
153        
154        super().__init__()
155
156        if len(in_dims) < 1: raise Exception(f"Size in_dims must be greater than 1, found {len(in_dims)}")
157        if ou_dim < 1: raise Exception(f"The parameter ou_dim must be greater than 1, found {ou_dim}")
158        if len(qlayers) < 1: raise Exception(f"Size of qlayers must be greater than 1, found {len(qlayers)}")
159        if len(qlayers) != len(in_dims): raise Exception(f"qlayers and in_dims must have the same lenght, found {len(qlayers)} and {len(in_dims)}")
160        for i, dim in enumerate(in_dims): 
161            if dim < 1: raise Exception(f"Element {i} of in_dims must be greater than 1, found {dim}")
162            else: pass
163        
164        self.fcs  = [torch.nn.Linear(dim, circ.n_qubits) for (dim, circ) in zip(in_dims, qlayers)]
165        self.fco  = torch.nn.Linear(qlayers[-1].n_qubits, ou_dim)
166        self.qcs  = [circ.qlayer for circ in qlayers]
167        self.tanh = torch.nn.Tanh()
168
169    def forward(self, x : torch.Tensor) -> torch.Tensor:
170        '''
171            Torch forward method  
172
173            Parameters:  
174            -----------  
175            - x : torch.Tensor  
176                input for the torch model  
177
178            Returns:  
179            --------  
180            - x : torch.Tensor  
181                output from the torch model  
182        '''
183        
184        for fc, qc in zip(self.fcs, self.qcs):
185            x = fc(x)
186            x = self.tanh(x)
187            x = qc(x)
188            x = self.tanh(x)
189        
190        x   = self.fco(x) 
191        out = self.tanh(x)
192        return out