fix: removed keras and pytorch interface import from __init__.py

src/qiboml/__init__.py

@@ -1,6 +1,5 @@
 import importlib.metadata as im
 
 from qiboml.backends.__init__ import MetaBackend
-from qiboml.models import keras, pytorch
 
 __version__ = im.version(__package__)

src/qiboml/models/abstract.py

@@ -7,7 +7,7 @@
 from qibo.config import raise_error
 from qibo.gates import abstract
 
-from qiboml.backends import TensorflowBackend as JaxBackend
+from qiboml.backends import JaxBackend
 
 
 @dataclass

src/qiboml/models/keras.py

@@ -14,25 +14,25 @@
 @dataclass
 class QuantumModel(keras.layers.Layer):  # pylint: disable=no-member
 
-    def __init__(self, layers: list[QuantumCircuitLayer]):
-        super().__init__()
-        nqubits = layers[0].circuit.nqubits
-        self.layers = layers
-        for layer in layers[1:]:
-            if layer.circuit.nqubits != nqubits:
+    layers: list[QuantumCircuitLayer]
+
+    def __post_init__(self):
+        super().__post_init__()
+        for layer in self.layers[1:]:
+            if layer.circuit.nqubits != self.nqubits:
                 raise_error(
                     RuntimeError,
-                    f"Layer \n{layer}\n has {layer.circuit.nqubits} qubits, but {nqubits} qubits was expected.",
+                    f"Layer \n{layer}\n has {layer.circuit.nqubits} qubits, but {self.nqubits} qubits was expected.",
                 )
             if layer.backend.name != self.backend.name:
                 raise_error(
                     RuntimeError,
                     f"Layer \n{layer}\n is using {layer.backend} backend, but {self.backend} backend was expected.",
                 )
-        if not isinstance(layers[-1], ed.QuantumDecodingLayer):
+        if not isinstance(self.layers[-1], ed.QuantumDecodingLayer):
             raise_error(
                 RuntimeError,
-                f"The last layer has to be a `QuantumDecodinglayer`, but is {layers[-1]}",
+                f"The last layer has to be a `QuantumDecodinglayer`, but is {self.layers[-1]}",
             )
 
     def call(self, x: tf.Tensor) -> tf.Tensor:

src/qiboml/models/pytorch.py

@@ -13,10 +13,11 @@
 @dataclass
 class QuantumModel(torch.nn.Module):
 
-    def __init__(self, layers: list[QuantumCircuitLayer]):
-        super().__init__()
-        self.layers = layers
-        for layer in layers[1:]:
+    layers: list[QuantumCircuitLayer]
+
+    def __post_init__(self):
+        super().__post_init__()
+        for layer in self.layers[1:]:
             if layer.circuit.nqubits != self.nqubits:
                 raise_error(
                     RuntimeError,
@@ -27,16 +28,16 @@ def __init__(self, layers: list[QuantumCircuitLayer]):
                     RuntimeError,
                     f"Layer \n{layer}\n is using {layer.backend} backend, but {self.backend} backend was expected.",
                 )
-        for layer in layers:
+        for layer in self.layers:
             if len(layer.circuit.get_parameters()) > 0:
                 self.register_parameter(
                     layer.__class__.__name__,
                     torch.nn.Parameter(torch.as_tensor(layer.circuit.get_parameters())),
                 )
-        if not isinstance(layers[-1], ed.QuantumDecodingLayer):
+        if not isinstance(self.layers[-1], ed.QuantumDecodingLayer):
             raise_error(
                 RuntimeError,
-                f"The last layer has to be a `QuantumDecodinglayer`, but is {layers[-1]}",
+                f"The last layer has to be a `QuantumDecodinglayer`, but is {self.layers[-1]}",
             )
 
     def forward(self, x: torch.Tensor):

tests/conftest.py

@@ -38,10 +38,14 @@ def get_backend(backend_name):
 
 
 def get_frontend(frontend_name):
-    import qiboml
-
-    frontend = getattr(qiboml, frontend_name)
-    setattr(frontend, "__str__", frontend_name)
+    from qiboml.models import keras, pytorch
+
+    if frontend_name == "keras":
+        frontend = keras
+    elif frontend_name == "pytorch":
+        frontend = pytorch
+    else:
+        raise RuntimeError(f"Unknown frontend {frontend_name}.")
     return frontend