unaiverse 0.1.6__cp314-cp314-macosx_10_15_x86_64.whl

unaiverse/streamlib/streamlib.py

@@ -0,0 +1,210 @@
+"""
+       █████  █████ ██████   █████           █████ █████   █████ ██████████ ███████████    █████████  ██████████
+      ░░███  ░░███ ░░██████ ░░███           ░░███ ░░███   ░░███ ░░███░░░░░█░░███░░░░░███  ███░░░░░███░░███░░░░░█
+       ░███   ░███  ░███░███ ░███   ██████   ░███  ░███    ░███  ░███  █ ░  ░███    ░███ ░███    ░░░  ░███  █ ░ 
+       ░███   ░███  ░███░░███░███  ░░░░░███  ░███  ░███    ░███  ░██████    ░██████████  ░░█████████  ░██████   
+       ░███   ░███  ░███ ░░██████   ███████  ░███  ░░███   ███   ░███░░█    ░███░░░░░███  ░░░░░░░░███ ░███░░█   
+       ░███   ░███  ░███  ░░█████  ███░░███  ░███   ░░░█████░    ░███ ░   █ ░███    ░███  ███    ░███ ░███ ░   █
+       ░░████████   █████  ░░█████░░████████ █████    ░░███      ██████████ █████   █████░░█████████  ██████████
+        ░░░░░░░░   ░░░░░    ░░░░░  ░░░░░░░░ ░░░░░      ░░░      ░░░░░░░░░░ ░░░░░   ░░░░░  ░░░░░░░░░  ░░░░░░░░░░ 
+                 A Collectionless AI Project (https://collectionless.ai)
+                 Registration/Login: https://unaiverse.io
+                 Code Repositories:  https://github.com/collectionlessai/
+                 Main Developers:    Stefano Melacci (Project Leader), Christian Di Maio, Tommaso Guidi
+"""
+import math
+import torch
+from unaiverse.streams import BufferedDataStream, DataProps
+
+
+class AllHotLabelStream(BufferedDataStream):
+    """
+    A buffered stream that simply repeat a single-element tensor valued "ones" (float), associated to some text labels
+    """
+
+    def __init__(self, labels: list[str],
+                 device: torch.device = torch.device('cpu')):
+        super().__init__(props=DataProps(name=AllHotLabelStream.__name__,
+                                         data_type="tensor",
+                                         data_desc="dummy stream",
+                                         tensor_shape=(1, len(labels)),
+                                         tensor_labels=labels,
+                                         tensor_dtype=str(torch.float),
+                                         tensor_labeling_rule="geq0.5"),
+                         is_static=True)
+
+        self.set(torch.ones((1, len(labels)), dtype=torch.float32, device=device))
+        self.restart()
+
+
+class Random(BufferedDataStream):
+
+    def __init__(self, std: float, shape: tuple[int] | None = (1,),
+                 device: torch.device = torch.device('cpu')):
+        super().__init__(props=DataProps(name="rand",
+                                         data_type="tensor",
+                                         data_desc="stream of random numbers",
+                                         tensor_shape=shape,
+                                         tensor_dtype=str(torch.float)))
+        self.std = std
+        self.device = device
+        self.restart()
+
+    def __getitem__(self, idx: int) -> torch.Tensor | None:
+        y = self.std * torch.rand(self.props.tensor_shape, device=self.device)
+        return self.props.adapt_tensor_to_tensor_labels(y), self.clock.get_cycle() - self.first_cycle
+
+
+class Sin(BufferedDataStream):
+
+    def __init__(self, freq: float, phase: float, delta: float,
+                 device: torch.device = torch.device('cpu')):
+        super().__init__(props=DataProps(name="sin",
+                                         data_type="tensor",
+                                         data_desc="stream of samples from the sin function",
+                                         tensor_shape=(1, 1),
+                                         tensor_dtype=str(torch.float)))
+        self.freq = freq
+        self.phase = phase
+        self.period = 1. / self.freq
+        self.delta = delta
+        self.device = device
+        self.restart()
+
+    def __getitem__(self, idx: int) -> torch.Tensor | None:
+        t = idx * self.delta + self.phase * self.period
+        y = torch.sin(torch.tensor([[2. * math.pi * self.freq * t]], device=self.device))
+        return self.props.adapt_tensor_to_tensor_labels(y), self.clock.get_cycle() - self.first_cycle
+
+
+class Square(BufferedDataStream):
+
+    def __init__(self, freq: float, ampl: float, phase: float, delta: float,
+                 device: torch.device = torch.device('cpu')):
+        super().__init__(props=DataProps(name="square",
+                                         data_type="tensor",
+                                         data_desc="stream of samples from the square function",
+                                         tensor_shape=(1, 1),
+                                         tensor_dtype=str(torch.float)))
+        self.freq = freq
+        self.ampl = ampl
+        self.phase = phase
+        self.period = 1. / self.freq
+        self.delta = delta
+        self.device = device
+        self.restart()
+
+    def __getitem__(self, idx: int) -> torch.Tensor | None:
+        t = idx * self.delta + self.phase * self.period
+        y = self.ampl * torch.tensor([[(-1.) ** (math.floor(2. * self.freq * t))]], device=self.device)
+        return self.props.adapt_tensor_to_tensor_labels(y), self.clock.get_cycle() - self.first_cycle
+
+
+class CombSin(BufferedDataStream):
+
+    def __init__(self, f_cap: float | list, c_cap: float | list, order: int, delta: float,
+                 device: torch.device = torch.device('cpu')):
+        super().__init__(props=DataProps(name="combsin",
+                                         data_type="tensor",
+                                         data_desc="stream of samples from combined sin functions",
+                                         tensor_shape=(1, 1),
+                                         tensor_dtype=str(torch.float)))
+        if isinstance(f_cap, float):
+            self.freqs = f_cap * torch.rand(order)
+        elif isinstance(f_cap, list):
+            self.freqs = torch.tensor(f_cap)
+        else:
+            raise Exception(f"expected float or list for f_cap, not {type(f_cap)}")
+        self.phases = torch.zeros_like(self.freqs)
+        if isinstance(c_cap, float):
+            self.coeffs = c_cap * (2 * torch.rand(order) - 1)
+        elif isinstance(c_cap, list):
+            self.coeffs = torch.tensor(c_cap)
+        else:
+            raise Exception(f"expected float or list for c_cap, not {type(c_cap)}")
+
+        # Check all the dimensions
+        assert len(self.coeffs) == len(self.freqs), \
+            (f"specify the same number of coefficients and frequencies (got {len(self.coeffs)} "
+             f"and {len(self.freqs)} respectively).")
+
+        self.delta = delta
+        self.device = device
+        self.restart()
+
+    def __getitem__(self, idx: int) -> torch.Tensor | None:
+        t = idx * self.delta
+        y = torch.sum(self.coeffs * torch.sin(2 * math.pi * self.freqs * t + self.phases)).view(1, 1)
+        return self.props.adapt_tensor_to_tensor_labels(y), self.clock.get_cycle() - self.first_cycle
+
+
+class SmoothHFHA(CombSin):
+    FEATURES = ['3sin', 'hf', 'ha']
+
+    def __init__(self, device: torch.device = torch.device('cpu')):
+        freqs = [0.11, 0.07, 0.05]
+        coeffs = [0.8, 0.16, 0.16]
+        super().__init__(f_cap=freqs, c_cap=coeffs, order=3, delta=0.1, device=device)
+        self.props.set_name("smoHfHa")
+
+
+class SmoothHFLA(CombSin):
+    FEATURES = ['3sin', 'hf', 'la']
+
+    def __init__(self, device: torch.device = torch.device('cpu')):
+        freqs = [0.11, 0.07, 0.05]
+        coeffs = [0.4, 0.08, 0.08]
+        super().__init__(f_cap=freqs, c_cap=coeffs, order=3, delta=0.1, device=device)
+        self.props.set_name("smoHfLa")
+
+
+class SmoothLFLA(CombSin):
+    FEATURES = ['3sin', 'lf', 'la']
+
+    def __init__(self, device: torch.device = torch.device('cpu')):
+        freqs = [0.11, 0.07, 0.05]
+        coeffs = [0.08, 0.08, 0.4]
+        super().__init__(f_cap=freqs, c_cap=coeffs, order=3, delta=0.1, device=device)
+        self.props.set_name("smoLfLa")
+
+
+class SmoothLFHA(CombSin):
+    FEATURES = ['3sin', 'lf', 'ha']
+
+    def __init__(self, device: torch.device = torch.device('cpu')):
+        freqs = [0.11, 0.07, 0.05]
+        coeffs = [0.16, 0.16, 0.8]
+        super().__init__(f_cap=freqs, c_cap=coeffs, order=3, delta=0.1, device=device)
+        self.props.set_name("smoLfHa")
+
+
+class SquareHFHA(Square):
+    FEATURES = ['square', 'hf', 'ha']
+
+    def __init__(self, device: torch.device = torch.device('cpu')):
+        super().__init__(freq=0.06, phase=0.5, ampl=1.0, delta=0.1, device=device)
+        self.props.set_name("squHfHa")
+
+
+class SquareHFLA(Square):
+    FEATURES = ['square', 'hf', 'la']
+
+    def __init__(self, device: torch.device = torch.device('cpu')):
+        super().__init__(freq=0.06, phase=0.5, ampl=0.5, delta=0.1, device=device)
+        self.props.set_name("squHfLa")
+
+
+class SquareLFHA(Square):
+    FEATURES = ['square', 'lf', 'ha']
+
+    def __init__(self, device: torch.device = torch.device('cpu')):
+        super().__init__(freq=0.03, phase=0.5, ampl=1.0, delta=0.1, device=device)
+        self.props.set_name("squLfHa")
+
+
+class SquareLFLA(Square):
+    FEATURES = ['square', 'lf', 'la']
+
+    def __init__(self, device: torch.device = torch.device('cpu')):
+        super().__init__(freq=0.03, phase=0.5, ampl=0.5, delta=0.1, device=device)
+        self.props.set_name("squLfLa")