lynxkite-graph-analytics/src/lynxkite_graph_analytics/pytorch_model_ops.py

"""Boxes for defining PyTorch models."""

import copy
import enum
import graphlib

import pydantic
from lynxkite.core import ops, workspace
from lynxkite.core.ops import Parameter as P
import torch
import torch_geometric.nn as pyg_nn
import dataclasses
from . import core

ENV = "PyTorch model"


def op(name, **kwargs):
    _op = ops.op(ENV, name, **kwargs)

    def decorator(func):
        _op(func)
        op = func.__op__
        for p in op.inputs.values():
            p.position = "bottom"
        for p in op.outputs.values():
            p.position = "top"
        return func

    return decorator


def reg(name, inputs=[], outputs=None, params=[]):
    if outputs is None:
        outputs = inputs
    return ops.register_passive_op(
        ENV,
        name,
        inputs=[ops.Input(name=name, position="bottom", type="tensor") for name in inputs],
        outputs=[ops.Output(name=name, position="top", type="tensor") for name in outputs],
        params=params,
    )


reg("Input: tensor", outputs=["output"], params=[P.basic("name")])
reg("Input: graph edges", outputs=["edges"])
reg("Input: sequential", outputs=["y"])

reg("LSTM", inputs=["x", "h"], outputs=["x", "h"])
reg(
    "Neural ODE",
    inputs=["x"],
    params=[
        P.basic("relative_tolerance"),
        P.basic("absolute_tolerance"),
        P.options(
            "method",
            [
                "dopri8",
                "dopri5",
                "bosh3",
                "fehlberg2",
                "adaptive_heun",
                "euler",
                "midpoint",
                "rk4",
                "explicit_adams",
                "implicit_adams",
            ],
        ),
    ],
)


reg("Attention", inputs=["q", "k", "v"], outputs=["x", "weights"])
reg("LayerNorm", inputs=["x"])
reg("Dropout", inputs=["x"], params=[P.basic("p", 0.5)])


@op("Linear")
def linear(x, *, output_dim=1024):
    return pyg_nn.Linear(-1, output_dim)


class ActivationTypes(enum.Enum):
    ReLU = "ReLU"
    Leaky_ReLU = "Leaky ReLU"
    Tanh = "Tanh"
    Mish = "Mish"


@op("Activation")
def activation(x, *, type: ActivationTypes = ActivationTypes.ReLU):
    return getattr(torch.nn.functional, type.name.lower().replace(" ", "_"))


@op("MSE loss")
def mse_loss(x, y):
    return torch.nn.functional.mse_loss


reg("Softmax", inputs=["x"])
reg(
    "Graph conv",
    inputs=["x", "edges"],
    outputs=["x"],
    params=[P.options("type", ["GCNConv", "GATConv", "GATv2Conv", "SAGEConv"])],
)
reg("Concatenate", inputs=["a", "b"], outputs=["x"])
reg("Add", inputs=["a", "b"], outputs=["x"])
reg("Subtract", inputs=["a", "b"], outputs=["x"])
reg("Multiply", inputs=["a", "b"], outputs=["x"])
reg("Triplet margin loss", inputs=["x", "x_pos", "x_neg"], outputs=["loss"])
reg("Cross-entropy loss", inputs=["x", "y"], outputs=["loss"])
reg(
    "Optimizer",
    inputs=["loss"],
    outputs=[],
    params=[
        P.options(
            "type",
            [
                "AdamW",
                "Adafactor",
                "Adagrad",
                "SGD",
                "Lion",
                "Paged AdamW",
                "Galore AdamW",
            ],
        ),
        P.basic("lr", 0.001),
    ],
)

ops.register_passive_op(
    ENV,
    "Repeat",
    inputs=[ops.Input(name="input", position="top", type="tensor")],
    outputs=[ops.Output(name="output", position="bottom", type="tensor")],
    params=[
        ops.Parameter.basic("times", 1, int),
        ops.Parameter.basic("same_weights", False, bool),
    ],
)

ops.register_passive_op(
    ENV,
    "Recurrent chain",
    inputs=[ops.Input(name="input", position="top", type="tensor")],
    outputs=[ops.Output(name="output", position="bottom", type="tensor")],
    params=[],
)


def _to_id(*strings: str) -> str:
    """Replaces all non-alphanumeric characters with underscores."""
    return "_".join("".join(c if c.isalnum() else "_" for c in s) for s in strings)


@dataclasses.dataclass
class Layer:
    """Temporary data structure used by ModelBuilder."""

    module: torch.nn.Module
    origin_id: str
    inputs: list[str]
    outputs: list[str]

    def for_sequential(self):
        inputs = ", ".join(self.inputs)
        outputs = ", ".join(self.outputs)
        return self.module, f"{inputs} -> {outputs}"


class ColumnSpec(pydantic.BaseModel):
    df: str
    column: str


class ModelMapping(pydantic.BaseModel):
    map: dict[str, ColumnSpec]


@dataclasses.dataclass
class ModelConfig:
    model: torch.nn.Module
    model_inputs: list[str]
    model_outputs: list[str]
    loss_inputs: list[str]
    loss: torch.nn.Module
    optimizer: torch.optim.Optimizer
    source_workspace: str | None = None
    trained: bool = False

    def num_parameters(self) -> int:
        return sum(p.numel() for p in self.model.parameters())

    def _forward(self, inputs: dict[str, torch.Tensor]) -> dict[str, torch.Tensor]:
        model_inputs = [inputs[i] for i in self.model_inputs]
        output = self.model(*model_inputs)
        if not isinstance(output, tuple):
            output = (output,)
        values = {k: v for k, v in zip(self.model_outputs, output)}
        return values

    def inference(self, inputs: dict[str, torch.Tensor]) -> dict[str, torch.Tensor]:
        # TODO: Do multiple batches.
        self.model.eval()
        return self._forward(inputs)

    def train(self, inputs: dict[str, torch.Tensor]) -> float:
        """Train the model for one epoch. Returns the loss."""
        # TODO: Do multiple batches.
        self.model.train()
        self.optimizer.zero_grad()
        values = self._forward(inputs)
        values.update(inputs)
        loss_inputs = [values[i] for i in self.loss_inputs]
        loss = self.loss(*loss_inputs)
        loss.backward()
        self.optimizer.step()
        return loss.item()

    def copy(self):
        """Returns a copy of the model."""
        c = dataclasses.replace(self)
        c.model = copy.deepcopy(self.model)
        return c

    def metadata(self):
        return {
            "type": "model",
            "model": {
                "inputs": self.model_inputs,
                "outputs": self.model_outputs,
                "loss_inputs": self.loss_inputs,
                "trained": self.trained,
            },
        }


def build_model(ws: workspace.Workspace) -> ModelConfig:
    """Builds the model described in the workspace."""
    builder = ModelBuilder(ws)
    return builder.build_model()


class ModelBuilder:
    """The state shared between methods that are used to build the model."""

    def __init__(self, ws: workspace.Workspace):
        self.catalog = ops.CATALOGS[ENV]
        optimizers = []
        self.nodes: dict[str, workspace.WorkspaceNode] = {}
        repeats: list[str] = []
        for node in ws.nodes:
            self.nodes[node.id] = node
            if node.data.title == "Optimizer":
                optimizers.append(node.id)
            elif node.data.title == "Repeat":
                repeats.append(node.id)
                self.nodes[f"START {node.id}"] = node
                self.nodes[f"END {node.id}"] = node
        assert optimizers, "No optimizer found."
        assert len(optimizers) == 1, f"More than one optimizer found: {optimizers}"
        [self.optimizer] = optimizers
        self.dependencies = {n: [] for n in self.nodes}
        self.in_edges: dict[str, dict[str, list[tuple[str, str]]]] = {n: {} for n in self.nodes}
        self.out_edges: dict[str, dict[str, list[tuple[str, str]]]] = {n: {} for n in self.nodes}
        for e in ws.edges:
            self.dependencies[e.target].append(e.source)
            self.in_edges.setdefault(e.target, {}).setdefault(e.targetHandle, []).append(
                (e.source, e.sourceHandle)
            )
            self.out_edges.setdefault(e.source, {}).setdefault(e.sourceHandle, []).append(
                (e.target, e.targetHandle)
            )
        # Split repeat boxes into start and end, and insert them into the flow.
        # TODO: Think about recursive repeats.
        for repeat in repeats:
            if not self.out_edges[repeat] or not self.in_edges[repeat]:
                continue
            start_id = f"START {repeat}"
            end_id = f"END {repeat}"
            # repeat -> first <- real_input
            # ...becomes...
            # real_input -> start -> first
            first, firsth = self.out_edges[repeat]["output"][0]
            [(real_input, real_inputh)] = [
                k for k in self.in_edges[first][firsth] if k != (repeat, "output")
            ]
            self.dependencies[first].remove(repeat)
            self.dependencies[first].append(start_id)
            self.dependencies[start_id] = [real_input]
            self.out_edges[real_input][real_inputh] = [
                k if k != (first, firsth) else (start_id, "input")
                for k in self.out_edges[real_input][real_inputh]
            ]
            self.in_edges[start_id] = {"input": [(real_input, real_inputh)]}
            self.out_edges[start_id] = {"output": [(first, firsth)]}
            self.in_edges[first][firsth] = [(start_id, "output")]
            # repeat <- last -> real_output
            # ...becomes...
            # last -> end -> real_output
            last, lasth = self.in_edges[repeat]["input"][0]
            [(real_output, real_outputh)] = [
                k for k in self.out_edges[last][lasth] if k != (repeat, "input")
            ]
            del self.dependencies[repeat]
            self.dependencies[end_id] = [last]
            self.dependencies[real_output].append(end_id)
            self.out_edges[last][lasth] = [(end_id, "input")]
            self.in_edges[end_id] = {"input": [(last, lasth)]}
            self.out_edges[end_id] = {"output": [(real_output, real_outputh)]}
            self.in_edges[real_output][real_outputh] = [
                k if k != (last, lasth) else (end_id, "output")
                for k in self.in_edges[real_output][real_outputh]
            ]
        self.inv_dependencies = {n: [] for n in self.nodes}
        for k, v in self.dependencies.items():
            for i in v:
                self.inv_dependencies[i].append(k)
        self.layers: list[Layer] = []
        # Clean up disconnected nodes.
        disconnected = set()
        for node_id in self.nodes:
            op = self.catalog[self.nodes[node_id].data.title]
            if len(self.in_edges[node_id]) != len(op.inputs):
                disconnected.add(node_id)
                disconnected |= self.all_upstream(node_id)
        for node_id in disconnected:
            del self.dependencies[node_id]
            del self.in_edges[node_id]
            del self.out_edges[node_id]
            del self.inv_dependencies[node_id]
            del self.nodes[node_id]

    def all_upstream(self, node: str) -> set[str]:
        """Returns all nodes upstream of a node."""
        deps = set()
        for dep in self.dependencies[node]:
            deps.add(dep)
            deps.update(self.all_upstream(dep))
        return deps

    def all_downstream(self, node: str) -> set[str]:
        """Returns all nodes downstream of a node."""
        deps = set()
        for dep in self.inv_dependencies[node]:
            deps.add(dep)
            deps.update(self.all_downstream(dep))
        return deps

    def run_node(self, node_id: str) -> None:
        """Adds the layer(s) produced by this node to self.layers."""
        node = self.nodes[node_id]
        t = node.data.title
        op = self.catalog[t]
        p = op.convert_params(node.data.params)
        match t:
            case "Repeat":
                if node_id.startswith("END "):
                    repeat_id = node_id.removeprefix("END ")
                    start_id = f"START {repeat_id}"
                    [last_output] = self.in_edges[node_id]["input"]
                    after_start = self.all_downstream(start_id)
                    after_end = self.all_downstream(node_id)
                    before_end = self.all_upstream(node_id)
                    affected_nodes = after_start - after_end - {node_id}
                    repeated_nodes = after_start & before_end
                    assert affected_nodes == repeated_nodes, (
                        f"edges leave repeated section '{repeat_id}':\n{affected_nodes - repeated_nodes}"
                    )
                    repeated_layers = [e for e in self.layers if e.origin_id in repeated_nodes]
                    assert p["times"] >= 1, f"Cannot repeat {repeat_id} {p['times']} times."
                    for i in range(p["times"] - 1):
                        # Copy repeat section's output to repeat section's input.
                        self.layers.append(
                            Layer(
                                torch.nn.Identity(),
                                origin_id=node_id,
                                inputs=[_to_id(*last_output)],
                                outputs=[_to_id(start_id, "output")],
                            )
                        )
                        # Repeat the layers in the section.
                        for layer in repeated_layers:
                            if p["same_weights"]:
                                self.layers.append(layer)
                            else:
                                self.run_node(layer.origin_id)
                self.layers.append(self.run_op(node_id, op, p))
            case "Optimizer" | "Input: tensor" | "Input: graph edges" | "Input: sequential":
                return
            case _:
                self.layers.append(self.run_op(node_id, op, p))

    def run_op(self, node_id: str, op: ops.Op, params) -> Layer:
        """Returns the layer produced by this op."""
        inputs = [_to_id(*i) for n in op.inputs for i in self.in_edges[node_id][n]]
        outputs = [_to_id(node_id, n) for n in op.outputs]
        if op.func == ops.no_op:
            module = torch.nn.Identity()
        else:
            module = op.func(*inputs, **params)
        return Layer(module, node_id, inputs, outputs)

    def build_model(self) -> ModelConfig:
        # Walk the graph in topological order.
        ts = graphlib.TopologicalSorter(self.dependencies)
        for node_id in ts.static_order():
            self.run_node(node_id)
        return self.get_config()

    def get_config(self) -> ModelConfig:
        # Split the design into model and loss.
        loss_nodes = set()
        for node_id in self.nodes:
            if "loss" in self.nodes[node_id].data.title:
                loss_nodes.add(node_id)
                loss_nodes |= self.all_downstream(node_id)
        layers = []
        loss_layers = []
        for layer in self.layers:
            if layer.origin_id in loss_nodes:
                loss_layers.append(layer)
            else:
                layers.append(layer)
        used_in_model = set(input for layer in layers for input in layer.inputs)
        used_in_loss = set(input for layer in loss_layers for input in layer.inputs)
        made_in_model = set(output for layer in layers for output in layer.outputs)
        made_in_loss = set(output for layer in loss_layers for output in layer.outputs)
        layers = [layer.for_sequential() for layer in layers]
        loss_layers = [layer.for_sequential() for layer in loss_layers]
        cfg = {}
        cfg["model_inputs"] = list(used_in_model - made_in_model)
        cfg["model_outputs"] = list(made_in_model & used_in_loss)
        cfg["loss_inputs"] = list(used_in_loss - made_in_loss)
        # Make sure the trained output is output from the last model layer.
        outputs = ", ".join(cfg["model_outputs"])
        layers.append((torch.nn.Identity(), f"{outputs} -> {outputs}"))
        # Create model.
        cfg["model"] = pyg_nn.Sequential(", ".join(cfg["model_inputs"]), layers)
        # Make sure the loss is output from the last loss layer.
        [(lossb, lossh)] = self.in_edges[self.optimizer]["loss"]
        lossi = _to_id(lossb, lossh)
        loss_layers.append((torch.nn.Identity(), f"{lossi} -> loss"))
        # Create loss function.
        cfg["loss"] = pyg_nn.Sequential(", ".join(cfg["loss_inputs"]), loss_layers)
        assert not list(cfg["loss"].parameters()), f"loss should have no parameters: {loss_layers}"
        # Create optimizer.
        op = self.catalog["Optimizer"]
        p = op.convert_params(self.nodes[self.optimizer].data.params)
        o = getattr(torch.optim, p["type"].name)
        cfg["optimizer"] = o(cfg["model"].parameters(), lr=p["lr"])
        return ModelConfig(**cfg)


def to_tensors(b: core.Bundle, m: ModelMapping | None) -> dict[str, torch.Tensor]:
    """Converts a tensor to the correct type for PyTorch. Ignores missing mappings."""
    if m is None:
        return {}
    tensors = {}
    for k, v in m.map.items():
        if v.df in b.dfs and v.column in b.dfs[v.df]:
            tensors[k] = torch.tensor(b.dfs[v.df][v.column].to_list(), dtype=torch.float32)
    return tensors