Create Gradio-based GUI

Dockerfile

@@ -1,7 +1,7 @@
 # This builds a dockerfile containing a working copy of PySR
 # with all pre-requisites installed.
 
-ARG JLVERSION=1.9.4
+ARG JLVERSION=1.10.0
 ARG PYVERSION=3.11.6
 ARG BASE_IMAGE=bullseye
 
@@ -12,28 +12,55 @@ FROM python:${PYVERSION}-${BASE_IMAGE}
 COPY --from=jl /usr/local/julia /usr/local/julia
 ENV PATH="/usr/local/julia/bin:${PATH}"
 
-# Install IPython and other useful libraries:
-RUN pip install --no-cache-dir ipython matplotlib
+# Install font used for GUI
+RUN mkdir -p /usr/local/share/fonts/IBM_Plex_Mono && \
+    curl -L https://github.com/IBM/plex/releases/download/v6.4.0/IBM-Plex-Mono.zip -o /tmp/IBM_Plex_Mono.zip && \
+    unzip /tmp/IBM_Plex_Mono.zip -d /usr/local/share/fonts/IBM_Plex_Mono && \
+    rm /tmp/IBM_Plex_Mono.zip
+RUN fc-cache -f -v
 
-WORKDIR /pysr
+# Set up a new user named "user" with user ID 1000
+RUN useradd -m -u 1000 user
+USER user
+WORKDIR /home/user/
+ENV HOME=/home/user
+ENV PATH=/home/user/.local/bin:$PATH
 
-# Caches install (https://stackoverflow.com/questions/25305788/how-to-avoid-reinstalling-packages-when-building-docker-image-for-python-project)
-ADD ./requirements.txt /pysr/requirements.txt
-RUN pip3 install --no-cache-dir -r /pysr/requirements.txt
+RUN python -m venv $HOME/.venv
 
-# Install PySR:
-# We do a minimal copy so it doesn't need to rerun at every file change:
-ADD ./pyproject.toml /pysr/pyproject.toml
-ADD ./setup.py /pysr/setup.py
-ADD ./pysr /pysr/pysr
-RUN pip3 install --no-cache-dir .
+ENV PYTHON="${HOME}/.venv/bin/python"
+ENV PIP="${PYTHON} -m pip"
+ENV PATH="${HOME}/.venv/bin:${PATH}"
+
+WORKDIR $HOME/pysr
+
+# Install all requirements, and then PySR itself
+COPY --chown=user ./requirements.txt $HOME/pysr/requirements.txt
+RUN $PIP install --no-cache-dir -r $HOME/pysr/requirements.txt
+
+COPY --chown=user ./gui/requirements.txt $HOME/pysr/gui/requirements.txt
+RUN $PIP install --no-cache-dir -r $HOME/pysr/gui/requirements.txt
+
+COPY --chown=user ./pyproject.toml $HOME/pysr/pyproject.toml
+COPY --chown=user ./setup.py $HOME/pysr/setup.py
+COPY --chown=user ./pysr $HOME/pysr/pysr
+RUN $PIP install --no-cache-dir .
 
 # Install Julia pre-requisites:
-RUN python3 -c 'import pysr'
+RUN $PYTHON -c 'import pysr'
+
+COPY --chown=user ./gui/*.py $HOME/pysr/gui/
+
+EXPOSE 7860
+ENV GRADIO_ALLOW_FLAGGING=never \
+	GRADIO_NUM_PORTS=1 \
+	GRADIO_SERVER_NAME=0.0.0.0 \
+	GRADIO_THEME=huggingface \
+	SYSTEM=spaces
 
 # metainformation
 LABEL org.opencontainers.image.authors = "Miles Cranmer"
 LABEL org.opencontainers.image.source = "https://github.com/MilesCranmer/PySR"
 LABEL org.opencontainers.image.licenses = "Apache License 2.0"
 
-CMD ["ipython"]
+CMD ["/home/user/.venv/bin/python", "/home/user/pysr/gui/app.py"]

README.md

@@ -1,3 +1,13 @@
+---
+title: PySR
+emoji: 🌍
+colorFrom: green
+colorTo: indigo
+sdk: docker
+pinned: false
+license: apache-2.0
+---
+
 [//]: # (Logo:)
 
 <div align="center">

gui/app.py

@@ -0,0 +1,300 @@
+from collections import OrderedDict
+
+import gradio as gr
+import numpy as np
+from data import TEST_EQUATIONS
+from gradio.components.base import Component
+from plots import plot_example_data, plot_pareto_curve
+from processing import processing, stop
+
+
+class ExampleData:
+    def __init__(self, demo: gr.Blocks) -> None:
+        with gr.Column(scale=1):
+            self.example_plot = gr.Plot()
+        with gr.Column(scale=1):
+            self.test_equation = gr.Radio(
+                TEST_EQUATIONS, value=TEST_EQUATIONS[0], label="Test Equation"
+            )
+            self.num_points = gr.Slider(
+                minimum=10,
+                maximum=1000,
+                value=200,
+                label="Number of Data Points",
+                step=1,
+            )
+            self.noise_level = gr.Slider(
+                minimum=0, maximum=1, value=0.05, label="Noise Level"
+            )
+            self.data_seed = gr.Number(value=0, label="Random Seed")
+
+        # Set up plotting:
+
+        eqn_components = [
+            self.test_equation,
+            self.num_points,
+            self.noise_level,
+            self.data_seed,
+        ]
+        for eqn_component in eqn_components:
+            eqn_component.change(
+                plot_example_data,
+                eqn_components,
+                self.example_plot,
+                show_progress=False,
+            )
+
+        demo.load(plot_example_data, eqn_components, self.example_plot)
+
+
+class UploadData:
+    def __init__(self) -> None:
+        self.file_input = gr.File(label="Upload a CSV File")
+        self.label = gr.Markdown(
+            "The rightmost column of your CSV file will be used as the target variable."
+        )
+
+
+class Data:
+    def __init__(self, demo: gr.Blocks) -> None:
+        with gr.Tab("Example Data"):
+            self.example_data = ExampleData(demo)
+        with gr.Tab("Upload Data"):
+            self.upload_data = UploadData()
+
+
+class BasicSettings:
+    def __init__(self) -> None:
+        self.binary_operators = gr.CheckboxGroup(
+            choices=["+", "-", "*", "/", "^", "max", "min", "mod", "cond"],
+            label="Binary Operators",
+            value=["+", "-", "*", "/"],
+        )
+        self.unary_operators = gr.CheckboxGroup(
+            choices=[
+                "sin",
+                "cos",
+                "tan",
+                "exp",
+                "log",
+                "square",
+                "cube",
+                "sqrt",
+                "abs",
+                "erf",
+                "relu",
+                "round",
+                "sign",
+            ],
+            label="Unary Operators",
+            value=["sin"],
+        )
+        self.niterations = gr.Slider(
+            minimum=1,
+            maximum=1000,
+            value=40,
+            label="Number of Iterations",
+            step=1,
+        )
+        self.maxsize = gr.Slider(
+            minimum=7,
+            maximum=100,
+            value=20,
+            label="Maximum Complexity",
+            step=1,
+        )
+        self.parsimony = gr.Number(
+            value=0.0032,
+            label="Parsimony Coefficient",
+        )
+
+
+class AdvancedSettings:
+    def __init__(self) -> None:
+        self.populations = gr.Slider(
+            minimum=2,
+            maximum=100,
+            value=15,
+            label="Number of Populations",
+            step=1,
+        )
+        self.population_size = gr.Slider(
+            minimum=2,
+            maximum=1000,
+            value=33,
+            label="Population Size",
+            step=1,
+        )
+        self.ncycles_per_iteration = gr.Number(
+            value=550,
+            label="Cycles per Iteration",
+        )
+        self.elementwise_loss = gr.Radio(
+            ["L2DistLoss()", "L1DistLoss()", "LogitDistLoss()", "HuberLoss()"],
+            value="L2DistLoss()",
+            label="Loss Function",
+        )
+        self.adaptive_parsimony_scaling = gr.Number(
+            value=20.0,
+            label="Adaptive Parsimony Scaling",
+        )
+        self.optimizer_algorithm = gr.Radio(
+            ["BFGS", "NelderMead"],
+            value="BFGS",
+            label="Optimizer Algorithm",
+        )
+        self.optimizer_iterations = gr.Slider(
+            minimum=1,
+            maximum=100,
+            value=8,
+            label="Optimizer Iterations",
+            step=1,
+        )
+        self.batching = gr.Checkbox(
+            value=False,
+            label="Batching",
+        )
+        self.batch_size = gr.Slider(
+            minimum=2,
+            maximum=1000,
+            value=50,
+            label="Batch Size",
+            step=1,
+        )
+
+
+class GradioSettings:
+    def __init__(self) -> None:
+        self.plot_update_delay = gr.Slider(
+            minimum=1,
+            maximum=100,
+            value=3,
+            label="Plot Update Delay",
+        )
+        self.force_run = gr.Checkbox(
+            value=False,
+            label="Ignore Warnings",
+        )
+
+
+class Settings:
+    def __init__(self):
+        with gr.Tab("Basic Settings"):
+            self.basic_settings = BasicSettings()
+        with gr.Tab("Advanced Settings"):
+            self.advanced_settings = AdvancedSettings()
+        with gr.Tab("Gradio Settings"):
+            self.gradio_settings = GradioSettings()
+
+
+class Results:
+    def __init__(self):
+        with gr.Tab("Pareto Front"):
+            self.pareto = gr.Plot()
+        with gr.Tab("Predictions"):
+            self.predictions_plot = gr.Plot()
+
+        self.df = gr.Dataframe(
+            headers=["complexity", "loss", "equation"],
+            datatype=["number", "number", "str"],
+            wrap=True,
+            column_widths=[75, 75, 200],
+            interactive=False,
+        )
+
+        self.messages = gr.Textbox(label="Messages", value="", interactive=False)
+
+
+def flatten_attributes(
+    component_group, absolute_name: str, d: OrderedDict
+) -> OrderedDict:
+    if not hasattr(component_group, "__dict__"):
+        return d
+
+    for name, elem in component_group.__dict__.items():
+        new_absolute_name = absolute_name + "." + name
+        if name.startswith("_"):
+            # Private attribute
+            continue
+        elif elem in d.values():
+            # Don't duplicate any tiems
+            continue
+        elif isinstance(elem, Component):
+            # Only add components to dict
+            d[new_absolute_name] = elem
+        else:
+            flatten_attributes(elem, new_absolute_name, d)
+
+    return d
+
+
+class AppInterface:
+    def __init__(self, demo: gr.Blocks) -> None:
+        with gr.Row():
+            with gr.Column(scale=2):
+                with gr.Row():
+                    self.data = Data(demo)
+                with gr.Row():
+                    self.settings = Settings()
+            with gr.Column(scale=2):
+                self.results = Results()
+                with gr.Row():
+                    with gr.Column(scale=1):
+                        self.stop = gr.Button(value="Stop")
+                    with gr.Column(scale=1, min_width=200):
+                        self.run = gr.Button()
+
+        # Update plot when dataframe is updated:
+        self.results.df.change(
+            plot_pareto_curve,
+            inputs=[self.results.df, self.settings.basic_settings.maxsize],
+            outputs=[self.results.pareto],
+            show_progress=False,
+        )
+
+        ignore = ["df", "predictions_plot", "pareto", "messages"]
+        self.run.click(
+            create_processing_function(self, ignore=ignore),
+            inputs=[
+                v
+                for k, v in flatten_attributes(self, "interface", OrderedDict()).items()
+                if last_part(k) not in ignore
+            ],
+            outputs=[
+                self.results.df,
+                self.results.predictions_plot,
+                self.results.messages,
+            ],
+            show_progress=True,
+        )
+        self.stop.click(stop)
+
+
+def last_part(k: str) -> str:
+    return k.split(".")[-1]
+
+
+def create_processing_function(interface: AppInterface, ignore=[]):
+    d = flatten_attributes(interface, "interface", OrderedDict())
+    keys = [k for k in map(last_part, d.keys()) if k not in ignore]
+    _, idx, counts = np.unique(keys, return_index=True, return_counts=True)
+    if np.any(counts > 1):
+        raise AssertionError("Bad keys: " + ",".join(np.array(keys)[idx[counts > 1]]))
+
+    def f(*components):
+        n = len(components)
+        assert n == len(keys)
+        for output in processing(**{keys[i]: components[i] for i in range(n)}):
+            yield output
+
+    return f
+
+
+def main():
+    with gr.Blocks(theme="default") as demo:
+        _ = AppInterface(demo)
+        demo.launch(debug=True)
+
+
+if __name__ == "__main__":
+    main()