Implement using plotly and pandas.

Something like this

Add option for superimposing extra particles, eg. those which aren't in final state.

Needs to be able to pass numpy arrays for everything, including masks.

Should include mass peaks, etc. More details to come.

Network visualisation

Should enable visualisation of an undirected network, similar to a scatter plot of the final state, but with edges connecting the particles.

This example was originally written to evaluate the success of a clustering model.

# assuming exist ParticleSet object at final_pcls
x = final_pcls.pmu.eta
y = final_pcls.pmu.phi
dR = gcl.matrix.delta_R_aff(final_pcls.pmu)
adj_matrix = gcl.matrix.cut_adj(dR, conf.dataset.radius)
degrees = adj_matrix.sum(axis=0)

def edge_pos(pcl_pos: npt.NDArray[np.float64], adj: npt.NDArray[np.bool_]) -> list[list[Optional[float]]]:
    edges = []
    for ux, row in zip(pcl_pos, adj):
        for vx in pcl_pos[row]:
            edges += [ux, vx, None]
    return edges

edge_trace = go.Scatter(
    x=edge_pos(x, adj_matrix), y=edge_pos(y, adj_matrix),
    line=dict(width=0.5, color="#686a72"),
    hoverinfo="none", mode="lines",
    showlegend=False,
)

node_trace = go.Scatter(
    x=list(x),
    y=list(y),
    mode="markers",
    hoverinfo="text",
#     text=list(final_pcls.pdg.name),
    text=list(map("degree={}, confidence={:.3f}".format, degrees, preds)),
    showlegend=False,
    marker=dict(
        showscale=True,
        symbol=labels,
        colorscale='Bluered',
        reversescale=True,
        color=degrees,
        size=10,
        colorbar=dict(
            thickness=15,
            title='degree',
            xanchor='left',
            titleside='right'
        ),
        line_width=1
    ),
)

fig = go.Figure(
    data=[edge_trace, node_trace],
    layout=go.Layout(
        title='Input graph data structure',
        titlefont_size=16,
        hovermode='closest',
        margin=dict(b=20,l=5,r=5,t=40),
        xaxis=dict(showgrid=True, zeroline=False, showticklabels=True, title_text="Pseudorapidity"),
        yaxis=dict(showgrid=True, zeroline=False, showticklabels=True, title_text="Azimuth"),
        )
    )
fig.show()
fig.write_html(fig_dir / f"{num:02}-preds-degree.html")

Features

Core

• connectivity: for simplicity, compute the radius-based edges automatically, rather than taking as input
• colour bar: provide input for colours, eg. $p_T$ or confidence score, etc.
• inputs: should take either full 4-momentum, 3-momentum, or $\phi, \eta$ as inputs
  • for the moment, just accept 4-momenta, but will do some refactoring later

Additional

• different symbols: allow particles to be assigned to different groups, with unique marker symbols
• annotations: enable additional information to be included on cursor hover, eg. particle name / PDG

The Histogram object itself should be at the top of the figure. Change the ordering so the overlays don't obscure it.

Add __eq__ and __add__ special methods to Histogram object for comparing and combining histogram data.

Histogram.update() currently takes one value at a time. Enable the val parameter to be an iterable s.t. pre-computed arrays can be efficiently binned, and the user doesn't have to resort to iteratively calling Histogram.update().

Add I/O to methods to Histogram s.t. they can be serialised and stored / read as JSON files.

Normalisation for the Histogram pdfs doesn't take into account bin width. Make sure values are normalised wrt to bin width.

Enable initialisation of Histogram from pre-computed histogram data, specifically in the format outputted from numpy.histogram(), ie. a two-tuple of arrays containing counts and bin edges, respectively. Should also contain an optional total parameter, which normalises the density with respect to the total number of values, rather than those just present in the bin range.

Dependency issues as the API for graphicle changed between versions 0.2 and 0.3. Additionally, can make use of the graphicle.select.clusters() function.

Maybe add some unit tests that periodically run, to ensure the integration of these packages is consistent.