plot.py

#!/usr/bin/env python
##############################################################################
#
# SrMise            by Luke Granlund
#                   (c) 2014 trustees of the Michigan State University.
#                   All rights reserved.
#
# File coded by:    Luke Granlund
#
# See LICENSE.txt for license information.
#
##############################################################################
"""plot extracted peaks and comparison to ideal distances (if given)"""

import optparse
import sys

import matplotlib.pyplot as plt
import mpl_toolkits.axisartist as AA
import numpy as np
from matplotlib.ticker import MultipleLocator
from mpl_toolkits.axes_grid1.inset_locator import inset_axes

from diffpy.srmise import PDFPeakExtraction, PeakStability
from diffpy.srmise.pdfpeakextraction import resample

# For a given figure, returns a label of interest
labeldict = {}

default_gobs_style = {
    "color": "b",
    "linestyle": "",
    "markeredgecolor": "b",
    "marker": "o",
    "markerfacecolor": "none",
    "markersize": 4,
}

default_gfit_style = {"color": "g"}
default_gind_style = {"facecolor": "green", "alpha": 0.2}
default_gres_style = {}
default_ep_style = {}
default_ip_style = {}

default_dg_style = {
    "linestyle": "none",
    "color": "black",
    "marker": "o",
    "markerfacecolor": "black",
    "markeredgecolor": "black",
    "markersize": 1,
    "antialiased": False,
}


def setfigformat(figsize):
    from matplotlib import rc

    rc("legend", numpoints=2)
    rc("figure", figsize=figsize)
    rc("axes", titlesize=12, labelsize=11)
    rc("xtick", labelsize=10)
    rc("ytick", labelsize=10)
    rc("lines", linewidth=0.75, markeredgewidth=0.5)
    return


def gr_formataxis(ax=None):
    if ax is None:
        ax = plt.gca()
    ax.xaxis.set_minor_locator(MultipleLocator(1))
    ax.yaxis.set_label_position("left")
    ax.yaxis.tick_left()
    ax.yaxis.set_ticks_position("both")
    return


def comparepositions(ppe, ip=None, **kwds):
    ax = kwds.get("ax", plt.gca())
    base = kwds.get("base", 0.0)
    yideal = kwds.get("yideal", -1.0)
    yext = kwds.get("yext", 1.0)
    ep_style = kwds.get("ep_style", default_ep_style)
    ip_style = kwds.get("ip_style", default_ip_style)
    yideal_label = kwds.get("yideal_label", r"ideal")
    yext_label = kwds.get("yext_label", r"found")
    pmin = kwds.get("pmin", -np.inf)
    pmax = kwds.get("pmax", np.inf)

    ep = [p["position"] for p in ppe.model]
    ep = [p for p in ep if p >= pmin and p <= pmax]

    if ip is not None:
        xi = np.NaN + np.zeros(3 * len(ip))
        xi[0::3] = ip
        xi[1::3] = ip
        yi = np.zeros_like(xi) + base
        yi[1::3] += yideal
        plt.plot(xi, yi, "b", lw=1.5, **ip_style)

    xe = np.NaN + np.zeros(3 * len(ep))
    xe[0::3] = ep
    xe[1::3] = ep
    ye = np.zeros_like(xe) + base
    ye[1::3] += yext
    plt.plot(xe, ye, "g", lw=1.5, **ep_style)

    if ip is not None:
        plt.axhline(base, linestyle=":", color="k")
        ax.yaxis.set_ticks([base + 0.5 * yideal, base + 0.5 * yext])
        ax.yaxis.set_ticklabels([yideal_label, yext_label])
    else:
        ax.yaxis.set_ticks([base + 0.5 * yext])
        ax.yaxis.set_ticklabels([yext_label])

    # Set ylim explicitly, for case where yext is empty.
    if ip is not None:
        plt.ylim(base + yideal, base + yext)
    else:
        plt.ylim(base, base + yext)

    for tick in ax.yaxis.get_major_ticks():
        tick.tick1line.set_markersize(0)
        tick.tick2line.set_markersize(0)
        tick.label1.set_verticalalignment("center")
        tick.label1.set_fontsize(8)
    ticks = ax.yaxis.get_major_ticks()
    ticks[-1].label1.set_color("green")
    if ip is not None:
        ticks[0].label1.set_color("blue")
    return


def dgseries(stability, **kwds):
    kwds.get("ax", plt.gca())
    dg_style = kwds.get("dg_style", default_dg_style)

    scale = kwds.get("scale", 1.0)

    dgmin = kwds.get("dgmin", stability.results[0][0]) * scale
    dgmax = kwds.get("dgmax", stability.results[-1][0]) * scale

    pmin = kwds.get("pmin", 0.0)
    pmax = kwds.get("pmax", np.inf)

    x = []
    y = []
    for dg, peaks, bl, dr in stability.results:
        if dg * scale < dgmin or dg * scale > dgmax:
            continue
        peakpos = [p["position"] for p in peaks]
        peakpos = [p for p in peakpos if p >= pmin and p <= pmax]
        x.extend(peakpos)
        y.extend(np.zeros_like(peakpos) + dg * scale)
    plt.plot(x, y, **dg_style)


def labelallsubplots():
    rv = []
    for i, c in enumerate("abcd"):
        plt.subplot(221 + i)
        s = "(%s)" % c
        ht = plt.text(
            0.04,
            0.95,
            s,
            horizontalalignment="left",
            verticalalignment="top",
            transform=plt.gca().transAxes,
            weight="bold",
        )
        rv.append(ht)
    return rv


def makeplot(ppe_or_stability, ip=None, **kwds):
    """Plot stuff"""
    if isinstance(ppe_or_stability, PeakStability):
        stability = ppe_or_stability
        ppe = stability.ppe
    else:
        stability = None
        ppe = ppe_or_stability

    if ppe.extracted is None:
        # Makeplot requires a ModelCluster, so whip one up.
        from diffpy.srmise import ModelCluster

        ppe.defaultvars()  # Make sure everything has some setting.  This
        # shouldn't have harmful side effects.
        rangeslice = ppe.getrangeslice()
        x = ppe.x[rangeslice]
        y = ppe.y[rangeslice]
        dy = ppe.effective_dy[rangeslice]
        mcluster = ModelCluster(ppe.initial_peaks, ppe.baseline, x, y, dy, None, ppe.error_method, ppe.pf)
        ext = mcluster
    else:
        ext = ppe.extracted

    figdict = {}

    # Range along x-axis
    xlo = kwds.get("xlo", ext.r_cluster[0])
    xhi = kwds.get("xhi", ext.r_cluster[-1])

    # Range of PDF to display
    # This is deferred until the defaults can be calculated
    # min_gr
    # max_gr

    # Define heights and interstitial offsets
    # All values in percent of main axis.
    top_offset = kwds.get("top_offset", 0.0)
    dg_height = kwds.get("dg_height", 15.0 if stability is not None else 0.0)
    cmp_height = kwds.get("cmp_height", 15.0 if ip is not None else 7.5)
    datatop_offset = kwds.get("datatop_offset", 3.0)
    # <- Data appears here ->
    databottom_offset = kwds.get("databottom_offset", 3.0)
    # <- Residual appears here ->
    bottom_offset = kwds.get("bottom_offset", 3.0)

    # Style options
    kwds.get("dg_style", default_dg_style)
    gobs_style = kwds.get("gobs_style", default_gobs_style)
    gfit_style = kwds.get("gfit_style", default_gfit_style)
    gind_style = kwds.get("gind_style", default_gind_style)
    gres_style = kwds.get("gres_style", default_gres_style)
    ep_style = kwds.get("ep_style", default_ep_style)
    ip_style = kwds.get("ip_style", default_ip_style)

    # Label options
    userxlabel = kwds.get("xlabel", r"r ($\mathrm{\AA}$)")
    userylabel = kwds.get("ylabel", r"G ($\mathrm{\AA^{-2}}$)")
    datalabelx = kwds.get("datalabelx", 0.04)
    yideal_label = kwds.get("yideal_label", r"ideal")
    yext_label = kwds.get("yext_label", r"found")

    # Other options
    datalabel = kwds.get("datalabel", None)
    dgformatstr = kwds.get("dgformatstr", r"$\delta$g=%f")
    kwds.get("dgformatpost", None)  # ->userfunction(string)
    show_fit = kwds.get("show_fit", True)
    show_individual = kwds.get("show_individual", True)
    fill_individual = kwds.get("fill_individual", True)
    show_observed = kwds.get("show_observed", True)
    show_residual = kwds.get("show_residual", True)
    mask_residual = kwds.get("mask_residual", False)  # -> number
    show_annotation = kwds.get("show_annotation", True)
    scale = kwds.get("scale", 1.0)  # Apply a global scaling factor to the data

    # Define the various data which will be plotted
    r = ext.r_cluster
    dr = (r[-1] - r[0]) / len(r)
    rexpand = np.concatenate((np.arange(r[0] - dr, xlo, -dr)[::-1], r, np.arange(r[-1] + dr, xhi + dr, dr)))
    rfine = np.arange(r[0], r[-1], 0.1 * dr)
    gr_obs = np.array(resample(ppe.x, ppe.y, rexpand)) * scale
    # gr_fit = resample(r, ext.value(), rfine)
    gr_fit = np.array(ext.value(rfine)) * scale
    gr_fit_baseline = np.array(ext.valuebl(rfine)) * scale
    gr_fit_ind = [gr_fit_baseline + np.array(p.value(rfine)) * scale for p in ext.model]
    gr_res = np.array(ext.residual()) * scale

    if mask_residual:
        gr_res = np.ma.masked_outside(gr_res, -mask_residual, mask_residual)

    all_gr = []
    if show_fit:
        all_gr.append(gr_fit)
    # if show_individual: all_gr.extend([gr_fit_baseline, gr_fit_ind])
    if show_individual:
        all_gr.append(gr_fit_baseline)
        if len(gr_fit_ind) > 0:
            all_gr.extend(gr_fit_ind)
    if show_observed:
        all_gr.append(gr_obs)

    # gr_fit_ind is a list of lists, so use np.min/max
    # The funky bit with scale makes sure that a user-specified value
    # has scale applied to it, without messing up the default values,
    # which are calculated from already scaled quantities.
    min_gr = kwds.get("min_gr", np.min([np.min(gr) for gr in all_gr]) / scale) * scale
    max_gr = kwds.get("max_gr", np.max([np.max(gr) for gr in all_gr]) / scale) * scale

    if show_residual:
        min_res = np.min(gr_res)
        max_res = np.max(gr_res)
    else:
        min_res = 0.0
        max_res = 0.0

    # Derive various y limits based on all the offsets
    rel_height = 100.0 - top_offset - dg_height - cmp_height - datatop_offset - databottom_offset - bottom_offset
    abs_height = 100 * ((max_gr - min_gr) + (max_res - min_res)) / rel_height

    yhi = max_gr + (top_offset + dg_height + cmp_height + datatop_offset) * abs_height / 100
    ylo = yhi - abs_height

    yhi = kwds.get("yhi", yhi)
    ylo = kwds.get("ylo", ylo)

    datatop = yhi - (yhi - ylo) * 0.01 * (top_offset + dg_height + cmp_height)
    datalabeltop = 1 - 0.01 * (top_offset + dg_height + cmp_height + datatop_offset)
    resbase = ylo + bottom_offset * abs_height / 100 - min_res

    resbase = kwds.get("resbase", resbase)

    fig = kwds.get("figure", plt.gcf())
    fig.clf()
    ax_data = AA.Subplot(fig, 111)
    fig.add_subplot(ax_data)
    figdict["fig"] = fig
    figdict["data"] = ax_data

    # Plot the data, fit, and residual
    if show_observed:
        plt.plot(rexpand, gr_obs, **gobs_style)
    if show_fit:
        plt.plot(rfine, gr_fit, **gfit_style)
    if fill_individual:
        for peak in gr_fit_ind:
            plt.fill_between(rfine, gr_fit_baseline, peak, **gind_style)
    if show_residual:
        plt.plot(r, gr_res + resbase, "r-", **gres_style)
        plt.plot((xlo, xhi), 2 * [resbase], "k:")

    # Format ax_data
    plt.xlim(xlo, xhi)
    plt.ylim(ylo, yhi)
    plt.xlabel(userxlabel)
    plt.ylabel(userylabel)
    ax_data.xaxis.set_minor_locator(plt.MultipleLocator(1))
    # ax_data.yaxis.set_minor_locator(plt.MultipleLocator(np.max([1,int((yhi-ylo)/20)])))
    ax_data.yaxis.set_label_position("left")
    ax_data.yaxis.tick_left()
    ax_data.yaxis.set_ticks_position("both")

    # Remove labels above where insets begin
    # ax_data.yaxis.set_ticklabels([str(int(loc)) for loc in ax_data.yaxis.get_majorticklocs() if loc < datatop])
    ax_data.yaxis.set_ticks([loc for loc in ax_data.yaxis.get_majorticklocs() if (loc < datatop and loc >= ylo)])

    # Dataset label
    if datalabel is not None:
        dl = plt.text(
            datalabelx,
            datalabeltop,
            datalabel,
            ha="left",
            va="top",
            transform=ax_data.transAxes,
            weight="bold",
        )
    else:
        dl = None
    figdict["datalabel"] = dl

    # Create new x axis at bottom edge of compare inset
    ax_data.axis["top"].set_visible(False)
    ax_data.axis["newtop"] = ax_data.new_floating_axis(0, datatop, axis_direction="bottom")  # "top" bugged?
    ax_data.axis["newtop"].toggle(all=False, ticks=True)
    ax_data.axis["newtop"].major_ticks.set_tick_out(True)
    ax_data.axis["newtop"].minor_ticks.set_tick_out(True)

    # New y-axis label, since AxisLabel positions cannot be set manually.
    # The original label is invisible, but we use its (dynamic) x position
    # to update the new label, which we define have the correct y position.
    # A bit of a tradeoff for the nice insets and ability to define new axes.
    newylabel = plt.text(
        -0.1,
        0.5 * (datatop - ylo) / (yhi - ylo),
        userylabel,
        ha="center",
        va="center",
        rotation="vertical",
        transform=ax_data.transAxes,
    )
    labeldict[fig] = newylabel  # so we can find the correct text object
    fig.canvas.mpl_connect("draw_event", on_draw)  # original label invisibility and updating

    # Compare extracted (and ideal, if provided) peak positions clearly.
    if cmp_height > 0:
        ax_cmp = inset_axes(
            ax_data,
            width="100%",
            height="%s%%" % cmp_height,
            loc=2,
            bbox_to_anchor=(0.0, -0.01 * (top_offset + dg_height), 1, 1),
            bbox_transform=ax_data.transAxes,
            borderpad=0,
        )

        figdict["cmp"] = ax_cmp
        plt.axes(ax_cmp)
        comparepositions(
            ext,
            ip,
            ep_style=ep_style,
            ip_style=ip_style,
            yideal_label=yideal_label,
            yext_label=yext_label,
        )
        plt.xlim(xlo, xhi)
        ax_cmp.set_xticks([])

    # Show how extracted peak positions change as dg is changed
    if dg_height > 0:

        ax_dg = inset_axes(
            ax_data,
            width="100%",
            height="%s%%" % dg_height,
            loc=2,
            bbox_to_anchor=(0, -0.01 * top_offset, 1, 1),
            bbox_transform=ax_data.transAxes,
            borderpad=0,
        )

        figdict["dg"] = ax_dg
        plt.axes(ax_dg)
        dgkwds = {}
        if "scale" in kwds:
            dgkwds["scale"] = kwds["scale"]
        if "dgmin" in kwds:
            dgkwds["dgmin"] = kwds["dgmin"]
        if "dgmax" in kwds:
            dgkwds["dgmax"] = kwds["dgmax"]
        dgseries(stability, base=0, pmin=r[0], pmax=r[-1], **dgkwds)
        plt.xlim(xlo, xhi)
        ax_dg.xaxis.set_major_locator(plt.NullLocator())
        ax_dg.yaxis.set_major_locator(plt.MaxNLocator(3))
        plt.ylabel(r"$\delta$g")

    # Annotate the actual dg shown
    if show_annotation:
        dg = np.mean(ext.error_cluster) * scale
        dgstr = dgformatstr % dg
        if "dgformatpost" in kwds:  # post-processing on dg annotation
            dgstr = kwds["dgformatpost"](dgstr)

        if len(ext.model) > 0:
            xpos = np.mean([xlo, ext.model[0]["position"]])  # OK for now.
        else:
            xpos = xlo + 0.1 * (xhi - xlo)
        if dg_height > 0 and cmp_height > 0:
            # Arrow, text in compare distances line
            ylo2, yhi2 = ax_dg.get_ylim()
            if ip is not None:
                ypos = ylo2 - 0.25 * cmp_height / dg_height * (yhi2 - ylo2)
            else:
                ypos = ylo2 - 0.5 * cmp_height / dg_height * (yhi2 - ylo2)
            plt.annotate(
                dgstr,
                xy=(xlo, dg),
                xycoords="data",
                va="center",
                ha="center",
                xytext=(xpos, ypos),
                textcoords="data",
                size=8,
                color="green",
                arrowprops=dict(
                    arrowstyle="->",
                    connectionstyle="angle,angleA=90,angleB=0,rad=10",
                    color="green",
                ),
            )

        elif dg_height > 0:
            # Arrow, and text located somewhere in main plot region
            # Must change axes
            pass
        elif cmp_height > 0:
            # No arrow, text in compare distances line
            # Must change axes
            plt.axes(ax_cmp)
            ylo2, yhi2 = ax_cmp.get_ylim()
            ypos = yhi2 / 2.0
            plt.text(xpos, ypos, dgstr, va="center", ha="center", size=8, color="green")
        else:
            # Text only in main plot region
            # Must change axes
            pass

    plt.draw()

    return figdict


# Bit of a kluge to make sure the label on the y-axis
# is placed correctly.  The "invisiblelabel" has correct
# x-position, but it's y-position cannot be manually set.
# The visiblelabel has correct y-position, and we update
# its x-position based on invisiblelabel.  Of course,
# invisiblelabel must be temporarily made visible to update
# its values.
_lastxpos = {}


def on_draw(event):
    global _lastxpos
    fig = event.canvas.figure
    ax_main = fig.get_axes()[0]
    invisiblelabel = ax_main.axis["left"].label
    invisiblelabel.set_visible(True)
    visiblelabel = labeldict[fig]
    bbox = invisiblelabel.get_window_extent(invisiblelabel._renderer)
    bbox = bbox.inverse_transformed(ax_main.transAxes)
    bbox = bbox.get_points()
    xpos = np.mean(np.transpose(bbox)[0])

    # This, and the whole lastxpos business, is so label is properly
    # updated when using the Agg backend to create a .png. (at least in
    # matplotlib 1.1.0)  For some reason the invisible label is not set
    # correctly when drawn with that backend unless redrawn at least once.
    # If it is kept visible the whole time this problem doesn't occur.
    # This problem doesn't occur onscreen (TkAgg) or printing PDFs, and
    # didn't occur in matplotlib 1.0.0.
    if abs(xpos - _lastxpos.get(fig, 0)) > 0.001:
        _lastxpos[fig] = xpos
        plt.draw()
    else:
        _lastxpos[fig] = xpos

    invisiblelabel.set_visible(False)
    xpos_old = visiblelabel.get_position()[0]
    if abs(xpos - xpos_old) > 0.001:
        labeldict[fig].set_x(xpos)
        plt.draw()
    return


def readcompare(filename):
    """Returns a list of distances read from filename, otherwise None."""

    # TODO: Make this safer
    try:
        datastring = open(filename, "rb").read()
    except Exception as err:
        raise err

    import re

    res = re.search(r"^[^#]", datastring, re.M)
    if res:
        datastring = datastring[res.end() :].strip()

    distances = []

    try:
        for line in datastring.split("\n"):
            distances.append(float(line))
    except (ValueError, IndexError):
        print("Could not read distances from '%s'. Ignoring file." % filename)

    if len(distances) == 0:
        return None
    else:
        return distances


def main():
    # configure options parsing
    usage = (
        "%prog srmise_file [options]\n"
        "srmise_file can be an extraction file saved by SrMise, "
        "or a data file saved by PeakStability."
    )
    descr = (
        "A very basic tool for somewhat prettier plotting than provided by "
        "the basic SrMise classes.  Can be used to compare peak positions "
        "with those from a list.\n"
        "NOTE: At this time the utility only works with peaks extracted using diffpy.srmise.PDFPeakExtraction."
    )

    parser = optparse.OptionParser(usage=usage, description=descr)
    parser.add_option(
        "--compare",
        type="string",
        help="Compare extracted distances to distances listed (1/line) in this file.",
    )
    parser.add_option(
        "--model",
        type="int",
        help="Plot given model from set.  Ignored if srmise_file is not a PeakStability file.",
    )
    parser.add_option("--show", action="store_true", help="execute pylab.show() blocking call")
    parser.add_option("-o", "--output", type="string", help="save plot to the specified file")
    parser.add_option("--format", type="string", default="eps", help="output format for plot saving")
    parser.allow_interspersed_args = True
    opts, args = parser.parse_args(sys.argv[1:])

    if len(args) != 1:
        parser.error("Exactly one argument required. \n" + usage)

    filename = args[0]

    if filename is not None:
        toplot = PDFPeakExtraction()
        try:
            toplot.read(filename)
        except Exception:
            toplot = PeakStability()
            try:
                toplot.load(filename)
            except Exception:
                print("File '%s' is not a .srmise or PeakStability data file." % filename)
                return

    if opts.model is not None:
        try:
            toplot.setcurrent(opts.model)
        except Exception:
            print("Ignoring model, %s is not a PeakStability file." % filename)

    distances = None
    if opts.compare is not None:
        # use baseline from existing file
        distances = readcompare(opts.compare)

    setfigformat(figsize=(6.0, 4.0))
    makeplot(toplot, distances)
    if opts.output:
        plt.savefig(opts.output, format=opts.format, dpi=600)
    if opts.show:
        plt.show()
    else:
        plt.draw()
    return


if __name__ == "__main__":
    main()