Added HistogramViewer panel

Author: dsblank

panels/HistogramViewer/HistogramViewer.py

@@ -0,0 +1,239 @@
+%pip install streamlit_plotly_events
+
+import comet_ml
+from comet_ml.data_structure import Histogram
+import streamlit as st
+import numbers
+import numpy as np
+import plotly.graph_objects as go
+import matplotlib.pyplot as plt
+import io
+import base64
+from streamlit_plotly_events import plotly_events
+
+st.set_page_config(layout="wide", initial_sidebar_state="collapsed")
+
+options = {
+    "start": None,
+    "stop": None,
+    "ybins": None,
+    "xbins": 50,
+    ## Colors are scaled from highest to lowest. You can add
+    ## additional values between 0 and 1 to add color ranges.
+    "colorScale": [
+        [0, "white"],  ## lower values
+        [0.5, "gray"],  ## middle value
+        [1, "blue"],  ## higher values
+    ],
+    "showScale": False,
+    "layout": {
+        "title": "",
+        "xaxis": {"ticks": "", "side": "bottom", "title": "Steps"},
+        "yaxis": {
+            "ticks": "", 
+            "ticksuffix": " ", 
+            "title": "Weights"
+        },
+    },
+}
+
+@st.dialog("Histogram by Step")
+def show_selected_data(selected_data, z, weight_labels, step_labels):
+    y, x = selected_data[0]["pointNumber"]
+    step = step_labels[x]
+    data = [column[x] for column in z]
+    fig = go.Figure(data=go.Bar(
+        y=data,
+        x=weight_labels,
+    ))
+    fig.update_layout(
+        title='Step %s' % step,
+        xaxis_title='Weight',
+        yaxis_title='Count',
+        barmode='group'
+    )
+    st.plotly_chart(fig)
+
+def generate_column_plot(column, z):
+    fig, ax = plt.subplots(figsize=(2, 2))
+    ax.plot([1, 2, 3, 2, 5, 2, 1])
+    ax.axis('off')
+
+    buf = io.BytesIO()
+    fig.savefig(buf, format='png', bbox_inches='tight', pad_inches=0)
+    img_data = base64.b64encode(buf.getbuffer()).decode("utf8")
+    plt.close(fig)
+    return f'<img src="data:image/png;base64,{img_data}">'
+
+def transpose(matrix):
+    return [list(row) for row in zip(*matrix)]
+
+def my_range(start, stop, step):
+    retval = []
+    i = start
+    while i <= stop:
+        retval.append(i)
+        i += step
+    return retval
+
+def getMinMax(histogram):
+    min = None
+    max = None
+    # Find the first/min value:
+    for i in range(len(histogram.counts)):
+        if histogram.counts[i] > 0:
+            min = histogram.values[i - 1]
+            break
+
+    # Find the last/max value:
+    for i in range(len(histogram.counts) - 1, -1, -1):
+        if histogram.counts[i] > 0:
+            max = histogram.values[i + 1]
+            break
+
+    if min is None and max is None:
+        min = -1.0
+        max = 1.0
+    #print(min, max)
+    return [min, max]
+
+
+
+
+def get_sample(length, max_steps):
+    """
+    Get selected/sampled indices
+    """
+    ## Start with all:
+    selected = range(length)
+    if length > max_steps:
+        ## need to sample
+        ## always include the first and last
+        selected = [0] + random.sample(selected[1:-1], max_steps - 2) + [length - 1]
+    return selected
+
+
+def get_histogram_values(asset, start, stop, bins, maxSteps):
+    ## First, collect them:
+    histograms = []
+    ## {'histograms': [{'step': num, 'histogram': {'index_values'}}, ...]
+    index = 0
+    selected_indices = get_sample(len(asset["histograms"]), maxSteps)
+    xValues = []
+    for hist in asset["histograms"]:
+        if index in selected_indices:
+            #print(hist)
+            h = Histogram.from_json(hist["histogram"])
+            xValues.append(hist["step"])
+            histograms.append(h)
+        index += 1
+    ## Next, find the start/stop
+    zValues = []
+    #xValues = []
+    if start is None or stop is None:
+        minimum = None
+        maximum = None
+        for histogram in histograms:
+            minmax = getMinMax(histogram)
+            minimum = min(minimum if minimum is not None else float("+inf"), minmax[0])
+            maximum = max(maximum if maximum is not None else float("-inf"), minmax[1])
+        if start is None:
+            start = minimum
+        if stop is None:
+            stop = maximum
+
+    if bins == None:
+        bins = 50
+
+    for histogram in histograms:
+        #print(start, stop)
+        data = histogram.get_counts(start, stop, (stop - start) / bins)
+        #print(data)
+        zValues.append(data)
+        #print(histogram)
+        #xValues.append(histogram.step)
+
+    yValues = my_range(start, stop, (stop - start) / bins)
+    #print("zValues:", transpose(zValues))
+    return [xValues, yValues, transpose(zValues)]
+
+
+def drawHistogram(asset):
+    [x, y, z] = get_histogram_values(
+        asset, options["start"], options["stop"], options["ybins"], options["xbins"]
+    )
+
+    if len(z[0]) == 0:
+        print("No histogram data available")
+        return
+
+    #sums = []
+    #for step in range(len(z[0])):
+    #    sum = 0.0
+    #    for bin in range(len(z)):
+    #        if isinstance(z[bin][step], numbers.Number):
+    #            sum += z[bin][step]
+    #    sums.append(sum)
+
+    ## This is run on every cell:
+    # hoverText = function(bin, step, item) {
+    #  stepPercentage = (sums[step] === 0
+    #    ? 0
+    #    : (item / sums[step]) * 100
+    #  ).toFixed(2)
+    # return `Step: ${x[step]}<br>
+    # Value: ${y[bin].toFixed(2)}</br>
+    # Count: ${item.toFixed(2)}<br>
+    # Count % of step: ${stepPercentage}%`
+    # }
+
+    #print(x)
+
+    #column_plots = [generate_column_plot(column, z) for column in x]
+    
+    fig = go.Figure(data=go.Heatmap(
+        z=z,
+        x=x,
+        y=y,
+        colorscale=options["colorScale"],
+        showscale=options["showScale"],
+    ))
+
+    fig.update_layout(**options["layout"])
+    selected_data = plotly_events(fig)
+    if selected_data:
+        show_selected_data(selected_data, z, y, x)
+
+
+api = comet_ml.API()
+
+experiments = api.get_panel_experiments()
+
+if len(experiments) == 0:
+    print("No available experiments")
+    st.stop()
+elif len(experiments) == 1:
+    experiment = experiments[0]
+else:
+    experiment = st.sidebar.selectbox(
+        "Experiment:",
+        experiments,
+        format_func=lambda experiment: experiment.name or experiment.id,
+    )
+
+assets = experiment.get_asset_list("histogram_combined_3d")
+
+if len(assets) == 0:
+    print("No available histograms")
+    st.stop()
+elif len(assets) == 1:
+    asset = assets[0]
+else:
+    asset = st.sidebar.selectbox(
+        "Histogram:",
+        sorted(assets, key=lambda item: item["fileName"]),
+        format_func=lambda asset: asset["fileName"],
+    )
+
+histogram = experiment.get_asset(asset["assetId"], return_type="json")
+drawHistogram(histogram)