now testing out multi-app with navigation

Procfile

@@ -1 +1 @@
-web: sh setup.sh && streamlit run beta_distribution.py
+web: sh setup.sh && streamlit run st_runner.py apps

apps/beta_distribution.py

@@ -1,8 +1,11 @@
 import streamlit as st
-from scipy.stats import beta, norm
+from scipy.stats import beta
 import numpy as np
 import matplotlib.pyplot as plt
+import hvplot.pandas  # noqa: F401
+import holoviews as hv
 
+hv.extension("bokeh")
 
 st.header("Beta Distribution Tutorial")
 
@@ -122,7 +125,8 @@ that best explain these three data points?
 The best ratio of alpha to beta is probably around 1:6.
 However, is it 1:6, or is it 6:36, or is it 15:90?
 
-Play around with different ratios to see which one maximizes the log likelihood.
+Play around with different ratios
+to see which one maximizes the log likelihood.
 """
 )
 
@@ -277,7 +281,7 @@ def process_data(data):
     # This is done by capturing the output of stdout and surfacing it to HTML.
     try:
         data = np.array([float(i) for i in data.split(", ")])
-    except ValueError as e:
+    except ValueError:
         raise ValueError("The data that you input must be castable as floats!")
     if not (np.all(data > 0) and np.all(data < 1)):
         raise ValueError("Your input data must be 0 < x < 1.")
@@ -300,7 +304,8 @@ Did you like this mini-tutorial?
 
 If you did, please give it a star on [GitHub](https://github.com/ericmjl/minimal-streamlit-example).
 
-This was hand-crafted using streamlit in under 3 hours.
+This was hand-crafted using streamlit in under 3 hours,
+2.5 of which were spent crafting prose.
 
 Created by [Eric J. Ma](https://ericmjl.github.io).
 """

apps/iris.py

@@ -0,0 +1,27 @@
+from sklearn.datasets import load_iris
+import pandas as pd
+import streamlit as st
+import holoviews as hv
+import hvplot.pandas  # noqa: F401
+
+
+st.header("Holoviews!")
+st.write("As a bonus, I'm going to show you a holoviews plot in streamlit!")
+
+
+def iris_data():
+    data = load_iris()
+    df = pd.DataFrame(data["data"], columns=data["feature_names"])
+    return df
+
+
+df = iris_data()
+
+x_column = st.selectbox("X axis", df.columns)
+y_column = st.selectbox("Y axis", df.columns)
+
+plot = (
+    iris_data().hvplot.scatter(x_column, y_column).opts(width=400, height=400)
+)
+
+st.bokeh_chart(hv.render(plot, backend="bokeh"))

environment.yml

@@ -16,5 +16,7 @@ dependencies:
 - pydocstyle
 - flake8
 - black
+- hvplot
+- bokeh
 - pip:
   - streamlit

st_runner.py

@@ -0,0 +1,51 @@
+import streamlit as st
+import os
+import sys
+import importlib.util
+
+# Parse command-line arguments.
+if len(sys.argv) > 1:
+    folder = os.path.abspath(sys.argv[1])
+else:
+    folder = os.path.abspath(os.getcwd())
+
+# Get filenames for all files in this path, excluding this script.
+
+this_file = os.path.abspath(__file__)
+fnames = []
+
+for basename in os.listdir(folder):
+    fname = os.path.join(folder, basename)
+
+    if fname.endswith(".py") and fname != this_file:
+        fnames.append(fname)
+
+# Make a UI to run different files.
+files = {f.split("/")[-1].split(".")[0]: f for f in fnames}
+filename = st.sidebar.selectbox("Select an app", list(files.keys()))
+fname_to_run = files[filename]
+
+# Create module from filepath and put in sys.modules, so Streamlit knows
+# to watch it for changes.
+
+fake_module_count = 0
+
+
+def load_module(filepath):
+    global fake_module_count
+
+    modulename = "_dont_care_%s" % fake_module_count
+    spec = importlib.util.spec_from_file_location(modulename, filepath)
+    module = importlib.util.module_from_spec(spec)
+    sys.modules[modulename] = module
+
+    fake_module_count += 1
+
+
+# Run the selected file.
+
+with open(fname_to_run) as f:
+    load_module(fname_to_run)
+    filebody = f.read()
+
+exec(filebody, {})