Create Benchmarking Jupytext.py

#334

demo/Benchmarking Jupytext.py

@@ -0,0 +1,92 @@
+# In this notebook, we benchmark the Jupytext formats for Jupyter notebooks against the base format
+# Open this script as a notebook in Jupyter to run it and see the plots
+
+import time
+import copy
+import pandas as pd
+import plotly.graph_objects as go
+from plotly.colors import DEFAULT_PLOTLY_COLORS
+
+import nbformat
+import jupytext
+
+# The notebook to be tested
+notebook = jupytext.read('World population.ipynb')
+
+# Same notebook, with no outputs, for a fair comparison
+notebook_no_outputs = copy.deepcopy(notebook)
+for cell in notebook_no_outputs.cells:
+    cell.outputs = []
+    cell.execution_count = None
+
+# +
+JUPYTEXT_FORMATS = ['ipynb', 'md', 'py:light', 'py:percent', 'py:sphinx']
+
+# Let's see if we have pandoc here
+try:
+    jupytext.writes(notebook, fmt='md:pandoc')
+    JUPYTEXT_FORMATS.append('md:pandoc')
+except jupytext.formats.JupytextFormatError as err:
+    print(str(err))
+
+
+# -
+
+
+def sample_perf(nb, n=30):
+    samples = pd.DataFrame(
+        pd.np.NaN,
+        index=pd.MultiIndex.from_product(
+            (range(n), ['nbformat'] + JUPYTEXT_FORMATS), names=['sample', 'implementation']),
+        columns=pd.Index(['size', 'read', 'write'], name='measure'))
+
+    for i, fmt in samples.index:
+        t0 = time.time()
+        if fmt == 'nbformat':
+            text = nbformat.writes(nb)
+        else:
+            text = jupytext.writes(nb, fmt)
+        t1 = time.time()
+        samples.loc[(i, fmt), 'write'] = t1 - t0
+        samples.loc[(i, fmt), 'size'] = len(text)
+        t0 = time.time()
+        if fmt == 'nbformat':
+            nbformat.reads(text, as_version=4)
+        else:
+            jupytext.reads(text, fmt)
+        t1 = time.time()
+        samples.loc[(i, fmt), 'read'] = t1 - t0
+    return samples
+
+
+def performance_plot(perf, title):
+    formats = ['nbformat'] + JUPYTEXT_FORMATS
+    mean = perf.groupby('implementation').mean().loc[formats]
+    std = perf.groupby('implementation').std().loc[formats]
+    data = [go.Bar(x=mean.index,
+                   y=mean[col],
+                   error_y=dict(
+                       type='data',
+                       array=std[col],
+                       color=color,
+                       thickness=0.5
+                   ) if col != 'size' else dict(),
+                   name=col,
+                   yaxis={'read': 'y1', 'write': 'y2', 'size': 'y3'}[col])
+            for col, color in zip(mean.columns, DEFAULT_PLOTLY_COLORS)]
+    layout = go.Layout(title=title,
+                       xaxis=dict(title='Implementation', anchor='y3'),
+                       yaxis=dict(domain=[0.7, 1], title='Read (secs)'),
+                       yaxis2=dict(domain=[0.35, .65], title='Write (secs)'),
+                       yaxis3=dict(domain=[0, .3], title='Size')
+                       )
+    return go.Figure(data=data, layout=layout)
+
+
+perf_no_outputs = sample_perf(notebook_no_outputs, 30)
+
+performance_plot(perf_no_outputs, 'Benchmarking Jupytext on the World Population notebook<br>(Outputs filtered)')
+
+perf = sample_perf(notebook, 30)
+
+performance_plot(perf, 'Benchmarking Jupytext on the World Population notebook<br>(With outputs)')