Use papermill for running tutorials (#1706)

Summary:
## Motivation

We are using nbconvert to run tutorials. nbconvert is not really made for this use case, but papermill is, so we have some handwritten code than can be handled by papermill. With papermill, we can go a bit further and use SMOKE_TEST as a [parameter](https://papermill.readthedocs.io/en/latest/usage-parameterize.html) rather than an environment variable. That would make it easy for people to work with the tutorials as notebooks.

Pull Request resolved: #1706

Test Plan:
Ran tutorials locally and made sure smoke-test flag was getting used appropriately.

## Related pull requests

Enabling papermill will make #1703, which automates running a notebook, a bit easier.

Reviewed By: saitcakmak

Differential Revision: D43631568

Pulled By: esantorella

fbshipit-source-id: e7bfeb68e221fff4f1633af8deb9a11d1ff1c0e6

Author: esantorella

scripts/run_tutorials.py

@@ -10,16 +10,13 @@
 import datetime
 import os
 import subprocess
-import tempfile
 import time
 from pathlib import Path
 from subprocess import CalledProcessError
 from typing import Any, Dict, Optional, Tuple
 
-import nbformat
 import pandas as pd
 from memory_profiler import memory_usage
-from nbconvert import PythonExporter
 
 
 IGNORE_ALWAYS = {  # ignored in smoke tests and full runs
@@ -64,33 +61,18 @@ def get_output_file_path(smoke_test: bool) -> str:
     return fname
 
 
-def parse_ipynb(file: Path) -> str:
-    with open(file, "r") as nb_file:
-        nb_str = nb_file.read()
-    nb = nbformat.reads(nb_str, nbformat.NO_CONVERT)
-    exporter = PythonExporter()
-    script, _ = exporter.from_notebook_node(nb)
-    return script
-
-
 def run_script(
-    script: str, timeout_minutes: int, env: Optional[Dict[str, str]] = None
+    tutorial: Path, timeout_minutes: int, env: Optional[Dict[str, str]] = None
 ) -> None:
-    # need to keep the file around & close it so subprocess does not run into I/O issues
-    with tempfile.NamedTemporaryFile(delete=False) as tf:
-        tf_name = tf.name
-        with open(tf_name, "w") as tmp_script:
-            tmp_script.write(script)
     if env is not None:
         env = {**os.environ, **env}
     run_out = subprocess.run(
-        ["ipython", tf_name],
+        ["papermill", tutorial, "|"],
         capture_output=True,
         text=True,
         env=env,
         timeout=timeout_minutes * 60,
     )
-    os.remove(tf_name)
     return run_out
 
 
@@ -102,13 +84,12 @@ def run_tutorial(
     them as a string, and returns runtime and memory information as a dict.
     """
     timeout_minutes = 5 if smoke_test else 30
-    script = parse_ipynb(tutorial)
     tic = time.monotonic()
     print(f"Running tutorial {tutorial.name}.")
     env = {"SMOKE_TEST": "True"} if smoke_test else None
     try:
         mem_usage, run_out = memory_usage(
-            (run_script, (script, timeout_minutes), {"env": env}),
+            (run_script, (tutorial, timeout_minutes), {"env": env}),
             retval=True,
             include_children=True,
         )

setup.py

@@ -23,6 +23,7 @@
     "kaleido",
     "matplotlib",
     "memory_profiler",
+    "papermill",
     "pykeops",
     "torchvision",
 ]