try/except selenium

Author: linogaliana

_quarto.yml

@@ -7,7 +7,6 @@ project:
     - content/manipulation/04a_webscraping_TP.qmd
     - content/modelisation/index.qmd
     - content/visualisation/index.qmd
-    - content/visualisation/matplotlib.qmd
     - content/annexes/evaluation.qmd
 
 profile:

content/getting-started/intro/_pourquoi_python_ia.qmd

@@ -33,9 +33,9 @@ Because coding is not just about writing lines of code. It’s about understandi
 
 Working with data is first and foremost an engineering process. Code is not the goal—it’s the tool that supports structured reasoning toward solving real-world problems. Just like an engineer designs a bridge to meet a practical need, a data scientist begins with an operational goal—such as building a recommendation system, evaluating the impact of a product launch, or forecasting sales—and reformulates it into an analytical task. This means translating scientific or business ideas into a set of questions, then breaking those down into logical steps, each of which can be executed by a computer.
 
-In this context, an LLM can act as a valuable assistant—but only when the problem is well formulated. If the task is vague or ill-defined, the model’s answers will be approximate or even useless. On standard problems, the results may appear accurate. But for more specific, non-standard tasks, it often becomes necessary to iterate, refine the prompt, reframe the problem… and sometimes still fail to get a satisfactory result. Not because the model is poor, but because good problem formulation—the essence of problem engineering—makes all the difference[^ai-enstein].
+In this context, an LLM can act as a valuable assistant—but only when the problem is well formulated. If the task is vague or ill-defined, the model’s answers will be approximate or even useless. On standard problems, the results may appear accurate. But for more specific, non-standard tasks, it often becomes necessary to iterate, refine the prompt, reframe the problem… and sometimes still fail to get a satisfactory result. Not because the model is poor, but because good problem formulation—the essence of problem engineering—makes all the difference[^ai-enstein-en].
 
-[^ai-enstein]: On this topic, see Thomas Wolf’s blog post [_The Einstein AI model_](https://thomwolf.io/blog/scientific-ai.html). Although the post focuses on disruptive innovation and pays less attention to incremental progress, it’s insightful in understanding that LLMs—despite bold predictions from tech influencers—are still just tools. They may excel at standardized tasks, but for now, they remain assistants.
+[^ai-enstein-en]: On this topic, see Thomas Wolf’s blog post [_The Einstein AI model_](https://thomwolf.io/blog/scientific-ai.html). Although the post focuses on disruptive innovation and pays less attention to incremental progress, it’s insightful in understanding that LLMs—despite bold predictions from tech influencers—are still just tools. They may excel at standardized tasks, but for now, they remain assistants.
 
 For instance, in the year 2025, [`uv`](https://docs.astral.sh/uv/) saw rapid adoption, as did [`ruff`](https://docs.astral.sh/ruff/) the year before. It will still be some time before generative AIs propose this environment manager on their own, rather than [`poetry`](https://python-poetry.org/). The existence of generative AIs does not, therefore, dispense us, as before, from keeping an active technical watch and being vigilant about changes in practices.
 :::

content/manipulation/04a_webscraping_TP.qmd

@@ -1601,10 +1601,33 @@ have performed the correct action:
 :::
 
 ```{python}
-#| output: false
+#| eval: false
 png = browser.get_screenshot_as_png()
 ```
 
+```{python}
+#| output: false
+#| echo: false
+import time
+from selenium.common.exceptions import WebDriverException
+
+def get_screenshot_retry(browser, max_retries=3, delay=2):
+    for attempt in range(1, max_retries + 1):
+        try:
+            png = browser.get_screenshot_as_png()
+            return png   # succès → on sort
+        except WebDriverException as e:
+            print(f"Échec {attempt}/{max_retries} : {e}")
+            if attempt < max_retries:
+                time.sleep(delay)
+            else:
+                raise  # après le dernier essai, on relance l’erreur
+
+# Utilisation
+png = get_screenshot_retry(browser)
+```
+
+
 ```{python}
 from IPython.display import Image
 Image(png, width='500')