Optimization fails for selected_times is not None (i.e. a subset time range)

[rajabell]

We are encountering an issue when trying to optimize for a subset of data:

When setting a time window that does not contain data for all of the channels in the model, the Summerizer works fine but the Optimizer frequently fails. It doesn't seem to matter if we force any channels to zero in the constraints:

[/usr/local/lib/python3.11/dist-packages/meridian/analysis/optimizer.py](https://localhost:8080/#) in _create_response_curves_section(self)
    867     response_curves = formatter.ChartSpec(
    868         id=summary_text.OPTIMIZED_RESPONSE_CURVES_CHART_ID,
--> 869         chart_json=self.plot_response_curves(
    870             n_top_channels=n_channels
    871         ).to_json(),

[/usr/local/lib/python3.11/dist-packages/meridian/analysis/optimizer.py](https://localhost:8080/#) in plot_response_curves(self, n_top_channels)
    418     else:
    419       title = summary_text.INC_KPI_LABEL
--> 420     df = self._get_plottable_response_curves_df(n_top_channels=n_top_channels)
    421     base = (
    422         alt.Chart(df)

[/usr/local/lib/python3.11/dist-packages/meridian/analysis/optimizer.py](https://localhost:8080/#) in _get_plottable_response_curves_df(self, n_top_channels)
    577     })
    578 
--> 579     response_curves_ds = self.get_response_curves()
    580     response_curves_df = (
    581         response_curves_ds.to_dataframe()

[/usr/local/lib/python3.11/dist-packages/meridian/analysis/optimizer.py](https://localhost:8080/#) in get_response_curves(self)
    536     # max upper spend constraint + padding.
    537     upper_limit = max(max(upper_bound) + c.SPEND_CONSTRAINT_PADDING, 2)
--> 538     spend_multiplier = np.arange(0, upper_limit, c.RESPONSE_CURVE_STEP_SIZE)
    539     # WARN: If `selected_times` is not None (i.e. a subset time range), this
    540     # response curve computation might take a significant amount of time.

ValueError: arange: cannot compute length
``

The error "ValueError: arange: cannot compute length" arises when np.arange cannot determine the number of elements to generate because the step size is zero or leads to an infinite loop. In this context, within the _get_plottable_response_curves_df method of the BudgetOptimizer class, the code uses np.arange(0, upper_limit, c.RESPONSE_CURVE_STEP_SIZE) to generate spend multipliers for plotting response curves. The error suggests that c.RESPONSE_CURVE_STEP_SIZE is likely zero or leading to an infinite loop, preventing the calculation of the range length.

[cpulavarthi]

Hello @rajabell,

Thank you for contacting us!

I am unable to pinpoint the root cause of the error you are experiencing and would like some additional information to better understand the issue. Please share the following details:

1. You have mentioned that the Optimizer fails frequently? Can you let me know if it is working fine sometimes but throws an error some other times?
2. Can you let me know the reason for unavailability of the data? Was there media execution during that period and you don’t have the data or was there no media execution for those channels in those time periods?
3. Can you share the code you are using to set the time window and how you are running the budget optimizer?

Please share the above requested information to help debug this issue.

Thank you

Google Meridian Support Team