need to correct IPW code for y and t input format

[judithabk6]

IPW implementation behaves badly if t and y dimension is (n, 1), and ok if t and y dimension is (n, 0), please fix

[sami6mz]

I just realized IPW is not the only one to fail.
g_computation, multiply_robust, simulation_based return ValueError if .ravel() is not applied to t and y.
I don't know about others estimators yet, but I will soon implement a test in response to #15.

[sami6mz]

I justed tested .ravel() robustness for all estimators
G_estimator badly behaves too (in addition to IPW, g_computation, multiply_robust, and simulation_based)

[sami6mz]

As for multiply_robust, (n,1) format makes the function to fail because some tr variable is not initialised :
l688 - l689

if len(t.shape) == 1:
        tr = t.reshape(-1, 1)

Instead of :

if len(t.shape) == 1:
        tr = t.reshape(-1, 1)
else:
        tr = np.copy(t)
        t = t.ravel()

Just like m.

Besides, why copying t or m is needed?

[sami6mz]

Besides, why copying t or m is needed?

Okay I realized get_interactions didn't support (n,) inputs so that's why the copying t and m was needed.
Issue was made in #34, and is to be solved in #35.

[sami6mz]

Just a reminder, as for now :

• IPW returns aberrant indirect effects if not given .ravel inputs
• g_computation, simulation_based, G_estimator return an error

[judithabk6]

need to modify the test accordingly when solved, in particular

med_bench/tests/estimation/test_get_estimation.py

Line 185 in 13c97da

return data[1].ravel()

and

med_bench/tests/estimation/test_get_estimation.py

Line 195 in 13c97da

return data[3].ravel() # same reason as t

(remove the .ravel())