psychometric reduce bins and log

Author: tudou2333333

To do.txt

@@ -22,8 +22,8 @@ It might improve with a better prompt, but it looks like a good plan for now. 
 and port where the animal comes from. Explore first the correlations of these regressors.
 For example, in correct trials, the three of them should be the same.√
 
-- Psychometric for auditory. Test using log value of the evidence strength.
-Reduce a bit the number of bins to make it less noisy in the middle. Explore other
+- Psychometric for auditory. Test using log value of the evidence strength.√(The values are under 1, so their values will be changed after log)
+Reduce a bit the number of bins to make it less noisy in the middle.√ Explore other
 properties of the cloud of tones, and the impact they have on decisions. For instance,
 several tones can sound simultaneously in the same timebin. One question is:
 are mice affected more by the total number of tones, or by the total number of timebins

lecilab_behavior_analysis/df_transforms.py

@@ -696,7 +696,7 @@ def mean_and_cis_of_rt_and_tbt(group):
 
 
 def get_choice_before(df):
-    df_copy = df.copy(deep=False)
+    df_copy = df.copy(deep=True)
     utils.column_checker(df_copy, required_columns={"first_choice", "last_choice", "previous_port_before_stimulus", "correct"})
 
     for mouse in df_copy['subject'].unique():
@@ -766,7 +766,7 @@ def parameters_for_fit(df):
     """
     Get the parameters for the fit
     """
-    df_copy = df.copy(deep=False)
+    df_copy = df.copy(deep=True)
     df_copy = add_mouse_first_choice(df_copy)
     df_copy = add_mouse_last_choice(df_copy)
     df_copy = add_port_where_animal_comes_from(df_copy)
@@ -816,7 +816,7 @@ def parameters_for_fit(df):
     return df_copy
 
 def get_time_kernel_impact(df:pd.DataFrame, y: str, max_lag, tau):
-    df_copy = df.copy(deep=False)
+    df_copy = df.copy(deep=True)
     if y == 'first_choice_numeric':
         df_copy = add_mouse_first_choice(df_copy)
         df_copy['first_choice_numeric'] = df_copy['first_choice'].apply(

lecilab_behavior_analysis/plots.py

@@ -342,24 +342,25 @@ def psychometric_plot(df: pd.DataFrame, x, y, ax: plt.Axes = None,
     column_checker(df, required_columns={x, y})
     df_copy = df.copy(deep=True)
     if valueType == 'discrete':
-        df_copy[x] = np.sign(df_copy[x]) * (np.log(abs(df_copy[x])).round(4))
+        df_copy[x + "_fit"] = np.sign(df_copy[x]) * (np.log(abs(df_copy[x])).round(4))
         ax.set_xlabel('log_' + x)
     else:
         # bin the continuous values when valueType is not discrete
-        df_copy[x + "_binned"] = pd.cut(df_copy[x], bins = 6, labels=False)
-        x = x + "_binned"
+        bins = pd.cut(df_copy[x], bins = 6)
+        labels = df_copy[x].groupby(bins).mean()
+        df_copy[x + "_fit"] = pd.cut(df_copy[x], bins = 6, labels = labels).astype(float)
+        # df_copy[x + "_fit"] = np.sign(df_copy[x + "_fit"]) * (np.log(abs(df_copy[x + "_fit"])).round(4))
         ax.set_xlabel(x)
     sns.pointplot(
-        x=x,
+        x=x + "_fit",
         y=y,
         data=df_copy,
         estimator=lambda x: np.mean(x),
         ax=ax,
         **point_kwargs
     )
-
-    xs = np.linspace(df_copy[x].min(), df_copy[x].max(), 100).reshape(-1, 1)
-    p_left, fitted_params = utils.fit_lapse_logistic_independent(df_copy[x], df_copy[y])
+    xs = np.linspace(df_copy[x + "_fit"].min(), df_copy[x + "_fit"].max(), 100).reshape(-1, 1)
+    p_left, fitted_params = utils.fit_lapse_logistic_independent(df_copy[x + "_fit"], df_copy[y])
     ax.plot(xs, p_left, **line_kwargs)
     ax.set_ylabel(y)
     ax.set_ylim(0, 1)