Add plot and function for test loss and accuracy

visualize.py

@@ -56,10 +56,23 @@ def filter_df(df):
     Args:
         df (pd.DataFrame): info issue from the json file
     """
+    df = df[df['test_losses'].isnull()]
     drop_indexes = df[df.val_losses.apply(lambda x:any([None in y for y in x]))].index
     return df.drop(drop_indexes)
 
 
+def filter_df_test(df):
+    """
+    return the same df while dropping any row whose validation loss contains Nan values
+    
+    Args:
+        df (pd.DataFrame): info issue from the json file
+    """
+    df = df[~df['test_losses'].isnull()]
+    drop_indexes = df[df.test_losses.apply(lambda x:any([None in y for y in x]))].index
+    return df.drop(drop_indexes)
+
+
 def plot_losses_fits(losses, 
                      ax, 
                      colors,
@@ -148,6 +161,21 @@ def get_concat_losses(df,train : bool):
         concat_losses = concat_losses + losses
     return concat_losses
 
+
+def get_concat_test_losses(df,train : bool):
+    """
+    concat the losses for the given df into a single list
+    
+    Args:
+        df (pd.DataFrame): info issue from the json file
+        train (Bool): train or validation set
+    """
+    concat_losses = []
+    tot_losses = df.train_losses if train else df.test_losses
+    for losses in tot_losses.values:
+        concat_losses = concat_losses + losses
+    return concat_losses
+
 def plot_grid_search(df,
                      ax,
                      plot_SGD=True,
@@ -192,3 +220,50 @@ def plot_grid_search(df,
     bottom,top = ax.get_ylim()
     ax.set_yticks(np.linspace(bottom,top,5))
     ax.get_yaxis().set_major_formatter(matplotlib.ticker.ScalarFormatter())
+
+
+def plot_grid_search_test(df,
+                         ax,
+                         plot_SGD=True,
+                         plot_Adam=True,
+                         plot_AdamW=True,
+                         plot_runs=True,
+                         plot_fit=False,
+                         plot_mean=False,
+                         fit_type="log",
+                         train=True):
+    """
+    plot the runs of the computed grid search according to many display parameters
+    
+    Args:
+        df (pd.DataFrame): info issue from the json file
+        ax (plt.ax): matplotlib ax
+        plot_XXX (bool): whether to plot the XXX optimizer
+        plot_runs (bool): plot the runs
+        plot_fit (bool): plot the fitted curve
+        plot_mean (bool): plot the mean of the runs
+        fit_type (bool): whether to fit the regression in a log-log or lin-log space
+        train (Bool): train or validation set    
+    """
+
+
+    args = [plot_runs,plot_fit,plot_mean,fit_type]
+    if plot_SGD:
+        sgd_df = df[df.optimizer == 'SGD']
+        sgd_concat_losses = get_concat_test_losses(sgd_df,train)
+        col1 = (colSGD,"#00ff3c")
+        reg1 = plot_losses_fits(sgd_concat_losses, ax, col1,*args,label="SGD")
+    if plot_Adam:
+        adam_df = df[df.optimizer == 'Adam']
+        adam_concat_losses = get_concat_test_losses(adam_df,train)
+        col2 = (coladam,"#08f0fc")
+        reg2 = plot_losses_fits(adam_concat_losses, ax,col2,*args,label="Adam")
+    if plot_AdamW:
+        adamW_df = df[df.optimizer == 'AdamW']
+        adamW_concat_losses = get_concat_test_losses(adamW_df,train)
+        col3 = (coladamW, "yellow")
+        reg3 = plot_losses_fits(adamW_concat_losses, ax,col3,*args,label="AdamW")
+    bottom,top = ax.get_ylim()
+    ax.set_yticks(np.linspace(bottom,top,5))
+    ax.get_yaxis().set_major_formatter(matplotlib.ticker.ScalarFormatter())
+