fix diagram (#4)

- fix diagram (0ece20e344b40a4fc527e79fffea56c1ceef4708)

naacl_demo/demo_utils.py

@@ -80,20 +80,20 @@ def prompt_boolq(passage, question, pattern):
 
 def advantage_text(advantage):
     model_type = (
-        """<span style="color: #4B0082">Head</span>"""
+        """<span style="color: #4B0082">分类头法</span>"""
         if advantage < 0
-        else """<span style="color: #daa520">Prompting</span>"""
+        else """<span style="color: #daa520">提示法</span>"""
     )
-    return f"""<b>{model_type}</b> advantage: <b>{abs(advantage):.2f}</b> data points"""
+    return f"""<b>{model_type}</b> 优势: <b>{abs(advantage):.2f}</b> 条样本"""
 
 
 def average_advantage_text(advantage):
     model_type = (
-        """<span style="color: #4B0082">head</span>"""
+        """<span style="color: #4B0082">分类头法</span>"""
         if advantage < 0
-        else """<span style="color: #daa520">prompting</span>"""
+        else """<span style="color: #daa520">提示法</span>"""
     )
-    return f"""<b>Average {model_type}</b> advantage: <b>{abs(advantage):.2f}</b> data points"""
+    return f"""<b>Average {model_type}</b> 优势: <b>{abs(advantage):.2f}</b> 条样本"""
 
 
 def naming_convention(task, seed, pvp_index=None, neutral=False):
@@ -296,13 +296,13 @@ def plot_polygons_bokeh(task, training_points, clf_results, pvp_results, clf_col
     middle_y = (full_range[0] + full_range[1]) / 2
 
     fig = figure(plot_height=400, plot_width=800, max_height=400, max_width=800,
-                 x_axis_type="log" if x_log_scale else "linear", title="Performance over training subset sizes of head and prompting methods")
+                 x_axis_type="log" if x_log_scale else "linear", title="分类头法及提示法在各规模的训练子集上的性能")
 
-    fig.circle(training_points, clf_results, color=clf_colors[0], legend="head run")
-    fig.circle(training_points, pvp_results, color=pvp_colors[0], legend="prompting run")
+    fig.circle(training_points, clf_results, color=clf_colors[0], legend="分类头法")
+    fig.circle(training_points, pvp_results, color=pvp_colors[0], legend="提示法")
     fig.line(training_points, clf_results, color=clf_colors[0], alpha=1)
     fig.line(training_points, pvp_results, color=pvp_colors[0], alpha=1)
-    fig.xaxis.axis_label = "training subset size"
+    fig.xaxis.axis_label = "训练子集规模"
     fig.yaxis.axis_label = task_metrics[task]
     fig.patch(
         [training_points[0], training_points[0], training_points[-1], training_points[-1]],
@@ -310,7 +310,7 @@ def plot_polygons_bokeh(task, training_points, clf_results, pvp_results, clf_col
         color="black",
         fill_alpha=0,
         line_width=0,
-        legend="comparison region",
+        legend="比较区域",
         hatch_alpha=0.14,
         hatch_scale=40,
         hatch_pattern="/",
@@ -356,7 +356,7 @@ def plot_polygons_bokeh(task, training_points, clf_results, pvp_results, clf_col
         location=training_points[-1], dimension="height", line_color="black", line_width=2.5, line_dash="dashed"
     )
     end_label = Label(
-        x=training_points[-1], y=middle_y, text="End of dataset", angle=90, angle_units="deg", text_align="center"
+        x=training_points[-1], y=middle_y, text="数据集总大小", angle=90, angle_units="deg", text_align="center"
     )
     fig.renderers.extend([vline, end_label])
 
@@ -374,12 +374,12 @@ def plot_three_polygons_bokeh(
     middle_y = (full_range[0] + full_range[1]) / 2
 
     fig = figure(plot_height=400, plot_width=800, max_height=400, max_width=800,
-                 x_axis_type="log" if x_log_scale else "linear", title="Performance over training subset sizes of head, prompting and prompting with a null verbalizer")
-    fig.xaxis.axis_label = "training subset size"
+                 x_axis_type="log" if x_log_scale else "linear", title="分类头法、提示法以及空言语器提示法在各规模的训练子集上的性能")
+    fig.xaxis.axis_label = "训练子集规模"
     fig.yaxis.axis_label = task_metrics[task]
-    fig.circle(training_points, clf_results, color=clf_colors[0], legend="head run")
-    fig.circle(training_points, pvp_results, color=pvp_colors[0], legend="prompting run")
-    fig.circle(training_points, ctl_results, color=ctl_colors[0], legend="null verbalizer run")
+    fig.circle(training_points, clf_results, color=clf_colors[0], legend="分类头法")
+    fig.circle(training_points, pvp_results, color=pvp_colors[0], legend="提示法")
+    fig.circle(training_points, ctl_results, color=ctl_colors[0], legend="空言语器提示法")
     fig.line(training_points, clf_results, color=clf_colors[0], alpha=1)
     fig.line(training_points, pvp_results, color=pvp_colors[0], alpha=1)
     fig.line(training_points, ctl_results, color=ctl_colors[0], alpha=1)
@@ -390,7 +390,7 @@ def plot_three_polygons_bokeh(
         color="black",
         fill_alpha=0,
         line_width=0,
-        legend="comparison region",
+        legend="比较区域",
         hatch_alpha=0.14,
         hatch_scale=40,
         hatch_pattern="/",
@@ -447,7 +447,7 @@ def plot_three_polygons_bokeh(
         location=training_points[-1], dimension="height", line_color="black", line_width=2.5, line_dash="dashed"
     )
     end_label = Label(
-        x=training_points[-1], y=middle_y, text="End of dataset", angle=90, angle_units="deg", text_align="center"
+        x=training_points[-1], y=middle_y, text="数据集总大小", angle=90, angle_units="deg", text_align="center"
     )
     fig.renderers.extend([vline, end_label])
 
@@ -458,7 +458,7 @@ def plot_three_polygons_bokeh(
 
 def pattern_graph(task):
     fig = figure(plot_height=400, plot_width=800, max_height=400, max_width=800, x_axis_type="log", title="Performance over training subset sizes of different prompt patterns")
-    fig.xaxis.axis_label = "training subset size"
+    fig.xaxis.axis_label = "训练子集规模"
     fig.yaxis.axis_label = task_metrics[task]
     url = f"https://raw.githubusercontent.com/TevenLeScao/pet/master/exported_results/{task.lower()}/wandb_export.csv"
     df = pd.read_csv(url)
@@ -488,7 +488,7 @@ def pattern_graph(task):
         y_max = list([np.max(training_point_df[task_metrics[task]]) for training_point, training_point_df in gby_training_points])
         y_min = list([np.min(training_point_df[task_metrics[task]]) for training_point, training_point_df in gby_training_points])
         y = list([np.median(training_point_df[task_metrics[task]]) for training_point, training_point_df in gby_training_points])
-        fig.circle(x, y, color=pattern_colors[i], alpha=1, legend=f"Pattern {i}")
+        fig.circle(x, y, color=pattern_colors[i], alpha=1, legend=f"模式 {i}")
         fig.line(x, y, color=pattern_colors[i], alpha=1)
         fig.varea(x=x, y1=y_max, y2=y_min, color=pattern_colors[i], alpha=0.11)
         # source = ColumnDataSource(data=dict(base=x, lower=y_min, upper=y_max))

naacl_demo/main.py

@@ -100,7 +100,7 @@ advantage_tabs = []
 advantage_all_figures = Tabs(tabs=advantage_tabs)
 
 advantage_box = Div(
-    text="Click within the comparison region to compute the data advantage for a performance level",
+    text="在比较区域内点击某点以计算该点对应的性能点上的数据优势",
     width=text_width,
     style=box_style,
     sizing_mode="scale_width",
@@ -200,13 +200,13 @@ def on_integrate_click():
         advantage_box.text = average_advantage_text(average_advantage)
 
 
-integrate = Button(width=175, max_width=175, label="Integrate over the whole region!")
+integrate = Button(width=175, max_width=175, label="对整个区域进行积分！")
 integrate.align = "center"
 integrate.on_click(on_integrate_click)
 
 
 def on_tab_change(attr, old, new):
-    advantage_box.text = "Click within the comparison region to compute the data advantage for a performance level"
+    advantage_box.text = "在比较区域内点击某点以计算该点对应的性能点上的数据优势"
 
 
 advantage_all_figures.on_change('active', on_tab_change)