Automatically adjust the number of subplots based on the number of li…

…braries

src/scripts/step4_analysis_retriever.py

@@ -1,41 +1,49 @@
-import pandas as pd
-import matplotlib.pyplot as plt
-
-df = pd.read_csv('plot/retriever_topk_results.csv')
-
-columns_to_keep = ['retrieved_api_nums', 'Validation Accuracy', 'Test Accuracy', 'val ambiguous Accuracy', 'test ambiguous Accuracy'] # 'Training Accuracy', 'Training ambiguous Accuracy', 
-
-labels = [
-    'Synthetic instruction', 'Annotation instruction',
-    'Synthetic instruction w/ ambiguity removal', 'Annotation instruction w/ ambiguity removal'
-]
-
-num_rows = 2
-num_cols = 2
-
-libs = df['LIB'].unique()[:num_rows * num_cols]
 
-fig, axs = plt.subplots(num_rows, num_cols, figsize=(15, 10))
-plt.subplots_adjust(bottom=0.25)  # 调整底部空间
+"""
+Author: Zhengyuan Dong
+Created Date: 2024-02-01
+Last Edited Date: 2024-04-08
+Description: 
+    Plot the comparison of the retriever accuracy results for different libraries and models.
+    Automatically adjust the number of subplots based on the number of libraries.
+"""
 
+import pandas as pd
+import matplotlib.pyplot as plt
+import numpy as np
+import math
+
+df = pd.read_csv('./output/retriever_topk_results.csv')
+columns_to_keep = ['retrieved_api_nums', 'Validation Accuracy', 'Test Accuracy', 'val ambiguous Accuracy', 'test ambiguous Accuracy']
+labels = ['Synthetic instruction', 'Annotation instruction',
+          'Synthetic instruction w/ ambiguity removal', 'Annotation instruction w/ ambiguity removal']
+libs = df['LIB'].unique()
+num_libs = len(libs)
+if num_libs <= 2:
+    fig, axs = plt.subplots(1, num_libs, figsize=(15 * num_libs, 6))
+    if num_libs == 1:
+        axs = np.array([axs])
+else:
+    num_rows = np.ceil(num_libs / 2).astype(int)
+    fig, axs = plt.subplots(num_rows, 2, figsize=(15, 6 * num_rows))
+    axs = axs.flatten()
 for index, lib in enumerate(libs):
-    ax = axs.flatten()[index]
-    lib_df = df[df['LIB'] == lib][columns_to_keep]
+    ax = axs[index]
+    lib_df = df[df['LIB'] == lib]
     for col, label in zip(columns_to_keep[1:], labels):
         ax.plot(lib_df['retrieved_api_nums'], lib_df[col], label=label)
     ax.set_title(f'Fine-tuned Retriever Accuracy v.s. Topk for {lib}')
     ax.set_xlabel('Topk')
     ax.set_ylabel('Accuracy')
-
-# 用flatten()将axs从2D数组转化为1D数组以便简单索引
+    ax.grid(True)
+if num_libs > 1:
+    for ax in axs[num_libs:]:
+        ax.set_visible(False)
 handles, labels = axs.flatten()[0].get_legend_handles_labels()
-
-# 在整个Figure的底部绘制一个共享的图例
 fig.legend(handles, labels, loc='upper center', bbox_to_anchor=(0.5, 0.1), ncol=4)
-
-# Tighten the layout with the rect parameter to fit the new legend
 plt.tight_layout(rect=[0, 0.1, 1, 1])
-
-plt.savefig('./plot/retriever_acc_lib_4.jpg')
+plt.subplots_adjust(hspace=0.4)
+plt_path = "./output/step4_analysis_retriever_acc_lib.jpg"
+plt.savefig(plt_path)
 plt.show()
-
+plt_path

src/scripts/step4_analysis_retriever.sh

@@ -3,11 +3,11 @@
 # usage: bash -x scripts/step4_analysis_retriever.sh
 
 export HUGGINGPATH=./hugging_models
-libs=("scanpy" "squidpy" "ehrapy" "snapatac2")
+libs=("scanpy" "squidpy" "ehrapy" "snapatac2") # "scanpy_subset"
 
 mkdir -p output
-mkdir -p plot/
-csv_file="plot/retriever_topk_results.csv"
+mkdir -p output/
+csv_file="output/retriever_topk_results.csv"
 
 echo "LIB,retrieved_api_nums,Training Accuracy,Validation Accuracy,Test Accuracy,Training ambiguous Accuracy,val ambiguous Accuracy,test ambiguous Accuracy" > $csv_file
 

src/scripts/step5_analysis_compare_retriever.py

@@ -1,75 +1,79 @@
+
+"""
+Author: Zhengyuan Dong
+Created Date: 2024-02-01
+Last Edited Date: 2024-04-08
+Description: 
+    Plot the comparison of the retriever accuracy results for different libraries and models.
+    Automatically adjust the number of subplots based on the number of libraries.
+"""
+
 import pandas as pd
 import matplotlib.pyplot as plt
 import numpy as np
-# Based on the provided data structure and the requirements for subplots, let's create the desired bar graph.
+import math
 
 # Load the CSV data
 df = pd.read_csv('output/retriever_accuracy_results.csv')
 default_colors = plt.rcParams['axes.prop_cycle'].by_key()['color']
 color_b = default_colors[:4]
-#color_b = ['blue', 'orange', 'green', 'red']
-
+legend_labels = ['Synthetic instruction', 'Annotated instruction',
+                 'Synthetic instruction w/ ambiguity removal', 'Annotated instruction w/ ambiguity removal']
 # Define the function for plotting
 def create_subplot_bar_graphs(df, ax, lib, title, colors, display_test=False):
-    # Columns for BM25, Un-Finetuned, and Finetuned (Val and Test)
     cols = [f'{lib} Val', f'{lib} Test', f'{lib} Ambiguous Val', f'{lib} Ambiguous Test']
-
     x_labels = ['BM25', 'SENTENCE-BERT \nw/o \nfine-tuning', 'SENTENCE-BERT \nw/ \nfine-tuning']
-
-    # The width of the bars
     bar_width = 0.15
-
-    # Set the positions of the bars
     indices = np.arange(len(x_labels))
-
-    # Plot bars
     for i, col_prefix in enumerate(['BM25', 'Un-Finetuned', 'Finetuned']):
-        #print('----------', col_prefix)
         val_acc = df.loc[df['LIB'] == lib, f'{col_prefix} Val'].values[0]
         test_acc = df.loc[df['LIB'] == lib, f'{col_prefix} Test'].values[0]
-        #if col_prefix!='BM25':
         val_am_acc = df.loc[df['LIB'] == lib, f'{col_prefix} Ambiguous Val'].values[0]
         test_am_acc = df.loc[df['LIB'] == lib, f'{col_prefix} Ambiguous Test'].values[0]
-
-        ax.bar(indices[i] - 3*bar_width/2, val_acc, bar_width, label='Synthetic instruction' if i==0 else "", color=colors[0])
+        ax.bar(indices[i] - 3*bar_width/2, val_acc, bar_width, color=colors[0])
         if display_test:
-            ax.bar(indices[i] - bar_width/2, test_acc, bar_width, label='Annotated instruction' if i==0 else "", color=colors[1])
-        #if col_prefix!='BM25':
-        ax.bar(indices[i] + bar_width/2, val_am_acc, bar_width, label='Synthetic instruction w/ ambiguity removal' if i==1 else "", color=colors[2])
+            ax.bar(indices[i] - bar_width/2, test_acc, bar_width, color=colors[1])
+        ax.bar(indices[i] + bar_width/2, val_am_acc, bar_width, color=colors[2])
         if display_test:
-            ax.bar(indices[i] + 3*bar_width/2, test_am_acc, bar_width, label='Annotated instruction w/ ambiguity removal' if i==1 else "", color=colors[3])
-
-        # Add data labels
+            ax.bar(indices[i] + 3*bar_width/2, test_am_acc, bar_width, color=colors[3])
         ax.text(indices[i] - 3*bar_width/2, val_acc, f'{val_acc:.2f}', ha='center', va='bottom', fontsize=8)
         if display_test:
             ax.text(indices[i] - bar_width/2, test_acc, f'{test_acc:.2f}', ha='center', va='bottom', fontsize=8)
-        #if col_prefix!='BM25':
         ax.text(indices[i] + bar_width/2, val_am_acc, f'{val_am_acc:.2f}', ha='center', va='bottom', fontsize=8)
         if display_test:
             ax.text(indices[i] + 3*bar_width/2, test_am_acc, f'{test_am_acc:.2f}', ha='center', va='bottom', fontsize=8)
-
-    # Set the title, x-ticks, and labels
     ax.set_title(f'{lib} {title}')
     ax.set_xticks(indices)
     ax.set_xticklabels(x_labels)
-
-    # Set the legend only for the first subplot
-    #if lib == df['LIB'].unique()[0]:
-    #    ax.legend()
-
-# Set up the subplots
-#fig, axs = plt.subplots(2, 2, figsize=(12, 10))
-fig, axs = plt.subplots(2, 2, figsize=(12, 10), constrained_layout=True)
-axs = axs.ravel()  # Flatten the array of axes for easier indexing
-
-fig.subplots_adjust(right=0.8)
-for i, lib in enumerate(df['LIB'].unique()):
+libs = df['LIB'].unique()
+num_libs = len(libs)
+# Calculate the number of rows and columns for the subplots
+num_cols = 2
+num_rows = math.ceil(num_libs / num_cols)
+if num_libs <= 2:
+    # Adjust subplot layout for 1 or 2 libs
+    fig, axs = plt.subplots(1, num_libs, figsize=(12 * num_libs, 5), constrained_layout=True)
+    if num_libs == 1:
+        axs = np.array([axs])
+else:
+    # Calculate the number of rows and columns for the subplots for more than 2 libs
+    num_cols = 2
+    num_rows = math.ceil(num_libs / num_cols)
+    fig, axs = plt.subplots(num_rows, num_cols, figsize=(12, num_rows * 5), constrained_layout=True)
+    axs = axs.ravel()
+for i, lib in enumerate(libs):
+    ax = axs[i]
     create_subplot_bar_graphs(df, axs[i], lib, 'Prediction Accuracy', color_b, display_test=True)
-
-handles, labels = axs[0].get_legend_handles_labels()
-fig.legend(handles, labels, loc='lower center', bbox_to_anchor=(0.5, 0), ncol=4, frameon=False)
-plt.subplots_adjust(bottom=0.2)
-#plt.tight_layout()
-plt.tight_layout(rect=[0, 0.05, 1, 1])
-plt.savefig('./output/step5_analysis_compare_retriever.jpg')
+for j in range(i + 1, len(axs)):
+    fig.delaxes(axs[j])
+# Adjust layout and add a global legend at the bottom
+plt.subplots_adjust(bottom=0.25, top=0.9)
+# Add a global legend at the bottom
+handles = [plt.Rectangle((0,0),1,1, color=color) for color in color_b[:len(legend_labels)]]
+fig.legend(handles, legend_labels, loc='upper center', bbox_to_anchor=(0.5, 0.1), ncol=4)
+plt.tight_layout(rect=[0, 0.1, 1, 1])
+plt.subplots_adjust(hspace=0.4)  # Adjust this value to make space for the legend
+plt.savefig('output/step5_analysis_compare_retriever.jpg')
 plt.show()
+
+

src/scripts/step5_analysis_compare_retriever.sh

@@ -3,7 +3,7 @@
 # Usage: bash -x scripts/step5_analysis_compare_retriever.sh
 
 export HUGGINGPATH=./hugging_models
-libs=("scanpy" "squidpy" "ehrapy" "snapatac2")
+libs=("scanpy" "squidpy" "ehrapy" "snapatac2") # scanpy_subset
 csv_file="output/retriever_accuracy_results.csv"
 
 # Header for CSV file