Add poster plot code.

scripts/result/plot_for_5m_poster.py

@@ -0,0 +1,126 @@
+import json
+import matplotlib.pyplot as plt
+import pandas as pd
+import os
+import numpy as np
+from matplotlib.backends.backend_pdf import PdfPages
+import matplotlib
+
+MODALITY2COLORSTR = {
+
+    'DNA to DNA': "tab:blue",
+    'Image to Image': "tab:red",
+    'Image to DNA': "tab:purple",
+}
+MODALITY2COLORSTRRGB = {k: matplotlib.colors.to_rgb(v) for k, v in MODALITY2COLORSTR.items()}
+
+color_map = MODALITY2COLORSTRRGB
+marker_map = {
+    'seen': 'o',
+    'unseen': 'X',
+    'harmonic mean': 's'
+}
+
+query_key_combinations = [['encoded_dna_feature', 'encoded_dna_feature'],
+                          ['encoded_image_feature', 'encoded_image_feature'],
+                          ['encoded_image_feature', 'encoded_dna_feature']]
+
+
+def add_harmonic_mean_acc_to_dict(acc_dict):
+    for query in acc_dict.keys():
+        for key in acc_dict[query].keys():
+
+            if "seen" not in acc_dict[query][key] or "unseen" not in acc_dict[query][key]:
+                continue
+            acc_dict[query][key]['harmonic mean'] = {}
+            for acc_type in ['micro_acc', 'macro_acc']:
+                acc_dict[query][key]['harmonic mean'][acc_type] = {}
+                for top_k in ['1', '3', '5']:
+                    acc_dict[query][key]['harmonic mean'][acc_type][top_k] = {}
+                    seen_acc = acc_dict[query][key]['seen'][acc_type][top_k]
+                    unseen_acc = acc_dict[query][key]['unseen'][acc_type][top_k]
+
+                    for level in ['order', 'family', 'genus', 'species']:
+                        if seen_acc[level] == 0 or unseen_acc[level] == 0:
+                            harmonic_mean = 0
+                        else:
+                            harmonic_mean = 2 / (1 / seen_acc[level] + 1 / unseen_acc[level])
+                        acc_dict[query][key]['harmonic mean'][acc_type][top_k][level] = harmonic_mean
+    return acc_dict
+
+
+def plot_accuracy(acc_dict, output_folder, experiment_name, acc_type='macro_acc', ):
+    plt.figure(figsize=(5, 3.5))
+    taxonomy_levels = ['order', 'family', 'genus', 'species']
+    color_lines = []
+
+    for color_key, color_value in color_map.items():
+        color_line, = plt.plot([], [], color=color_value, label=color_key)
+        color_lines.append(color_line)
+
+    marker_lines = []
+
+    for (query, key), color_map_key in zip(query_key_combinations, color_map.keys()):
+        for split in ['seen', 'unseen', 'harmonic mean']:
+            try:
+                curr_acc_dict_with_four_level = acc_dict[query][key][split][acc_type]['1']
+            except:
+                continue
+
+            curr_acc_list = [curr_acc_dict_with_four_level[level] for level in taxonomy_levels]
+            # time 100 to get percentage
+            curr_acc_list = [acc * 100 for acc in curr_acc_list]
+            plt.plot(taxonomy_levels, curr_acc_list, color=color_map[color_map_key], marker=marker_map[split])
+
+            marker_line, = plt.plot([], [], color='gray', marker=marker_map[split], label=split)
+            marker_lines.append(marker_line)
+
+    if experiment_name == "Image + DNA + Taxonomy":
+
+        color_legend = plt.legend(handles=color_lines, loc='lower left', fontsize=12)
+        plt.gca().add_artist(color_legend)
+
+        marker_legend = plt.legend(handles=marker_lines[:3], loc='lower left', bbox_to_anchor=(0.285, 0), fontsize=12)
+
+    plt.title(f"{experiment_name}", fontsize=14)
+    plt.ylabel("Macro-accuracy (%)", fontsize=14)
+    plt.grid(True, axis="y", which="major", linestyle="-", alpha=0.8)
+    plt.grid(True, axis="y", which="minor", linestyle="-", alpha=0.2)
+    plt.xticks(ticks=np.arange(len(taxonomy_levels)), labels=taxonomy_levels, rotation=45, fontsize=14)
+    plt.tick_params(axis="both", which="major", labelsize=14)
+    plt.grid(True)
+    plt.ylim(0, 100)
+    plt.tight_layout()
+    # save to output folder, in pdf
+    # plt.savefig(f"{output_folder}/{experiment_name}.png")
+    plt.savefig(f"{output_folder}/{experiment_name}.pdf")
+    plt.close()
+
+if __name__ == '__main__':
+    json_root = '/localhome/zmgong/second_ssd/projects/bioscan-clip/extracted_embedding/bioscan_5m/acc_json'
+    list_of_acc_val_json = [
+        f"{json_root}/image_dna_text_4gpu/acc_dict_test.json",
+        f"{json_root}/image_text_4gpu/acc_dict_test.json",
+        f"{json_root}/image_dna_4gpu/acc_dict_test.json",
+        f"{json_root}/no_alignment_baseline/acc_dict_test.json"
+    ]
+
+    experiment_names = ["Image + DNA + Taxonomy",
+                        "Image + DNA",
+                        "Image + Taxonomy",
+                        "no align baseline",]
+
+    line_descriptions = ['DNA to DNA',
+                        'Image to Image',
+                        'Image to DNA']
+    output_folder = 'plots'
+
+    acc_type = 'macro_acc'
+
+    os.makedirs(output_folder, exist_ok=True)
+
+    for path, experiment_name in zip(list_of_acc_val_json, experiment_names):
+        with open(path, "r") as f:
+            acc_dict = json.load(f)
+            acc_dict = add_harmonic_mean_acc_to_dict(acc_dict)
+            plot_accuracy(acc_dict, output_folder, experiment_name, acc_type)