File size: 22,017 Bytes
78f20e6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
import os
import shutil
import re
import json
import random
from collections import Counter
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import tempfile
import zipfile


# #############################################################################
# CORE LOGIC FROM UTILITY SCRIPTS
# #############################################################################

# --- From organise_dataset.py ---
def util_get_class_name_from_filename(filename):
    match = re.match(r'(.+?)_\d{3,}', filename)
    return match.group(1) if match else None

def util_create_class_mapping(train_txt_path, log_capture):
    class_mapping = {}
    try:
        with open(train_txt_path, 'r') as f:
            for line in f:
                parts = line.strip().split()
                if len(parts) == 2:
                    filename, class_id_str = parts
                    class_id = int(class_id_str)
                    if class_id not in class_mapping:
                        class_name = util_get_class_name_from_filename(filename)
                        if class_name:
                            class_mapping[class_id] = class_name
    except FileNotFoundError:
        print(f"Error: {train_txt_path} not found.", file=log_capture)
        return None
    return class_mapping

def util_process_dataset(annotations_path, source_dir, dest_dir, class_mapping, log_capture):
    if not os.path.exists(annotations_path):
        print(f"Warning: Annotations file not found: {annotations_path}", file=log_capture)
        return
    if not os.path.exists(source_dir):
        print(f"Warning: Source directory not found: {source_dir}", file=log_capture)
        return
    with open(annotations_path, 'r') as f:
        for line in f:
            parts = line.strip().split()
            if len(parts) == 2:
                filename, class_id_str = parts
                class_id = int(class_id_str)
                if class_id in class_mapping:
                    class_name = class_mapping[class_id]
                    class_dir = os.path.join(dest_dir, class_name)
                    os.makedirs(class_dir, exist_ok=True)
                    source_path = os.path.join(source_dir, filename)
                    dest_path = os.path.join(class_dir, filename)
                    if os.path.exists(source_path):
                        print(f"Moving {source_path} to {dest_path}", file=log_capture)
                        shutil.move(source_path, dest_path)
                    else:
                        print(f"Warning: Source file not found: {source_path}", file=log_capture)
                else:
                    print(f"Warning: Class ID {class_id} for file {filename} not in mapping.", file=log_capture)

# --- From normalise_class_names.py ---
def util_normalise_class_names(target_dir, log_capture):
    print(f"\nStarting normalisation in '{target_dir}'...", file=log_capture)
    try:
        # Make a static list of directories to process, as we're modifying the contents of the target_dir
        subdirectories = [d for d in os.listdir(target_dir) if os.path.isdir(os.path.join(target_dir, d))]
        if not subdirectories:
            print("No subdirectories found to normalise.", file=log_capture)
            return
        for old_name in subdirectories:
            new_name = old_name.lower()
            if old_name == new_name:
                print(f"Skipping '{old_name}' as it is already normalised.", file=log_capture)
                continue
            old_path = os.path.join(target_dir, old_name)
            new_path = os.path.join(target_dir, new_name)
            # On case-sensitive systems, check if a different directory with the new name already exists.
            if os.path.exists(new_path) and not os.path.samefile(old_path, new_path):
                print(f"Warning: Cannot rename '{old_name}' to '{new_name}' because a different directory with that name already exists. Skipping.", file=log_capture)
                continue
            try:
                # Two-stage rename to handle case-insensitivity issues on some filesystems
                temp_name = old_name + "_temp_rename"
                temp_path = os.path.join(target_dir, temp_name)
                if os.path.exists(temp_path):
                    print(f"Warning: Temporary path '{temp_path}' already exists. Skipping rename for '{old_name}'.", file=log_capture)
                    continue
                os.rename(old_path, temp_path)
                os.rename(temp_path, new_path)
                print(f"Renamed '{old_name}' to '{new_name}'.", file=log_capture)
            except OSError as e:
                print(f"Error renaming '{old_name}': {e}", file=log_capture)
        print("\nNormalisation complete.", file=log_capture)
    except Exception as e:
        print(f"An unexpected error occurred: {e}", file=log_capture)

# --- From normalise_image_names.py ---
def util_lowercase_filenames(target_dir, log_capture):
    file_count, renamed_count = 0, 0
    for dirpath, _, filenames in os.walk(target_dir):
        for old_name in filenames:
            file_count += 1
            new_name = old_name.lower()
            if old_name == new_name: continue
            old_path, new_path = os.path.join(dirpath, old_name), os.path.join(dirpath, new_name)
            if os.path.exists(new_path) and not os.path.samefile(old_path, new_path):
                print(f"Warning: Cannot rename '{old_path}' to '{new_path}'. Skipping.", file=log_capture)
                continue
            try:
                temp_name = old_name + "_temp_rename"
                temp_path = os.path.join(dirpath, temp_name)
                if os.path.exists(temp_path):
                    print(f"Warning: Temp path '{temp_path}' exists. Skipping '{old_name}'.", file=log_capture)
                    continue
                os.rename(old_path, temp_path)
                os.rename(temp_path, new_path)
                print(f"Renamed '{os.path.basename(old_path)}' to '{os.path.basename(new_path)}'", file=log_capture)
                renamed_count += 1
            except OSError as e:
                print(f"Error renaming '{old_path}': {e}", file=log_capture)
    print(f"\nProcessed {file_count} files, renamed {renamed_count}.", file=log_capture)

def util_standardise_filenames(target_dir, to_lowercase, log_capture):
    total_renamed = 0
    for dirpath, _, filenames in os.walk(target_dir):
        if not filenames: continue
        class_name = os.path.basename(dirpath).lower() if to_lowercase else os.path.basename(dirpath)
        print(f"\nProcessing directory: {class_name}", file=log_capture)
        rename_map = []
        for i, old_name in enumerate(sorted(filenames)):
            _, extension = os.path.splitext(old_name)
            if to_lowercase: extension = extension.lower()
            new_name = f"{class_name}_{i+1:04d}{extension}"
            rename_map.append({'old_path': os.path.join(dirpath, old_name), 'new_path': os.path.join(dirpath, new_name)})
        for item in rename_map:
            item['temp_path'] = item['old_path'] + '.tmp'
            os.rename(item['old_path'], item['temp_path'])
        for item in rename_map:
            os.rename(item['temp_path'], item['new_path'])
            print(f"Renamed '{os.path.basename(item['old_path'])}' to '{os.path.basename(item['new_path'])}'", file=log_capture)
            total_renamed += 1
    print(f"\nStandardised {total_renamed} files.", file=log_capture)

# --- From autotrain_dataset_splitter.py ---
def util_split_image_dataset(source_dir, output_dir, min_images_per_split, log_capture):
    main_output_path = os.path.join(os.getcwd(), output_dir)
    os.makedirs(main_output_path, exist_ok=True)
    print(f"Created output directory: {main_output_path}", file=log_capture)

    try:
        with tempfile.TemporaryDirectory(prefix="autotrain_split_train_") as train_dir, \
             tempfile.TemporaryDirectory(prefix="autotrain_split_val_") as validation_dir:

            class_dirs = [r for r, d, f in os.walk(source_dir) if not d and f]
            image_extensions = {".jpg", ".jpeg", ".png", ".bmp", ".gif"}
            required_total = min_images_per_split * 2
            class_image_data = {cd: [f for f in os.listdir(cd) if os.path.splitext(f)[1].lower() in image_extensions] for cd in class_dirs}
            valid_class_count = sum(1 for images in class_image_data.values() if len(images) >= required_total)
            if valid_class_count < 2:
                print(f"Error: Dataset splitting requires at least 2 classes with >= {required_total} images each. Found {valid_class_count} valid classes.", file=log_capture)
                return

            manifest = {"included_classes": {}, "skipped_classes": {}}
            processed_class_names = set()
            for class_dir, images in class_image_data.items():
                base_class_name = os.path.basename(class_dir)
                class_name, counter = base_class_name, 1
                while class_name in processed_class_names:
                    class_name = f"{base_class_name}_{counter}"; counter += 1
                processed_class_names.add(class_name)
                if len(images) < required_total:
                    manifest["skipped_classes"][class_name] = {"count": len(images), "reason": f"Not enough images ({len(images)}), required {required_total}."}
                    continue
                random.shuffle(images)
                num_val_ratio = round(len(images) * 0.2)
                num_train_ratio = len(images) - num_val_ratio
                num_validation = num_val_ratio if num_val_ratio >= min_images_per_split and num_train_ratio >= min_images_per_split else min_images_per_split
                validation_images, train_images = images[:num_validation], images[num_validation:]
                manifest["included_classes"][class_name] = {"train": len(train_images), "validation": len(validation_images)}
                for split_dir, split_images in [(train_dir, train_images), (validation_dir, validation_images)]:
                    split_class_dir = os.path.join(split_dir, class_name)
                    os.makedirs(split_class_dir, exist_ok=True)
                    for image in split_images:
                        shutil.copy(os.path.join(class_dir, image), os.path.join(split_class_dir, image))

            manifest_filename = f"{output_dir}-manifest.json"
            manifest_path = os.path.join(main_output_path, manifest_filename)
            with open(manifest_path, "w") as f: json.dump(manifest, f, indent=4)
            print(f"Manifest saved to: {manifest_path}", file=log_capture)

            train_zip_filename = f"{output_dir}-train"
            train_zip_path = os.path.join(main_output_path, train_zip_filename)
            shutil.make_archive(train_zip_path, 'zip', train_dir)
            print(f"Training data zip created: {train_zip_path}.zip", file=log_capture)

            validation_zip_filename = f"{output_dir}-validation"
            validation_zip_path = os.path.join(main_output_path, validation_zip_filename)
            shutil.make_archive(validation_zip_path, 'zip', validation_dir)
            print(f"Validation data zip created: {validation_zip_path}.zip", file=log_capture)

        print(f"\nDataset splitting and packaging complete. Files are in '{main_output_path}'.", file=log_capture)
    except Exception as e:
        print(f"An error occurred during dataset splitting: {e}", file=log_capture)

# --- From directory_manifest.py ---
def util_generate_manifest(directory, save_manifest, manifest_path, log_capture):
    ignored_dirs = {'.git', '__pycache__', '.vscode', '.idea', 'node_modules', 'venv', '.venv'}
    ignored_files = {os.path.basename(manifest_path)} if save_manifest else set()
    ignored_extensions = {'.pyc', '.zip', '.log', '.tmp', '.bak', '.swp'}
    manifest_content = []
    try:
        for root, dirs, files in os.walk(directory, topdown=True):
            dirs[:] = sorted([d for d in dirs if d not in ignored_dirs])
            files.sort()
            for filename in files:
                if filename in ignored_files or os.path.splitext(filename)[1].lower() in ignored_extensions:
                    continue
                relative_path = os.path.relpath(os.path.join(root, filename), directory).replace(os.sep, '/')
                manifest_content.append(f"- {relative_path}\n")
        
        manifest_string = "".join(manifest_content)

        print("--- Manifest Content ---", file=log_capture)
        print(manifest_string, file=log_capture)
        print("------------------------", file=log_capture)

        if save_manifest:
            with open(manifest_path, "w", encoding="utf-8") as f:
                f.write(manifest_string)
            print(f"Manifest file created at: {manifest_path}", file=log_capture)
        else:
            print("Manifest generated but not saved.", file=log_capture)
    except Exception as e:
        print(f"An error occurred: {e}", file=log_capture)

# --- From check_balance.py ---
def util_get_class_from_line(line: str):
    return line.strip().lstrip('- ').split('/')[0] if '/' in line else None

def util_analyse_balance(manifest_path):
    if not os.path.exists(manifest_path):
        raise FileNotFoundError(f"Error: Manifest file not found at '{manifest_path}'")
    with open(manifest_path, 'r', encoding='utf-8') as f:
        lines = f.readlines()
    is_class_count_manifest = any('|---|' in line for line in lines[:5]) or (lines and lines[0].strip() == "# Class Count Manifest")
    class_counts = Counter()
    if is_class_count_manifest:
        for line in lines:
            line = line.strip()
            if line.startswith('|'):
                parts = [p.strip() for p in line.split('|')]
                if len(parts) >= 4:
                    class_name, count_str = parts[1], parts[2]
                    if class_name.lower() not in ('class name', '---') and count_str.isdigit():
                        class_counts[class_name] = int(count_str)
    else:
        class_counts = Counter(c for line in lines if (c := util_get_class_from_line(line)))
    if not class_counts:
        return "No classes found in the manifest file.", None
    counts = list(class_counts.values())
    imbalance_ratio = max(counts) / min(counts)
    summary = (
        f"Dataset Balance Analysis\n"
        f"=========================\n"
        f"Total classes: {len(class_counts)}\n"
        f"Total images: {sum(counts)}\n"
        f"Images per class:\n"
        f"  - Minimum: {min(counts)}\n"
        f"  - Maximum: {max(counts)}\n"
        f"  - Average: {np.mean(counts):.2f}\n"
        f"  - Std Dev: {np.std(counts):.2f}\n"
        f"Imbalance Ratio (Max/Min): {imbalance_ratio:.2f}:1"
    )
    sorted_classes = sorted(class_counts.keys())
    sorted_counts = [class_counts[c] for c in sorted_classes]
    fig, ax = plt.subplots(figsize=(20, 10))
    ax.bar(sorted_classes, sorted_counts)
    ax.set_xlabel('Class'); ax.set_ylabel('Number of Images'); ax.set_title('Image Distribution Across Classes')
    plt.xticks(rotation=90, fontsize='small'); plt.tight_layout()
    return summary, fig

# --- From count_classes.py ---
def util_count_classes(target_dir, save_to_manifest, manifest_path, log_capture):
    if not os.path.isdir(target_dir):
        print(f"Error: Directory not found at '{target_dir}'", file=log_capture)
        return
    try:
        class_dirs = [e for e in os.listdir(target_dir) if os.path.isdir(os.path.join(target_dir, e))]
        if not class_dirs:
            print(f"No class subdirectories found in '{target_dir}'.", file=log_capture)
            return
        class_counts = Counter({name: len([f for f in os.listdir(os.path.join(target_dir, name)) if os.path.isfile(os.path.join(target_dir, name, f))]) for name in class_dirs})
        sorted_counts = sorted(class_counts.items())
        print(f"Found {len(class_dirs)} classes.", file=log_capture)
        print("-" * 20, file=log_capture)
        for class_name, count in sorted_counts: print(f"{class_name}: {count} items", file=log_capture)
        print("-" * 20, file=log_capture)
        if save_to_manifest:
            with open(manifest_path, 'w') as f:
                f.write(f"# Class Count Manifest\n\n**Total classes:** {len(class_dirs)}\n\n| Class Name | Item Count |\n|---|---|\n")
                for class_name, count in sorted_counts: f.write(f"| {class_name} | {count} |\n")
            print(f"Manifest saved to {manifest_path}", file=log_capture)
    except OSError as e:
        print(f"Error accessing directory '{target_dir}': {e}", file=log_capture)

# --- From plot_metrics.py ---
def util_plot_training_metrics(json_path):
    with open(json_path, 'r', encoding='utf-8') as f: data = json.load(f)
    df = pd.DataFrame(data.get('log_history', []))
    if df.empty: raise ValueError("No 'log_history' found.")
    train_df = df[df['loss'].notna()].copy()
    eval_df = df[df['eval_loss'].notna()].copy()
    figures = {}
    # Plot Loss
    fig_loss, ax = plt.subplots(figsize=(10, 6)); ax.set_title('Training vs. Evaluation Loss')
    if 'loss' in train_df: ax.plot(train_df['step'], train_df['loss'], label='Training Loss', marker='o')
    if 'eval_loss' in eval_df: ax.plot(eval_df['step'], eval_df['eval_loss'], label='Evaluation Loss', marker='x')
    ax.set_xlabel('Step'); ax.set_ylabel('Loss')
    ax.legend(); ax.grid(True); figures['Loss'] = fig_loss
    # Plot Accuracy
    fig_acc, ax = plt.subplots(figsize=(10, 6)); ax.set_title('Evaluation Accuracy')
    if 'eval_accuracy' in eval_df: ax.plot(eval_df['step'], eval_df['eval_accuracy'], label='Evaluation Accuracy', marker='o', color='g')
    ax.set_xlabel('Step'); ax.set_ylabel('Accuracy')
    ax.legend(); ax.grid(True); figures['Accuracy'] = fig_acc
    # Plot Learning Rate
    fig_lr, ax = plt.subplots(figsize=(10, 6)); ax.set_title('Learning Rate Schedule')
    if 'learning_rate' in train_df: ax.plot(train_df['step'], train_df['learning_rate'], label='Learning Rate', marker='o', color='r')
    ax.set_xlabel('Step'); ax.set_ylabel('Learning Rate')
    ax.legend(); ax.grid(True); figures['Learning Rate'] = fig_lr
    # Plot Grad Norm
    fig_gn, ax = plt.subplots(figsize=(10, 6)); ax.set_title('Gradient Norm')
    if 'grad_norm' in train_df: ax.plot(train_df['step'], train_df['grad_norm'], label='Grad Norm', marker='o', color='purple')
    ax.set_xlabel('Step'); ax.set_ylabel('Gradient Norm')
    ax.legend(); ax.grid(True); figures['Gradient Norm'] = fig_gn
    # Plot F1
    fig_f1, ax = plt.subplots(figsize=(10, 6)); ax.set_title('Evaluation F1 Scores')
    if 'eval_f1_macro' in eval_df: ax.plot(eval_df['step'], eval_df['eval_f1_macro'], label='F1 Macro', marker='o')
    if 'eval_f1_micro' in eval_df: ax.plot(eval_df['step'], eval_df['eval_f1_micro'], label='F1 Micro', marker='x')
    if 'eval_f1_weighted' in eval_df: ax.plot(eval_df['step'], eval_df['eval_f1_weighted'], label='F1 Weighted', marker='s')
    ax.set_xlabel('Step'); ax.set_ylabel('F1 Score')
    ax.legend(); ax.grid(True); figures['F1 Scores'] = fig_f1
    # Plot Precision
    fig_prec, ax = plt.subplots(figsize=(10, 6)); ax.set_title('Evaluation Precision Scores')
    if 'eval_precision_macro' in eval_df: ax.plot(eval_df['step'], eval_df['eval_precision_macro'], label='Precision Macro', marker='o')
    if 'eval_precision_micro' in eval_df: ax.plot(eval_df['step'], eval_df['eval_precision_micro'], label='Precision Micro', marker='x')
    if 'eval_precision_weighted' in eval_df: ax.plot(eval_df['step'], eval_df['eval_precision_weighted'], label='Precision Weighted', marker='s')
    ax.set_xlabel('Step'); ax.set_ylabel('Precision')
    ax.legend(); ax.grid(True); figures['Precision'] = fig_prec
    # Plot Recall
    fig_recall, ax = plt.subplots(figsize=(10, 6)); ax.set_title('Evaluation Recall Scores')
    if 'eval_recall_macro' in eval_df: ax.plot(eval_df['step'], eval_df['eval_recall_macro'], label='Recall Macro', marker='o')
    if 'eval_recall_micro' in eval_df: ax.plot(eval_df['step'], eval_df['eval_recall_micro'], label='Recall Micro', marker='x')
    if 'eval_recall_weighted' in eval_df: ax.plot(eval_df['step'], eval_df['eval_recall_weighted'], label='Recall Weighted', marker='s')
    ax.set_xlabel('Step'); ax.set_ylabel('Recall')
    ax.legend(); ax.grid(True); figures['Recall'] = fig_recall
    # Plot Epoch
    fig_epoch, ax = plt.subplots(figsize=(10, 6)); ax.set_title('Epoch Progression')
    if 'epoch' in df:
        epoch_df = df[['step', 'epoch']].dropna().drop_duplicates('step').sort_values('step')
        ax.plot(epoch_df['step'], epoch_df['epoch'], label='Epoch', marker='.')
    ax.set_xlabel('Step'); ax.set_ylabel('Epoch')
    ax.legend(); ax.grid(True); figures['Epoch'] = fig_epoch
    # Plot Eval Runtime
    fig_runtime, ax = plt.subplots(figsize=(10, 6)); ax.set_title('Evaluation Runtime')
    if 'eval_runtime' in eval_df: ax.plot(eval_df['step'], eval_df['eval_runtime'], label='Eval Runtime', marker='o')
    ax.set_xlabel('Step'); ax.set_ylabel('Runtime (s)')
    ax.legend(); ax.grid(True); figures['Eval Runtime'] = fig_runtime
    # Plot Eval Samples Per Second
    fig_sps, ax = plt.subplots(figsize=(10, 6)); ax.set_title('Evaluation Samples Per Second')
    if 'eval_samples_per_second' in eval_df: ax.plot(eval_df['step'], eval_df['eval_samples_per_second'], label='Eval Samples/sec', marker='o')
    ax.set_xlabel('Step'); ax.set_ylabel('Samples / Second')
    ax.legend(); ax.grid(True); figures['Eval Samples/sec'] = fig_sps
    # Plot Eval Steps Per Second
    fig_steps_ps, ax = plt.subplots(figsize=(10, 6)); ax.set_title('Evaluation Steps Per Second')
    if 'eval_steps_per_second' in eval_df: ax.plot(eval_df['step'], eval_df['eval_steps_per_second'], label='Eval Steps/sec', marker='o')
    ax.set_xlabel('Step'); ax.set_ylabel('Steps / Second')
    ax.legend(); ax.grid(True); figures['Eval Steps/sec'] = fig_steps_ps
    return figures