File size: 4,935 Bytes
f71c233
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import matplotlib.pyplot as plt
import matplotlib.colors as mcolors
import numpy as np
import json
import os
import os.path as osp
import pickle

# LOAD FINAL RESULTS:
datasets = ["circle", "dino", "line", "moons"]
folders = os.listdir("./")
final_results = {}
train_info = {}


def smooth(x, window_len=10, window='hanning'):
    s = np.r_[x[window_len - 1:0:-1], x, x[-2:-window_len - 1:-1]]
    if window == 'flat':  # moving average
        w = np.ones(window_len, 'd')
    else:
        w = getattr(np, window)(window_len)
    y = np.convolve(w / w.sum(), s, mode='valid')
    return y


for folder in folders:
    if folder.startswith("run") and osp.isdir(folder):
        with open(osp.join(folder, "final_info.json"), "r") as f:
            final_results[folder] = json.load(f)
        all_results = pickle.load(open(osp.join(folder, "all_results.pkl"), "rb"))
        train_info[folder] = all_results

# Create a programmatic color palette
def generate_color_palette(n):
    cmap = plt.get_cmap('tab20')  # You can change 'tab20' to other colormaps like 'Set1', 'Set2', 'Set3', etc.
    return [mcolors.rgb2hex(cmap(i)) for i in np.linspace(0, 1, n)]

# CREATE LEGEND
labels = {
    "run_0": "Baseline",
    "run_1": "10x10 Grid",
    "run_2": "20x20 Grid",
    "run_3": "Multi-scale Grid",
    "run_4": "Multi-scale + L1 Reg",
    "run_5": "Adjusted L1 Reg"
}

# Only plot the runs that are both in the labels dictionary and in the final_results
runs = [run for run in final_results.keys() if run in labels]
colors = generate_color_palette(len(runs))


# CREATE PLOTS


# Get the list of runs and generate the color palette
runs = list(final_results.keys())
colors = generate_color_palette(len(runs))

# Plot 1: Line plot of training loss for each dataset across the runs with labels
fig, axs = plt.subplots(2, 2, figsize=(14, 8), sharex=True)

for j, dataset in enumerate(datasets):
    row = j // 2
    col = j % 2
    for i, run in enumerate(runs):
        mean = train_info[run][dataset]["train_losses"]
        mean = smooth(mean, window_len=25)
        axs[row, col].plot(mean, label=labels[run], color=colors[i])
        axs[row, col].set_title(dataset)
        axs[row, col].legend()
        axs[row, col].set_xlabel("Training Step")
        axs[row, col].set_ylabel("Loss")

plt.tight_layout()
plt.savefig("train_loss.png")
plt.show()

# Plot 2: Visualize generated samples
# If there is more than 1 run, these are added as extra rows.
num_runs = len(runs)
fig, axs = plt.subplots(num_runs, 4, figsize=(14, 3 * num_runs))

for i, run in enumerate(runs):
    for j, dataset in enumerate(datasets):
        images = train_info[run][dataset]["images"]
        if num_runs == 1:
            axs[j].scatter(images[:, 0], images[:, 1], alpha=0.2, color=colors[i])
            axs[j].set_title(dataset)
        else:
            axs[i, j].scatter(images[:, 0], images[:, 1], alpha=0.2, color=colors[i])
            axs[i, j].set_title(dataset)
    if num_runs == 1:
        axs[0].set_ylabel(labels[run])
    else:
        axs[i, 0].set_ylabel(labels[run])

plt.tight_layout()
plt.savefig("generated_images.png")
plt.show()

# Plot 3: Bar plot of evaluation metrics
metrics = ['eval_loss', 'kl_divergence', 'training_time', 'inference_time']
fig, axs = plt.subplots(2, 2, figsize=(16, 12))
fig.suptitle("Evaluation Metrics Across Runs", fontsize=16)

for i, metric in enumerate(metrics):
    row = i // 2
    col = i % 2
    data = [final_results[run][dataset][metric] for run in runs for dataset in datasets]
    x = np.arange(len(datasets) * len(runs))
    axs[row, col].bar(x, data, color=colors)
    axs[row, col].set_title(metric.replace('_', ' ').title())
    axs[row, col].set_xticks(x + 0.5 * (len(runs) - 1))
    axs[row, col].set_xticklabels(datasets * len(runs), rotation=45)
    axs[row, col].legend(labels.values(), loc='upper left', bbox_to_anchor=(1, 1))

plt.tight_layout()
plt.savefig("evaluation_metrics.png")
plt.show()

# Plot 4: Grid variance comparison (for runs 3 and 4)
if 'run_3' in runs and 'run_4' in runs:
    fig, axs = plt.subplots(1, 2, figsize=(14, 6))
    fig.suptitle("Grid Variance Comparison", fontsize=16)

    for i, grid_type in enumerate(['coarse_grid_variance', 'fine_grid_variance']):
        data_run3 = [final_results['run_3'][dataset][grid_type] for dataset in datasets]
        data_run4 = [final_results['run_4'][dataset][grid_type] for dataset in datasets]
        
        x = np.arange(len(datasets))
        width = 0.35
        
        axs[i].bar(x - width/2, data_run3, width, label='Multi-scale Grid', color=colors[3])
        axs[i].bar(x + width/2, data_run4, width, label='Multi-scale + L1 Reg', color=colors[4])
        
        axs[i].set_title(grid_type.replace('_', ' ').title())
        axs[i].set_xticks(x)
        axs[i].set_xticklabels(datasets)
        axs[i].legend()

    plt.tight_layout()
    plt.savefig("grid_variance_comparison.png")
    plt.show()