File size: 8,003 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 |
import matplotlib.pyplot as plt
import matplotlib.colors as mcolors
import numpy as np
import json
import os
import os.path as osp
# LOAD FINAL RESULTS:
datasets = ["x_div_y", "x_minus_y", "x_plus_y", "permutation"]
folders = os.listdir("./")
final_results = {}
results_info = {}
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)
results_dict = np.load(
osp.join(folder, "all_results.npy"), allow_pickle=True
).item()
print(results_dict.keys())
run_info = {}
for dataset in datasets:
run_info[dataset] = {}
val_losses = []
train_losses = []
val_accs = []
train_accs = []
for k in results_dict.keys():
if dataset in k and "val_info" in k:
run_info[dataset]["step"] = [
info["step"] for info in results_dict[k]
]
val_losses.append([info["val_loss"] for info in results_dict[k]])
val_accs.append([info["val_accuracy"] for info in results_dict[k]])
if dataset in k and "train_info" in k:
train_losses.append(
[info["train_loss"] for info in results_dict[k]]
)
train_accs.append(
[info["train_accuracy"] for info in results_dict[k]]
)
mean_val_losses = np.mean(val_losses, axis=0)
mean_train_losses = np.mean(train_losses, axis=0)
mean_val_accs = np.mean(val_accs, axis=0)
mean_train_accs = np.mean(train_accs, axis=0)
if len(val_losses) > 0:
sterr_val_losses = np.std(val_losses, axis=0) / np.sqrt(
len(val_losses)
)
stderr_train_losses = np.std(train_losses, axis=0) / np.sqrt(
len(train_losses)
)
sterr_val_accs = np.std(val_accs, axis=0) / np.sqrt(len(val_accs))
stderr_train_accs = np.std(train_accs, axis=0) / np.sqrt(
len(train_accs)
)
else:
sterr_val_losses = np.zeros_like(mean_val_losses)
stderr_train_losses = np.zeros_like(mean_train_losses)
sterr_val_accs = np.zeros_like(mean_val_accs)
stderr_train_accs = np.zeros_like(mean_train_accs)
run_info[dataset]["val_loss"] = mean_val_losses
run_info[dataset]["train_loss"] = mean_train_losses
run_info[dataset]["val_loss_sterr"] = sterr_val_losses
run_info[dataset]["train_loss_sterr"] = stderr_train_losses
run_info[dataset]["val_acc"] = mean_val_accs
run_info[dataset]["train_acc"] = mean_train_accs
run_info[dataset]["val_acc_sterr"] = sterr_val_accs
run_info[dataset]["train_acc_sterr"] = stderr_train_accs
results_info[folder] = run_info
# CREATE LEGEND -- ADD RUNS HERE THAT WILL BE PLOTTED
labels = {
"run_0": "Baseline",
"run_1": "Xavier (Glorot)",
"run_2": "He",
"run_3": "Orthogonal",
"run_4": "Kaiming Normal",
"run_5": "Uniform",
}
# Create a programmatic color palette
def generate_color_palette(n):
cmap = plt.get_cmap("tab20")
return [mcolors.rgb2hex(cmap(i)) for i in np.linspace(0, 1, n)]
# Get the list of runs and generate the color palette
runs = list(labels.keys())
colors = generate_color_palette(len(runs))
# Plot 1: Line plot of training loss for each dataset across the runs with labels
for dataset in datasets:
plt.figure(figsize=(10, 6))
for i, run in enumerate(runs):
iters = results_info[run][dataset]["step"]
mean = results_info[run][dataset]["train_loss"]
sterr = results_info[run][dataset]["train_loss_sterr"]
plt.plot(iters, mean, label=labels[run], color=colors[i])
plt.fill_between(iters, mean - sterr, mean + sterr, color=colors[i], alpha=0.2)
plt.title(f"Training Loss Across Runs for {dataset} Dataset")
plt.xlabel("Update Steps")
plt.ylabel("Training Loss")
plt.legend()
plt.grid(True, which="both", ls="-", alpha=0.2)
plt.tight_layout()
plt.savefig(f"train_loss_{dataset}.png")
plt.close()
# Plot 2: Line plot of validation loss for each dataset across the runs with labels
for dataset in datasets:
plt.figure(figsize=(10, 6))
for i, run in enumerate(runs):
iters = results_info[run][dataset]["step"]
mean = results_info[run][dataset]["val_loss"]
sterr = results_info[run][dataset]["val_loss_sterr"]
plt.plot(iters, mean, label=labels[run], color=colors[i])
plt.fill_between(iters, mean - sterr, mean + sterr, color=colors[i], alpha=0.2)
plt.title(f"Validation Loss Across Runs for {dataset} Dataset")
plt.xlabel("Update Steps")
plt.ylabel("Validation Loss")
plt.legend()
plt.grid(True, which="both", ls="-", alpha=0.2)
plt.tight_layout()
plt.savefig(f"val_loss_{dataset}.png")
plt.close()
# Plot 3: Line plot of training acc for each dataset across the runs with labels
for dataset in datasets:
plt.figure(figsize=(10, 6))
for i, run in enumerate(runs):
iters = results_info[run][dataset]["step"]
mean = results_info[run][dataset]["train_acc"]
sterr = results_info[run][dataset]["train_acc_sterr"]
plt.plot(iters, mean, label=labels[run], color=colors[i])
plt.fill_between(iters, mean - sterr, mean + sterr, color=colors[i], alpha=0.2)
plt.title(f"Training Accuracy Across Runs for {dataset} Dataset")
plt.xlabel("Update Steps")
plt.ylabel("Training Acc")
plt.legend()
plt.grid(True, which="both", ls="-", alpha=0.2)
plt.tight_layout()
plt.savefig(f"train_acc_{dataset}.png")
plt.close()
# Plot 2: Line plot of validation acc for each dataset across the runs with labels
for dataset in datasets:
plt.figure(figsize=(10, 6))
for i, run in enumerate(runs):
iters = results_info[run][dataset]["step"]
mean = results_info[run][dataset]["val_acc"]
sterr = results_info[run][dataset]["val_acc_sterr"]
plt.plot(iters, mean, label=labels[run], color=colors[i])
plt.fill_between(iters, mean - sterr, mean + sterr, color=colors[i], alpha=0.2)
plt.title(f"Validation Loss Across Runs for {dataset} Dataset")
plt.xlabel("Update Steps")
plt.ylabel("Validation Acc")
plt.legend()
plt.grid(True, which="both", ls="-", alpha=0.2)
plt.tight_layout()
plt.savefig(f"val_acc_{dataset}.png")
plt.close()
# Plot 5: Summary plot comparing all initialization methods across datasets
def plot_summary(final_results, labels, datasets):
metrics = ['final_train_acc_mean', 'final_val_acc_mean', 'step_val_acc_99_mean']
fig, axs = plt.subplots(len(metrics), 1, figsize=(12, 5*len(metrics)), sharex=True)
x = np.arange(len(datasets))
width = 0.15
n_runs = len(labels)
for i, metric in enumerate(metrics):
for j, (run, label) in enumerate(labels.items()):
values = [final_results[run][dataset]['means'][metric] for dataset in datasets]
axs[i].bar(x + (j - n_runs/2 + 0.5) * width, values, width, label=label)
axs[i].set_ylabel(metric.replace('_', ' ').title())
axs[i].set_xticks(x)
axs[i].set_xticklabels(datasets)
axs[i].legend(loc='upper left', bbox_to_anchor=(1, 1))
axs[i].grid(True, which="both", ls="-", alpha=0.2)
plt.tight_layout()
plt.savefig("summary_plot.png", bbox_inches='tight')
plt.close()
# Call the summary plot function
plot_summary(final_results, labels, datasets)
|