Spaces:
Build error
Build error
File size: 14,528 Bytes
4fa8bcb |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 |
# gradio_full_system.py
import gradio as gr
import torch
import pandas as pd
import random
import matplotlib.pyplot as plt
from model_training.model_torch import EncoderDecoder
from model_training.train import train_model
from model_training.generate_training_data import generate_training_data
from model_training.train import PhysicsTrajectoryDataset
from video_sequencer.generate_frames_and_video import generate_frames_and_video
import os
import torch.nn as nn
import numpy as np
from config import normalize_input, denormalize_input, get_input_fields, get_physics_types, get_param_ranges
from utils.path_utils import resolve_path
# --- Available Physics Types ---
physics_types = get_physics_types()
# --- Inspect Training Trajectories ---
def inspect_training_trajectories(physics_type, frame_size=64):
dataset_path = resolve_path(f"{physics_type}_data.pkl")
dataset = PhysicsTrajectoryDataset(dataset_path, physics_type)
num_samples = 3
fig, axes = plt.subplots(num_samples, 3, figsize=(10, 3 * num_samples), constrained_layout=True)
for i in range(num_samples):
row = dataset.df.iloc[i]
input_features, trajectory = dataset[i] # trajectory: [T, 2]
T = trajectory.shape[0]
angle = row.get("angle", None)
for j, t in enumerate([0, T // 2, T - 1]):
ax = axes[i, j] if num_samples > 1 else axes[j]
frame = torch.zeros((frame_size, frame_size))
# Get (x, y) and denormalize back to pixel space
x, y = trajectory[t].numpy()
px = int(x * (frame_size - 1))
py = int(y * (frame_size - 1))
# Plot the point
ax.imshow(frame, cmap="gray")
ax.plot(px, py, "ro", markersize=5)
# Initialize title
title = f"t={t}"
# Dynamically append available metadata fields
for key in ['mass', 'angle', 'friction', 'initial_velocity', 'acceleration', 'gravity']:
if key in row:
val = row[key]
if key == 'angle':
title += f"\n{key.capitalize()}: {val:.1f}Β°"
else:
title += f"\n{key.capitalize()}: {val:.2f}"
if angle is not None:
title += f"\nAngle: {angle:.1f}Β°"
ax.set_title(title, pad=10)
ax.axis("off")
return fig
# --- Predict Normalized Coordinates ---
def predict_trajectory(physics_type, *inputs, debug=False):
model_file = f"{physics_type}_model.pth"
model_path = resolve_path(model_file)
if not os.path.exists(model_path):
return None
sample_data = pd.read_pickle(resolve_path(f"{physics_type}_data.pkl"))
input_dim = len(sample_data.columns) - 1
output_seq_len = len(sample_data.iloc[0]['trajectory']) # [T, 2]
model = EncoderDecoder(
input_dim=input_dim,
output_seq_len=output_seq_len,
output_shape=None # Use coordinate mode
)
checkpoint = torch.load(model_path)
model.load_state_dict(checkpoint['model_state'])
model.eval()
inputs_tensor = torch.tensor([inputs], dtype=torch.float32)
with torch.no_grad():
prediction = model(inputs_tensor)
prediction = prediction.cpu().numpy()[0] # [T, 2]
if debug:
print("π Debug output β predicted (x, y) per frame:")
for t, (x, y) in enumerate(prediction):
print(f"t={t}: ({x:.3f}, {y:.3f})")
return prediction # [T, 2]
# --- Plotting for Normalized (x, y) ---
def plot_coordinates_over_time(coords, title="Predicted Dot Trajectory", frame_size=64):
T = len(coords)
fig, axes = plt.subplots(1, 3, figsize=(12, 4))
indices = [0, T // 2, T - 1]
for ax, idx in zip(axes, indices):
x, y = coords[idx]
px = int(x * (frame_size - 1))
py = int(y * (frame_size - 1))
frame = np.zeros((frame_size, frame_size))
ax.imshow(frame, cmap='gray')
ax.plot(px, py, 'ro')
ax.set_title(f"t={idx} | ({px}, {py})")
ax.axis('off')
fig.suptitle(title)
return fig
def plot_multiple_predictions(predictions, inputs_list, frame_size=64, denorm=None):
N = len(predictions)
fig, axes = plt.subplots(N, 3, figsize=(12, 3 * N), constrained_layout=True)
for row in range(N):
indices = [0, len(predictions[row]) // 2, len(predictions[row]) - 1]
inp = inputs_list[row]
# Denormalize if provided
labels = denorm(inp) if denorm else inp
for col, t in enumerate(indices):
x, y = predictions[row][t]
px = int(x * (frame_size - 1))
py = int(y * (frame_size - 1))
px = round(px, 2)
py = round(py, 2)
frame = np.zeros((frame_size, frame_size))
ax = axes[row, col] if N > 1 else axes[col]
ax.imshow(frame, cmap='gray')
ax.plot(px, py, 'ro')
ax.set_title(f"t={t} | inputs={labels}\nDot: ({px}, {py})")
ax.axis("off")
return fig
def plot_trajectory(physics_type, *inputs, debug=False):
pred = predict_trajectory(physics_type, *inputs, debug=debug)
if pred is None:
return None
fig = plot_coordinates_over_time(pred, title=f"{physics_type.title()} Prediction")
return fig, pred
# --- Video Generation ---
def predict_plot_video(physics_type, *inputs, debug=False):
norm_input = normalize_input(physics_type, *inputs)
fig, pred = plot_trajectory(physics_type, *norm_input, debug=debug)
if pred is None:
return None, None
video_mp4_path = generate_frames_and_video(pred)
return fig, video_mp4_path
def test_input_sensitivity(physics_type):
model_path = resolve_path(f"{physics_type}_model.pth")
if not os.path.exists(model_path):
return None
# Load sample data to get dimensions
sample_data = pd.read_pickle(f"data/{physics_type}_data.pkl")
input_dim = len(sample_data.columns) - 1
output_seq_len = len(sample_data.iloc[0]['trajectory']) # [T, 2]
# Load trained model
model = EncoderDecoder(
input_dim=input_dim,
output_seq_len=output_seq_len,
output_shape=None
)
checkpoint = torch.load(model_path)
model.load_state_dict(checkpoint['model_state'])
model.eval()
# Get parameter ranges and field names
param_ranges = get_param_ranges(physics_type)
param_specs = get_input_fields(physics_type)
# Dynamically generate test inputs (midpoints or meaningful values)
test_inputs = []
for variation in range(4):
base = []
for field in param_specs:
min_val, max_val = param_ranges[field]
mid = (min_val + max_val) / 2
val = mid + ((variation - 1.5) * (max_val - min_val) / 4)
val = max(min_val, min(max_val, val)) # Clamp
base.append(val)
test_inputs.append(normalize_input(physics_type, *base))
preds = []
for inp in test_inputs:
inp_tensor = torch.tensor([inp], dtype=torch.float32)
with torch.no_grad():
output = model(inp_tensor).cpu().numpy()[0] # [T, 2]
preds.append(output)
# Create a closure that captures physics_type
def denorm_fn(inputs):
return denormalize_input(physics_type, inputs)
fig = plot_multiple_predictions(preds, test_inputs, denorm=denorm_fn)
return fig
def load_uploaded_dataset(uploaded_file, physics_type):
tmp_path = os.path.join("/tmp", f"{physics_type}_data.pkl")
with open(uploaded_file.name, "rb") as src, open(tmp_path, "wb") as dst:
dst.write(src.read())
return f"β
Uploaded and saved dataset to /tmp for '{physics_type}'"
def load_uploaded_model(uploaded_file, physics_type):
tmp_path = os.path.join("/tmp", "", f"{physics_type}_model.pth")
os.makedirs(os.path.dirname(tmp_path), exist_ok=True)
with open(uploaded_file.name, "rb") as src, open(tmp_path, "wb") as dst:
dst.write(src.read())
return f"β
Uploaded and saved model to /tmp for '{physics_type}'"
# --- Gradio Interface ---
with gr.Blocks() as demo:
gr.Markdown("# π§ Full Physics ML System")
gr.Markdown("## π§ͺ Data β Train β Predict")
gr.Markdown("### π¨βπ» Developed by [Miles Exner](https://www.linkedin.com/in/milesexner/)")
# π Global dropdown visible to user
physics_dropdown = gr.Dropdown(
choices=physics_types,
label="Physics Type",
value=physics_types[0]
)
with gr.Tab("Data Generation"):
with gr.Row():
num_samples = gr.Slider(100, 5000, value=1000, label="Number of Samples", step=100)
time_steps = gr.Slider(5, 100, value=50, label="Time Steps", step=5)
gen_output = gr.Textbox(label="Output Log")
generate_btn = gr.Button("Generate Data")
generate_btn.click(
fn=generate_training_data,
inputs=[physics_dropdown, num_samples, time_steps],
outputs=gen_output
)
gr.Markdown("#### π€ Upload Existing Dataset (.pkl)")
upload_data = gr.File(file_types=[".pkl"], label="Upload .pkl")
upload_data.upload(
fn=load_uploaded_dataset,
inputs=[upload_data, physics_dropdown],
outputs=gen_output
)
gr.Markdown("#### π₯ Download Generated Dataset")
download_data_btn = gr.Button("Download Dataset")
download_data_file = gr.File(label="Download Link")
def return_dataset_path(physics_type):
return resolve_path(f"{physics_type}_data.pkl", write_mode=True)
download_data_btn.click(
fn=return_dataset_path,
inputs=[physics_dropdown],
outputs=download_data_file
)
with gr.Tab("Data Inspection"):
gr.Markdown("Visualize 3 samples from the training dataset to debug dot position.")
inspect_btn = gr.Button("Show Sample Trajectories")
output_fig = gr.Plot()
inspect_btn.click(
fn=inspect_training_trajectories,
inputs=[physics_dropdown],
outputs=output_fig
)
with gr.Tab("Training"):
with gr.Row():
epochs = gr.Slider(5, 100, value=20, label="Epochs", step=1)
early_stopping_checkbox = gr.Checkbox(label="Enable Early Stopping", value=False)
patience_slider = gr.Slider(1, 20, value=5, label="Patience Steps", step=1)
train_output = gr.Textbox(label="Training Log")
loss_plot = gr.Plot(label="Training Loss Curve")
train_btn = gr.Button("Train Model")
def run_training(physics_type, epochs, early_stopping, patience):
msg, losses = train_model(
physics_type=physics_type,
epochs=epochs,
early_stopping=early_stopping,
patience=patience
)
fig, ax = plt.subplots()
ax.plot(losses)
ax.set_title("Training Loss Curve")
ax.set_xlabel("Epoch")
ax.set_ylabel("Loss")
return msg, fig
train_btn.click(
fn=run_training,
inputs=[physics_dropdown, epochs, early_stopping_checkbox, patience_slider],
outputs=[train_output, loss_plot]
)
gr.Markdown("#### π€ Upload Trained Model (.pth)")
upload_model = gr.File(file_types=[".pth"], label="Upload .pth")
upload_model.upload(
fn=load_uploaded_model,
inputs=[upload_model, physics_dropdown],
outputs=train_output
)
gr.Markdown("#### π₯ Download Trained Model")
download_model_btn = gr.Button("Download Model")
download_model_file = gr.File(label="Download Link")
def return_model_path(physics_type):
return resolve_path(f"{physics_type}_model.pth", write_mode=True)
download_model_btn.click(
fn=return_model_path,
inputs=[physics_dropdown],
outputs=download_model_file
)
with gr.Tab("Input Sensitivity Test"):
gr.Markdown("Compare how different inputs affect predicted trajectories.")
sensitivity_btn = gr.Button("Run Sensitivity Test")
output_plot = gr.Plot()
sensitivity_btn.click(
fn=test_input_sensitivity,
inputs=[physics_dropdown],
outputs=output_plot
)
with gr.Tab("Prediction"):
with gr.Row():
slider_outputs = [
gr.Slider(visible=False),
gr.Slider(visible=False),
gr.Slider(visible=False)
]
debug_checkbox = gr.Checkbox(label="Debug", value=False)
predict_btn = gr.Button("Predict Trajectory")
pred_plot = gr.Plot(label="Trajectory Prediction")
pred_video = gr.Video(label="Generated Video")
def refresh_inputs(physics_type):
sliders = get_input_fields(physics_type) # Returns configured sliders
updates = []
param_ranges = get_param_ranges(physics_type)
for template, target in zip(sliders, slider_outputs):
min_val, max_val = param_ranges[template]
default_val = (min_val + max_val) / 2
updates.append(gr.update(
visible=True,
label=template.replace('_', ' ').title(),
minimum=min_val,
maximum=max_val,
value=default_val
))
# Hide any unused sliders
for _ in range(len(sliders), len(slider_outputs)):
updates.append(gr.update(visible=False))
return updates
physics_dropdown.change(
fn=refresh_inputs,
inputs=physics_dropdown,
outputs=slider_outputs
)
demo.load(fn=refresh_inputs, inputs=physics_dropdown, outputs=slider_outputs)
def predict_switch(physics_type, *args):
*slider_vals, debug = args
return predict_plot_video(physics_type, *slider_vals, debug=debug)
predict_btn.click(
fn=predict_switch,
inputs=[physics_dropdown] + slider_outputs + [debug_checkbox],
outputs=[pred_plot, pred_video]
)
if __name__ == "__main__":
demo.launch()
|