Run it with
python -m src.random_ball_falling --pause
The we want restore configuration file in XML format and use them for training afterwards. The configuration file includes bodies and contacts
Note:
python -m src.gen_data.generate_data -s 30 -p 'path' -n 10
to generate the training data
<body index="86" type="free">
<mass value="3.14159274101"/>
<position x="7.79289388657" y="2.62924313545"/>
<velocity x="2.7878344059" y="-1.45545887947"/>
<orientation theta="-0.115291565657"/>
<inertia value="1.57079637051"/>
<spin omega="-2.33787894249"/>
<shape value="circle"/>
</body>
...
<contact index="1" master="2" master_shape="b2CircleShape(childCount=1,
pos=b2Vec2(0,0),
radius=1.2000000476837158,
type=0,
)" slave="97" slave_shape="b2CircleShape(childCount=1,
pos=b2Vec2(0,0),
radius=1.2000000476837158,
type=0,
)">
<position x="0.21963849663734436" y="13.875240325927734"/>
<normal normal="b2Vec2(-1,2.9819e-05)"/>
<impulse n="0.005236322991549969" t="-0.002184529323130846"/>
</contact>
Then you can transform the xml data to grid ones which will return to np.array
python -m src.gen_data.load_xml_save_grid
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_1 (Conv2D) (None, 41, 41, 64) 2944
_________________________________________________________________
batch_normalization_1 (Batch (None, 41, 41, 64) 256
_________________________________________________________________
activation_1 (Activation) (None, 41, 41, 64) 0
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 21, 21, 64) 0
_________________________________________________________________
conv2d_2 (Conv2D) (None, 21, 21, 128) 73856
_________________________________________________________________
batch_normalization_2 (Batch (None, 21, 21, 128) 512
_________________________________________________________________
activation_2 (Activation) (None, 21, 21, 128) 0
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 11, 11, 128) 0
_________________________________________________________________
conv2d_3 (Conv2D) (None, 11, 11, 256) 295168
_________________________________________________________________
batch_normalization_3 (Batch (None, 11, 11, 256) 1024
_________________________________________________________________
activation_3 (Activation) (None, 11, 11, 256) 0
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 6, 6, 256) 0
_________________________________________________________________
conv2d_4 (Conv2D) (None, 6, 6, 512) 1180160
_________________________________________________________________
batch_normalization_4 (Batch (None, 6, 6, 512) 2048
_________________________________________________________________
activation_4 (Activation) (None, 6, 6, 512) 0
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 3, 3, 512) 0
_________________________________________________________________
dropout_1 (Dropout) (None, 3, 3, 512) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 4608) 0
_________________________________________________________________
dense_1 (Dense) (None, 4000) 18436000
_________________________________________________________________
activation_5 (Activation) (None, 4000) 0
_________________________________________________________________
dropout_2 (Dropout) (None, 4000) 0
_________________________________________________________________
dense_2 (Dense) (None, 3362) 13451362
_________________________________________________________________
activation_6 (Activation) (None, 3362) 0
=================================================================
Total params: 33,443,330
Trainable params: 33,441,410
Non-trainable params: 1,920
_________________________________________________________________
Training Config
batch_size=1000,
metrics=[losses.mean_absolute_error],
optimizer=optimizers.SGD(
lr=0.01, decay=1e-6, momentum=0.9, nesterov=True,
),
loss_func=losses.mean_squared_error,
Train on 18208 samples, validate on 4553 samples. Training Size
Input Size: 41 * 41 *5
Output Size: 3362 * 1
Start training by
python cnn_training.py -p src/gen_data/data/grid -n 30
After training finished, you can check the training logs by
tensorboard --logdir ./log/${date_folder}