Apache MXNet Cheat Sheet Sunil Mallya: smallya@amazon.com Multilayer Perceptron (MLP) Lenet (CNN example) data = mx.symbol.Variable(’data’) conv1 = mx.sym.Convolution(data=data, kernel=(5,5), num_filter=20) tanh1 = mx.sym.Activation(data=conv1, act_type="tanh") pool1 = mx.sym.Pooling(data=tanh1, pool_type="max", kernel =(2,2), stride=(2,2)) conv2 = mx.sym.Convolution(data=pool1, kernel=(5,5), num_filter=50) tanh2 = mx.sym.Activation(data=conv2, act_type="tanh") pool2 = mx.sym.Pooling(data=tanh2, pool_type="max", kernel =(2,2), stride=(2,2)) flatten = mx.sym.Flatten(data=pool2) fc1 = mx.symbol.FullyConnected(data=flatten, num_hidden=500) tanh3 = mx.sym.Activation(data=fc1, act_type="tanh") fc2 = mx.sym.FullyConnected(data=tanh3, num_hidden=10) lenet = mx.sym.SoftmaxOutput(data=fc2, name=’softmax’) mx.viz.plot_network(symbol=lenet, shape=shape) Image Iterator data_iter = mx.io.ImageRecordIter(path_imgrec="data.rec", data_shape=(3, 227, 227), batch_size=4) S3 Image Iterator #Append USE_S3=1 to config.mk before building MXNet s3_data_iter = mx.io.ImageRecordIter( path_imgrec="s3://bucketname/train/data.rec", data_shape=(3, 227, 227), batch_size=4, resize=256) Scoring mod.score(data.get_iter(batch_size), [’mse’, ’acc’]) net = mx.sym.Variable(’data’) net = mx.sym.FullyConnected(net, name=’fc1’, num_hidden=64) net = mx.sym.Activation(net, name=’relu1’, act_type="relu") net = mx.sym.FullyConnected(net, name=’fc2’, num_hidden=10) mlp = mx.sym.SoftmaxOutput(net, name=’softmax’) NDArray Iterator mx.io.NDArrayIter(data=np.zeros((1200, 3, 224, 224), dtype=’float32’), label={’annot’: np.zeros((1200, 80), dtype=’int8’), ’label’: np.zeros((1200, 1), dtype=’int8’)}, batch_size=10) CSV Iterator csv_iter = mx.io.CSVIter(data_csv=’data.csv’, data_shape = (3,), batch_size = 3) Linear Regression X = mx.sym.Variable(’data’) Y = mx.symbol.Variable(’lin_reg_label’) fully_connected_layer = mx.sym.FullyConnected(data=X, name=’fc1’, num_hidden = 1) lro = mx.sym.LinearRegressionOutput( data=fully_connected_layer, label=Y, name="lro") Logistic Regression data = mx.sym.Variable("data") target = mx.sym.Variable("target") fc = mx.sym.FullyConnected(data=data, num_hidden=1, name=’fc’) pred = mx.sym.LogisticRegressionOutput( data=fc, label=target) Save, Checkpoint, Load and Predict # save prefix = ’mymodel’ iteration = 100 model.save(prefix, iteration) # periodic checkpoint mod.fit(X, Y, eval_metric="acc", ... Convolution Factory def ConvFactory(data, num_filter, kernel, stride=(1,1), pad=(0, 0), name=None, suffix=’’): conv = mx.symbol.Convolution(data=data, num_filter=num_filter, kernel=kernel, bn = mx.symbol.BatchNorm(data=conv, name=’bn_%s%s’ %(name, suffix)) act = mx.symbol.Activation(data=bn, act_type=’relu’, name=’relu_%s%s’ %(name, suffix)) return act stride=stride, pad=pad, name=’conv_%s%s’ %(name, suffix)) prev = mx.symbol.Variable(name="Previos Output") conv_comp = ConvFactory(data=prev, num_filter=64, kernel=(7,7), stride=(2, 2)) Multiple GPUs (one-liner) Module devices = [mx.gpu(i) for i in range(num_device)] mod = mx.module.Module(context=devices, ...) mod = mx.mod.Module(symbol=net, context=mx.cpu(), data_names=[’data’], label_names=[’softmax_label’]) Fine-Tuning Learing rate scheduler sym, arg_params, aux_params = mx.model.load_checkpoint(prefix, epoch) all_layers = sym.get_internals() net = all_layers[layer_name+’_output’] net = mx.symbol.FullyConnected(data=net, num_hidden=num_classes, name=’fc’) net = mx.symbol.SoftmaxOutput(data=net, name=’softmax’) new_args = dict({k:arg_params[k] for k in arg_params if ’fc’ not in k}) lr_sch = mx.lr_scheduler.FactorScheduler(step=100, factor=0.9) mod.init_optimizer(optimizer=’sgd’, optimizer_params=((’learning_rate’, 0.1), (’ lr_scheduler’, lr_sch))) RNN: LSTM Module Low Level API epoch_end_callback=mx.callback.save.checkpoint("mymodel") # load model sym, arg_params, aux_params = mx.model.load_checkpoint(prefix, epoch_num) mod = mx.mod.Module(symbol=sym, context=mx.gpu()) mod.bind([(’data’, (BATCH_SIZE, NDARRAY_SHAPE))], for_training=False) mod.set_params(arg_params=arg_params, aux_params=aux_params) data = mx.sym.var("data") target = mx.sym.var("target") lstm1 = mx.rnn.FusedRNNCell(num_hidden=1, mode="lstm", prefix="lstm1_") mod = mx.mod.Module(symbol=net) train_iter = data.get_iter(batch_size) mod.bind(data_shapes=train_iter.provide_data, label_shapes=train_iter.provide_label) L1, L1_states = lstm1.unroll(length=seq_len, inputs=data, merge_outputs=True, layout="TNC") mod.init_params(initializer=mx.init.Xavier(magnitude=2)) mod.init_optimizer(optimizer=’sgd’, pred = mx.sym.LinearRegressionOutput(data=L1, label=target) optimizer_params=((’learning_rate’, 0.1), )) # Prediction Batch = collections.namedtuple(’Batch’, [’data’]) mod.forward(Batch(batch), is_train=False) out = mod.get_outputs()[0].asnumpy() metric = mx.metric.create(’acc’) for batch in train_iter: mod.forward(batch, is_train=True) mod.update_metric(metric, batch.label) mod.backward() mod.update()