Update: write and read the model, train and predict perspectively

src/test/java/org/apache/sysds/test/gpu/multigpu/GPUTest.java

@@ -40,7 +40,8 @@ public abstract class GPUTest extends AutomatedTestBase {
     protected static final String TEST_CLASS_DIR = TEST_DIR + MultiGPUTest.class.getSimpleName() + "/";
     protected static final String SINGLE_GPU_TEST = "SingleGPUTest";
     protected static final String MULTI_GPUS_TEST = "MultiGPUsTest";
-    protected static final String TEST_NAME = "GPUTest";
+    protected static final String TEST_NAME = "InferenceScript";
+    protected static final String TRAIN_SCRIPT = "TrainScript";
     protected static final String DATA_SET = DATASET_DIR + "MNIST/mnist_test.csv";
     protected static final String SINGLE_TEST_CONFIG = CONFIG_DIR + "SystemDS-SingleGPU-config.xml";
     protected static final String MULTI_TEST_CONFIG = CONFIG_DIR + "SystemDS-config.xml";
@@ -90,6 +91,24 @@ protected void runMultiGPUsTest(boolean multiGPUs, int numTestImages) {
         appender.clearLogMessages();
     }
 
+    /**
+     * Run the training script
+     */
+    protected void runTrainingScript(boolean multiGPUs, int numTestImages) {
+        getAndLoadTestConfiguration(multiGPUs ? MULTI_GPUS_TEST : SINGLE_GPU_TEST);
+
+        String HOME = SCRIPT_DIR + TEST_DIR;
+        fullDMLScriptName = HOME + TRAIN_SCRIPT + ".dml";
+        programArgs = new String[] { "-args", DATA_SET, output("R"), Integer.toString(numTestImages), "-config",
+                multiGPUs ? MULTI_TEST_CONFIG : SINGLE_TEST_CONFIG };
+        fullRScriptName = HOME + TEST_NAME + ".R";
+
+        rCmd = null;
+        InMemoryAppender appender = configureLog4j();
+
+        runTest(true, false, null, -1);
+    }
+
     protected static InMemoryAppender configureLog4j() {
         Logger rootLogger = Logger.getRootLogger();
         rootLogger.setLevel(Level.ERROR);

src/test/java/org/apache/sysds/test/gpu/multigpu/SingleGPUTest.java

@@ -64,6 +64,9 @@ public void test01_gpuTest_500k() {
 
     @Override
     protected void runMultiGPUsTest(boolean multiGPUs, int numTestImages) {
+        // Train the model first
+        super.runTrainingScript(multiGPUs, numTestImages);
+
         long startTime = System.nanoTime();
         super.runMultiGPUsTest(multiGPUs, numTestImages);
         long endTime = System.nanoTime();

src/test/scripts/gpu/InferenceScript.dml

@@ -0,0 +1,73 @@
+#-------------------------------------------------------------
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+#-------------------------------------------------------------
+
+source("src/test/scripts/applications/nn/component/efficientNet.dml") as eff
+
+# Read training data
+data = read($1, format="csv")
+
+N = nrow(data)
+
+# Extract images and labels
+images = data[,2:ncol(data)]
+labels = data[,1]
+
+# Scale images to [0,1], and one-hot encode the labels
+images = images / 255.0
+labels = table(seq(1, N), labels+1, N, 10)
+
+# Load the trained model
+model = read("output/model.txt", format="text")
+
+# Generate random data for predicting
+num_test_images = as.integer($3)
+test_images = rand(rows=num_test_images, cols=28*28, min=0, max=1, sparsity=1.0, pdf="uniform", seed=42)
+test_labels = rand(rows=num_test_images, cols=1, min=0, max=9, sparsity=1.0, pdf="uniform", seed=42)
+test_labels = round(test_labels)
+
+# One-hot encode the test labels
+test_labels = table(seq(1, num_test_images), test_labels+1, num_test_images, 10)
+
+# Materialize intermediates by computing and printing their sums
+print("Sum of test_images: " + sum(test_images))
+print("Sum of test_labels: " + sum(test_labels))
+
+# Also Predict in Batches since otherwise we can run into Memory Issues
+# Could be unnecessary on more powerful machines :)
+batch_size = 1024  # Adjust the batch size to a larger value
+iters = ceil(num_test_images / batch_size)
+partial_accuracies = matrix(0, rows=iters, cols=1)
+
+# Start timing the parfor loop
+parfor_start_time = time()
+parfor(i in 1:iters) {
+  beg = ((i-1) * batch_size) %% num_test_images + 1
+  end = min(num_test_images, beg + batch_size - 1)
+  X_batch = test_images[beg:end,]
+  y_batch = test_labels[beg:end,]
+
+  pred = eff::netPredict(X_batch, model, 1, 28, 28)
+  partial_accuracies[i,1] = mean(rowIndexMax(pred) == rowIndexMax(y_batch))
+}
+# End timing the parfor loop
+parfor_end_time = time()
+parfor_execution_time = floor((parfor_end_time-parfor_start_time)/1000000000)
+print("Parfor Execution Time: " + parfor_execution_time)

src/test/scripts/gpu/TrainScript.dml

@@ -0,0 +1,45 @@
+#-------------------------------------------------------------
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+#-------------------------------------------------------------
+
+source("src/test/scripts/applications/nn/component/efficientNet.dml") as eff
+
+# Read training data
+data = read($1, format="csv")
+
+N = nrow(data)
+
+# Extract images and labels
+images = data[,2:ncol(data)]
+labels = data[,1]
+
+# Scale images to [0,1], and one-hot encode the labels
+images = images / 255.0
+labels = table(seq(1, N), labels+1, N, 10)
+
+# Initialize and train the model
+model = eff::initNetwork(1, 10, -1)
+epochs = 1
+batch_size = 256
+model = eff::netTrain(model, images, 1, 28, 28, labels, epochs, batch_size, 0.025, 0.9, TRUE)
+
+# Save the trained model to disk
+write(model, "output/model.txt", format="text")
+print("Trained model saved to output/model.bin")