feat: add a base model to the repository

Author: kkweon

model/ver20220624/model.json

@@ -0,0 +1 @@
+{"modelTopology":{"class_name":"Sequential","config":{"name":"sequential_1","layers":[{"class_name":"Dense","config":{"units":32,"activation":"relu","use_bias":true,"kernel_initializer":{"class_name":"VarianceScaling","config":{"scale":1,"mode":"fan_avg","distribution":"normal","seed":null}},"bias_initializer":{"class_name":"Zeros","config":{}},"kernel_regularizer":null,"bias_regularizer":null,"activity_regularizer":null,"kernel_constraint":null,"bias_constraint":null,"name":"dense_Dense1","trainable":true,"batch_input_shape":[null,512]}},{"class_name":"BatchNormalization","config":{"axis":-1,"momentum":0.99,"epsilon":0.001,"center":true,"scale":true,"beta_initializer":{"class_name":"Zeros","config":{}},"gamma_initializer":{"class_name":"Ones","config":{}},"moving_mean_initializer":{"class_name":"Zeros","config":{}},"moving_variance_initializer":{"class_name":"Ones","config":{}},"beta_regularizer":null,"gamma_regularizer":null,"beta_constraint":null,"gamma_constraint":null,"name":"batch_normalization_BatchNormalization1","trainable":true}},{"class_name":"Dense","config":{"units":32,"activation":"relu","use_bias":true,"kernel_initializer":{"class_name":"VarianceScaling","config":{"scale":1,"mode":"fan_avg","distribution":"normal","seed":null}},"bias_initializer":{"class_name":"Zeros","config":{}},"kernel_regularizer":null,"bias_regularizer":null,"activity_regularizer":null,"kernel_constraint":null,"bias_constraint":null,"name":"dense_Dense2","trainable":true}},{"class_name":"BatchNormalization","config":{"axis":-1,"momentum":0.99,"epsilon":0.001,"center":true,"scale":true,"beta_initializer":{"class_name":"Zeros","config":{}},"gamma_initializer":{"class_name":"Ones","config":{}},"moving_mean_initializer":{"class_name":"Zeros","config":{}},"moving_variance_initializer":{"class_name":"Ones","config":{}},"beta_regularizer":null,"gamma_regularizer":null,"beta_constraint":null,"gamma_constraint":null,"name":"batch_normalization_BatchNormalization2","trainable":true}},{"class_name":"Dense","config":{"units":1,"activation":"sigmoid","use_bias":true,"kernel_initializer":{"class_name":"VarianceScaling","config":{"scale":1,"mode":"fan_avg","distribution":"normal","seed":null}},"bias_initializer":{"class_name":"Zeros","config":{}},"kernel_regularizer":null,"bias_regularizer":null,"activity_regularizer":null,"kernel_constraint":null,"bias_constraint":null,"name":"dense_Dense3","trainable":true}}]},"keras_version":"tfjs-layers 3.18.0","backend":"tensor_flow.js"},"weightsManifest":[{"paths":["weights.bin"],"weights":[{"name":"dense_Dense1/kernel","shape":[512,32],"dtype":"float32"},{"name":"dense_Dense1/bias","shape":[32],"dtype":"float32"},{"name":"batch_normalization_BatchNormalization1/gamma","shape":[32],"dtype":"float32"},{"name":"batch_normalization_BatchNormalization1/beta","shape":[32],"dtype":"float32"},{"name":"dense_Dense2/kernel","shape":[32,32],"dtype":"float32"},{"name":"dense_Dense2/bias","shape":[32],"dtype":"float32"},{"name":"batch_normalization_BatchNormalization2/gamma","shape":[32],"dtype":"float32"},{"name":"batch_normalization_BatchNormalization2/beta","shape":[32],"dtype":"float32"},{"name":"dense_Dense3/kernel","shape":[32,1],"dtype":"float32"},{"name":"dense_Dense3/bias","shape":[1],"dtype":"float32"},{"name":"batch_normalization_BatchNormalization1/moving_mean","shape":[32],"dtype":"float32"},{"name":"batch_normalization_BatchNormalization1/moving_variance","shape":[32],"dtype":"float32"},{"name":"batch_normalization_BatchNormalization2/moving_mean","shape":[32],"dtype":"float32"},{"name":"batch_normalization_BatchNormalization2/moving_variance","shape":[32],"dtype":"float32"},{"name":"iter","shape":[],"dtype":"int32","group":"optimizer"}]}],"format":"layers-model","generatedBy":"TensorFlow.js tfjs-layers v3.18.0","convertedBy":null,"trainingConfig":{"loss":{},"metrics":["binary_accuracy","precision","recall"],"optimizer_config":{"class_name":"Adam","config":{"learningRate":0.001,"beta1":0.9,"beta2":0.999,"epsilon":1e-7}}}}

model/ver20220624/weights.bin

@@ -0,0 +1,31 @@
+import * as tf from "@tensorflow/tfjs-node";
+
+export async function getModel(
+  modelPath: string
+): Promise<tf.LayersModel | tf.Sequential> {
+  try {
+    console.info(`Trying to load a model from ${modelPath}`);
+    return await tf.loadLayersModel(modelPath);
+  } catch (e) {
+    console.warn(`Unable to load a model. Creating a new model`);
+    return tf.sequential({
+      layers: [
+        tf.layers.dense({
+          inputDim: 512,
+          units: 32,
+          activation: "relu",
+        }),
+        tf.layers.batchNormalization(),
+        tf.layers.dense({
+          units: 32,
+          activation: "relu",
+        }),
+        tf.layers.batchNormalization(),
+        tf.layers.dense({
+          units: 1,
+          activation: "sigmoid",
+        }),
+      ],
+    });
+  }
+}

trainer/model.ts

@@ -4,7 +4,7 @@
   "description": "",
   "main": "index.js",
   "scripts": {
-    "build": "npx ts-node trainer.ts"
+    "start": "npx ts-node trainer.ts"
   },
   "keywords": [],
   "author": "",

trainer/package.json

@@ -1,9 +1,19 @@
 import * as use from "@tensorflow-models/universal-sentence-encoder";
 import * as tf from "@tensorflow/tfjs-node";
 
+import path from "path";
+import { getModel } from "./model";
+
+const MODEL_PATH = `file://${path.join(
+  __dirname,
+  "..",
+  "model",
+  "ver20220624"
+)}`;
+
 async function main() {
   const encoder = await use.load();
-  const trainData = tf.data
+  const trainData = await tf.data
     .csv(
       "https://raw.githubusercontent.com/smilegate-ai/korean_unsmile_dataset/main/unsmile_train_v1.0.tsv",
       {
@@ -28,10 +38,11 @@ async function main() {
         ys: Object.values(data.ys),
       };
     })
+    .prefetch(10000)
     .batch(32)
     .shuffle(32);
 
-    const valData = tf.data
+  const valData = tf.data
     .csv(
       "https://raw.githubusercontent.com/smilegate-ai/korean_unsmile_dataset/main/unsmile_valid_v1.0.tsv",
       {
@@ -56,38 +67,36 @@ async function main() {
         ys: Object.values(data.ys),
       };
     })
+    .prefetch(10000)
     .batch(32)
-    .shuffle(32);    
+    .shuffle(32);
 
-  const model = tf.sequential({
-    layers: [
-      tf.layers.dense({
-        inputDim: 512,
-        units: 512,
-        activation: "relu",
-      }),
-      tf.layers.batchNormalization(),
-      tf.layers.dense({
-        units: 512,
-        activation: "relu",
-      }),
-      tf.layers.batchNormalization(),
-      tf.layers.dense({
-        units: 1,
-        activation: "sigmoid",
-      }),
-    ],
-  });
+  const model = await getModel(MODEL_PATH);
 
   model.compile({
     optimizer: tf.train.adam(),
     loss: tf.losses.sigmoidCrossEntropy,
-    metrics: [tf.metrics.binaryAccuracy],
+    metrics: [
+      tf.metrics.binaryAccuracy,
+      tf.metrics.precision,
+      tf.metrics.recall,
+    ],
   });
 
+  model.summary();
+
   model.fitDataset(trainData, {
     epochs: 5,
-    validationData: valData
+    validationData: valData,
+    callbacks: [
+      tf.callbacks.earlyStopping({
+        patience: 1,
+      }),
+    ],
+  });
+
+  model.save(MODEL_PATH, {
+    includeOptimizer: true,
   });
 }
 

trainer/trainer.ts

@@ -11,7 +11,7 @@
     // "disableReferencedProjectLoad": true,             /* Reduce the number of projects loaded automatically by TypeScript. */
 
     /* Language and Environment */
-    "target": "es2016",                                  /* Set the JavaScript language version for emitted JavaScript and include compatible library declarations. */
+    "target": "es2016" /* Set the JavaScript language version for emitted JavaScript and include compatible library declarations. */,
     // "lib": [],                                        /* Specify a set of bundled library declaration files that describe the target runtime environment. */
     // "jsx": "preserve",                                /* Specify what JSX code is generated. */
     // "experimentalDecorators": true,                   /* Enable experimental support for TC39 stage 2 draft decorators. */
@@ -25,7 +25,7 @@
     // "moduleDetection": "auto",                        /* Control what method is used to detect module-format JS files. */
 
     /* Modules */
-    "module": "commonjs",                                /* Specify what module code is generated. */
+    "module": "commonjs" /* Specify what module code is generated. */,
     // "rootDir": "./",                                  /* Specify the root folder within your source files. */
     // "moduleResolution": "node",                       /* Specify how TypeScript looks up a file from a given module specifier. */
     // "baseUrl": "./",                                  /* Specify the base directory to resolve non-relative module names. */
@@ -57,7 +57,7 @@
     // "downlevelIteration": true,                       /* Emit more compliant, but verbose and less performant JavaScript for iteration. */
     // "sourceRoot": "",                                 /* Specify the root path for debuggers to find the reference source code. */
     // "mapRoot": "",                                    /* Specify the location where debugger should locate map files instead of generated locations. */
-    // "inlineSourceMap": true,                          /* Include sourcemap files inside the emitted JavaScript. */
+    "inlineSourceMap": true /* Include sourcemap files inside the emitted JavaScript. */,
     // "inlineSources": true,                            /* Include source code in the sourcemaps inside the emitted JavaScript. */
     // "emitBOM": true,                                  /* Emit a UTF-8 Byte Order Mark (BOM) in the beginning of output files. */
     // "newLine": "crlf",                                /* Set the newline character for emitting files. */
@@ -71,12 +71,12 @@
     /* Interop Constraints */
     // "isolatedModules": true,                          /* Ensure that each file can be safely transpiled without relying on other imports. */
     // "allowSyntheticDefaultImports": true,             /* Allow 'import x from y' when a module doesn't have a default export. */
-    "esModuleInterop": true,                             /* Emit additional JavaScript to ease support for importing CommonJS modules. This enables 'allowSyntheticDefaultImports' for type compatibility. */
+    "esModuleInterop": true /* Emit additional JavaScript to ease support for importing CommonJS modules. This enables 'allowSyntheticDefaultImports' for type compatibility. */,
     // "preserveSymlinks": true,                         /* Disable resolving symlinks to their realpath. This correlates to the same flag in node. */
-    "forceConsistentCasingInFileNames": true,            /* Ensure that casing is correct in imports. */
+    "forceConsistentCasingInFileNames": true /* Ensure that casing is correct in imports. */,
 
     /* Type Checking */
-    "strict": true,                                      /* Enable all strict type-checking options. */
+    "strict": true /* Enable all strict type-checking options. */,
     // "noImplicitAny": true,                            /* Enable error reporting for expressions and declarations with an implied 'any' type. */
     // "strictNullChecks": true,                         /* When type checking, take into account 'null' and 'undefined'. */
     // "strictFunctionTypes": true,                      /* When assigning functions, check to ensure parameters and the return values are subtype-compatible. */
@@ -98,6 +98,6 @@
 
     /* Completeness */
     // "skipDefaultLibCheck": true,                      /* Skip type checking .d.ts files that are included with TypeScript. */
-    "skipLibCheck": true                                 /* Skip type checking all .d.ts files. */
+    "skipLibCheck": true /* Skip type checking all .d.ts files. */
   }
 }