Added more gpt analysis unit tests. Improved prompts so that answers are given for every question for better result parsing

Author: samtirz07

server/src/gpt-controller.ts

@@ -49,7 +49,8 @@ export class GPTController {
       const loopPromises = Array.from({ length: 3 }, async (_) => {
         const assistant = await this.createAssistant(assistantParams);
         const thread = await this.createThread(threadMessage);
-        // Run the assistant on the thread and get the prompt results. Think non-stream results are better?
+        
+        // Run the assistant on the thread and get the prompt results
         let run = await GPTController.client.beta.threads.runs.createAndPoll(
           thread.id,
           {
@@ -67,19 +68,20 @@ export class GPTController {
               var result = message.content[0].text.value;
               if(n % 2 == 0) {                                // Every second message has the data values
                 // console.log(`${message.role} > ${result}`); // FOR TESTING
-                let preres = result.split("ø").map((s) => s.replace("\n", ""));
+                let preres = result.split("ø").map((s) => s.replace("\n", "") && s.replace(/^\s+/g, ""));   // Trimming each string
                 console.log(preres)
                 var resvalues: GPTData =  {
                   paper_name: preres[0],
                   year: parseInt(preres[1]),
-                  author: preres[2].split(",").map((s) => s.replace(/^\s+/g, "")),
+                  author: preres[2].split("¶").map((s) => s.replace(/^\s+/g, "")),
                   part_no: preres[3],
                   type: preres[4],
                   manufacturer: preres[5],
                   testing_location: <TestLocation>preres[6],
-                  testing_type: <Testing>preres[7],     // TODO: preres[7] is a list ("TID, TID, DD") sometimes so the cast may fail
-                                                        // Produces weird output: "SEE【4:0†source】"
-                  data_type: 0                          // TODO: add a prompt to get data_type
+                  testing_type: <Testing>preres[7],
+                  // TODO: preres[7] is a list ("TID, TID, DD") if the paper has more than one testing type, so the cast may fail
+                          // Produces weird output: "SEE【4:0†source】"
+                  data_type: 0                          // TODO: add a prompt to get number data_type. What is it?
                 };
                 console.log(resvalues)
                 threadResults.push(resvalues);

server/src/index.ts

@@ -42,7 +42,7 @@ async function initializeSystem(): Promise<{dbController: DatabaseController, gp
     filename: "./database.db",
     driver: sqlite3.Database,
   });
-  return {dbController: new DatabaseController(db), gptController: new GPTController(GPTModel.GPT3_5Turbo)};
+  return {dbController: new DatabaseController(db), gptController: new GPTController(GPTModel.GPT4Turbo)};
 }
 
 initializeSystem().then(({dbController, gptController}) => {

server/src/prompts.data.ts

@@ -1,8 +1,8 @@
 export const questions = [
   "What is the title of the paper",
   "Which year was the paper published",
-  "What are all of the author's names, in the format (J. Doe) in a comma separated list",
-  "What is the part number (or name if that is not available) of the part or parts studied by the paper",
+  "What are all of the author's names, in the format (J. Doe) in a \"¶\" separated list",
+  "What is the part no. of the part studied by the paper",        // Returns same thing as part type if part name is mentioned
   "What is the type of part or parts (eg, switching regulator) mentioned in the paper",
   "Who is the manufacturer",
   "What is the type of testing location: Respond to this question with \"Terrestrial\" for a terrestial testing location, or \"Flight\" for a flight testing location",
@@ -11,8 +11,8 @@ export const questions = [
 
 export const prompt = `Please answer the following questions, as concisely as possible, and with a heavy emphasis on numbers instead of words.\n
     Use standard text and do not provide citations for each of your answers. 
-    Answer each question, and separate the answers with a "ø" character as a delimiter.
-    If you are unable to answer the question accurately, provide the answer N/A.\n`;
+    If you are unable to answer the question accurately, provide the answer N/A.
+    Answer each question, and separate the answers with a "ø" character as a delimiter.\n`;
 
 export const Other_targeted_questions = [
   "What type was the radiation source",

server/test/gpt-controller.test.ts

@@ -32,3 +32,37 @@ test("Insertion of a paper with one author", async () => {
 
   await expect(testGPT.runGPTAnalysis([paper])).resolves.not.toThrow();
 }, 60000);
+
+test('Insertion of a paper with five authors', async () => {
+  const paper: string = path.resolve(__dirname, 
+    './testfiles/SEE_in-flight_data_for_two_static_32KB_memories_on_high_earth_orbit.pdf');
+  fs.readFileSync(paper);
+
+  await expect(testGPT.runGPTAnalysis([paper])).resolves.not.toThrow();
+}, 60000);
+
+// Below gpt tests commented out to reduce token usage during testing
+
+// test('Insertion of a paper with two authors', async () => {
+//   const paper: string = path.resolve(__dirname, 
+//     './testfiles/Solar_flare_proton_environment_for_estimating_downtime_of_spacecraft_CCDs.pdf');
+//   fs.readFileSync(paper);
+
+//   await expect(testGPT.runGPTAnalysis([paper])).resolves.not.toThrow();
+// }, 60000);
+
+// test('Insertion of a paper with three authors. 1 of 2', async () => {
+//   const paper: string = path.resolve(__dirname, 
+//     './testfiles/A_radiation_tolerant_video_camera_for_high_total_dose_environments.pdf');
+//   fs.readFileSync(paper);
+
+//   await expect(testGPT.runGPTAnalysis([paper])).resolves.not.toThrow();
+// }, 60000);
+
+// test('Insertion of a paper with three authors. 2 of 2', async () => {
+//   const paper: string = path.resolve(__dirname, 
+//     './testfiles/Radiation_effects_predicted_observed_and_compared_for_spacecraft_systems.pdf');
+//   fs.readFileSync(paper);
+
+//   await expect(testGPT.runGPTAnalysis([paper])).resolves.not.toThrow();
+// }, 60000);