feat: Add actual_result column to LevaRunnerResults and LevaDatasetRe…

…cords This commit adds the `actual_result` column to the `leva_runner_results` and `leva_dataset_records` tables in the database. The `actual_result` column will store the actual result of each dataset record, allowing for better analysis and comparison with the predicted results. This change improves the functionality and accuracy of the Leva application. Related to recent code changes that added the `actual_result` column to the `Leva::RunnerResult` and `Leva::DatasetRecord` models, and updated the migrations and schema file to reflect these changes. Also aligns with recent repository commits that updated foreign key references in migration files and added the `actual_result` attribute to the `Leva::RunnerResult` and `Leva::DatasetRecord` models.
kieranklaassen · Aug 21, 2024 · 7931e4e · 7931e4e
1 parent b3a0e40
commit 7931e4e
Show file tree

Hide file tree

Showing 12 changed files with 102 additions and 17 deletions.
diff --git a/app/controllers/leva/runner_results_controller.rb b/app/controllers/leva/runner_results_controller.rb
@@ -0,0 +1,8 @@
+module Leva
+  class RunnerResultsController < ApplicationController
+    def show
+      @experiment = Experiment.find(params[:experiment_id])
+      @runner_result = @experiment.runner_results.find(params[:id])
+    end
+  end
+end
diff --git a/app/models/leva/dataset_record.rb b/app/models/leva/dataset_record.rb
@@ -8,6 +8,7 @@
 #  updated_at      :datetime         not null
 #  dataset_id      :integer          not null
 #  recordable_id   :integer          not null
+#  actual_result   :text
 #
 # Indexes
 #
@@ -26,6 +27,8 @@ class DatasetRecord < ApplicationRecord
     has_many :runner_results, dependent: :destroy
     has_many :evaluation_results, dependent: :destroy, through: :runner_results
 
+    validates :actual_result, presence: true
+
     # @return [Hash] A hash of attributes to be displayed in the dataset records index
     def index_attributes
       if recordable.respond_to?(:index_attributes)

diff --git a/app/models/leva/runner_result.rb b/app/models/leva/runner_result.rb
@@ -3,6 +3,7 @@
 # Table name: leva_runner_results
 #
 #  id                :integer          not null, primary key
+#  actual_result     :text
 #  prediction        :text
 #  prompt_version    :integer
 #  created_at        :datetime         not null
@@ -31,6 +32,7 @@ class RunnerResult < ApplicationRecord
     has_many :evaluation_results, dependent: :destroy
 
     validates :prediction, presence: true
+    validates :actual_result, presence: true
     validates :prompt, presence: true
   end
 end
diff --git a/app/views/leva/experiments/show.html.erb b/app/views/leva/experiments/show.html.erb
@@ -49,6 +49,7 @@
               <th class="px-6 py-3 text-left text-xs font-medium text-indigo-300 uppercase tracking-wider">Dataset Record</th>
               <th class="px-6 py-3 text-left text-xs font-medium text-indigo-300 uppercase tracking-wider">Prompt</th>
               <th class="px-6 py-3 text-left text-xs font-medium text-indigo-300 uppercase tracking-wider">Prediction</th>
+              <th class="px-6 py-3 text-left text-xs font-medium text-indigo-300 uppercase tracking-wider">Actual Result</th>
               <% @experiment.evaluation_results.group_by(&:evaluator_class).keys.each do |evaluator_class| %>
                 <th class="px-6 py-3 text-left text-xs font-medium text-indigo-300 uppercase tracking-wider"><%= evaluator_class %></th>
               <% end %>
@@ -61,7 +62,8 @@
               <tr>
                 <td class="px-6 py-4 whitespace-nowrap text-sm text-gray-300"><%= runner_result.dataset_record.display_name %></td>
                 <td class="px-6 py-4 whitespace-nowrap text-sm text-gray-300"><%= runner_result.prompt.name %> (v<%= runner_result.prompt_version %>)</td>
-                <td class="px-6 py-4 text-sm text-gray-300"><%= truncate(runner_result.prediction, length: 50) %></td>
+                <td class="px-6 py-4 text-sm text-gray-300"><%= truncate(runner_result.prediction, length: 30) %></td>
+                <td class="px-6 py-4 text-sm text-gray-300"><%= truncate(runner_result.actual_result, length: 30) %></td>
                 <% @experiment.evaluation_results.group_by(&:evaluator_class).keys.each do |evaluator_class| %>
                   <% eval_result = runner_result.evaluation_results.find_by(evaluator_class: evaluator_class) %>
                   <td class="px-6 py-4 whitespace-nowrap text-sm">
@@ -83,24 +85,14 @@
                 <% end %>
                 <td class="px-6 py-4 whitespace-nowrap text-sm text-gray-300"><%= runner_result.created_at.strftime("%Y-%m-%d %H:%M:%S") %></td>
                 <td class="px-6 py-4 whitespace-nowrap text-sm text-gray-300">
+                  <%= link_to 'View Details', experiment_runner_result_path(@experiment, runner_result), class: 'text-indigo-400 hover:underline mr-2' %>
                   <%= link_to 'Experiment', workbench_index_path(prompt_id: runner_result.prompt_id, dataset_record_id: runner_result.dataset_record_id, runner: @experiment.runner_class), class: 'text-indigo-400 hover:underline' %>
                 </td>
               </tr>
             <% end %>
           </tbody>
         </table>
       </div>
-      <div class="mt-4 text-sm text-gray-400">
-        <p>Score color legend:</p>
-        <ul class="list-disc list-inside">
-          <li><span class="text-red-500 font-semibold">0.00 - 0.19</span>: Very Poor</li>
-          <li><span class="text-orange-500 font-semibold">0.20 - 0.39</span>: Poor</li>
-          <li><span class="text-yellow-500 font-semibold">0.40 - 0.59</span>: Fair</li>
-          <li><span class="text-lime-500 font-semibold">0.60 - 0.79</span>: Good</li>
-          <li><span class="text-green-400 font-semibold">0.80 - 0.99</span>: Very Good</li>
-          <li><span class="text-green-300 font-semibold">1.00+</span>: Excellent</li>
-        </ul>
-      </div>
     <% else %>
       <p class="text-gray-400">No runner results available yet.</p>
     <% end %>

diff --git a/app/views/leva/runner_results/show.html.erb b/app/views/leva/runner_results/show.html.erb
@@ -0,0 +1,59 @@
+<% content_for :title, "Runner Result Details" %>
+<div class="container mx-auto px-4 py-8 bg-gray-950 text-white">
+  <div class="mb-8">
+    <h1 class="text-3xl font-bold text-indigo-400 mb-2">Runner Result Details</h1>
+    <%= link_to "Back to Experiment", experiment_path(@experiment), class: "text-indigo-400 hover:underline" %>
+  </div>
+  <div class="grid grid-cols-1 md:grid-cols-2 gap-6 mb-8">
+    <div class="bg-gray-800 rounded-lg shadow-lg p-6">
+      <h2 class="text-2xl font-semibold text-indigo-300 mb-4">Details</h2>
+      <p class="text-gray-400">
+        <strong class="text-indigo-300">Dataset Record:</strong>
+        <%= link_to @runner_result.dataset_record.display_name, dataset_dataset_record_path(@runner_result.dataset_record.dataset, @runner_result.dataset_record), class: "text-indigo-400 hover:underline" %>
+      </p>
+      <p class="text-gray-400">
+        <strong class="text-indigo-300">Prompt:</strong>
+        <%= link_to "#{@runner_result.prompt.name} (v#{@runner_result.prompt_version})", prompt_path(@runner_result.prompt), class: "text-indigo-400 hover:underline" %>
+      </p>
+      <p class="text-gray-400"><strong class="text-indigo-300">Created At:</strong> <%= @runner_result.created_at.strftime("%Y-%m-%d %H:%M:%S") %></p>
+      <%= link_to 'Run in Workbench', workbench_index_path(prompt_id: @runner_result.prompt_id, dataset_record_id: @runner_result.dataset_record_id, runner: @experiment.runner_class), class: 'mt-4 inline-block px-4 py-2 bg-indigo-600 text-white rounded hover:bg-indigo-700 transition-colors duration-200' %>
+    </div>
+    <div class="bg-gray-800 rounded-lg shadow-lg p-6">
+      <h2 class="text-2xl font-semibold text-indigo-300 mb-4">Evaluation Results</h2>
+      <% if @runner_result.evaluation_results.any? %>
+        <div class="space-y-4">
+          <% @runner_result.evaluation_results.each do |eval_result| %>
+            <div class="bg-gray-700 rounded-lg p-4">
+              <h3 class="text-lg font-semibold text-indigo-200 mb-2"><%= eval_result.evaluator_class %></h3>
+              <% score = eval_result.score %>
+              <% color_class = case score
+                 when 0...0.2 then 'text-red-500'
+                 when 0.2...0.4 then 'text-orange-500'
+                 when 0.4...0.6 then 'text-yellow-500'
+                 when 0.6...0.8 then 'text-lime-500'
+                 when 0.8...1.0 then 'text-green-400'
+                 else 'text-green-300'
+                 end %>
+              <p class="text-xl font-bold <%= color_class %>"><%= sprintf('%.2f', score) %></p>
+            </div>
+          <% end %>
+        </div>
+      <% else %>
+        <p class="text-gray-400">No evaluation results available.</p>
+      <% end %>
+    </div>
+  </div>
+  <div class="bg-gray-800 rounded-lg shadow-lg p-6 mb-8">
+    <h2 class="text-2xl font-semibold text-indigo-300 mb-4">Prediction and Actual Result</h2>
+    <div class="grid grid-cols-1 md:grid-cols-2 gap-6">
+      <div>
+        <h3 class="text-xl font-semibold text-indigo-200 mb-2">Prediction</h3>
+        <pre class="bg-gray-700 p-4 rounded-lg mt-2 text-sm text-gray-300 whitespace-pre-wrap"><%= @runner_result.prediction %></pre>
+      </div>
+      <div>
+        <h3 class="text-xl font-semibold text-indigo-200 mb-2">Actual Result</h3>
+        <pre class="bg-gray-700 p-4 rounded-lg mt-2 text-sm text-gray-300 whitespace-pre-wrap"><%= @runner_result.actual_result %></pre>
+      </div>
+    </div>
+  </div>
+</div>
diff --git a/app/views/leva/workbench/_results_section.html.erb b/app/views/leva/workbench/_results_section.html.erb
@@ -75,8 +75,17 @@
         <% if @dataset_record %>
           <% evaluation_result = @dataset_record.evaluation_results.for_evaluator(evaluator_class).last %>
           <% if evaluation_result %>
-            <div class="text-sm text-gray-300">
-              Score: <%= evaluation_result.score %>
+            <% score = evaluation_result.score %>
+            <% color_class = case score
+               when 0...0.2 then 'text-red-500'
+               when 0.2...0.4 then 'text-orange-500'
+               when 0.4...0.6 then 'text-yellow-500'
+               when 0.6...0.8 then 'text-lime-500'
+               when 0.8...1.0 then 'text-green-400'
+               else 'text-green-300'
+               end %>
+            <div class="text-sm <%= color_class %> font-semibold">
+              Score: <%= sprintf('%.2f', score) %>
             </div>
           <% else %>
             <div class="text-sm text-gray-500">

diff --git a/config/routes.rb b/config/routes.rb
@@ -4,7 +4,9 @@
   resources :datasets do
     resources :dataset_records, path: 'records', only: [:index, :show]
   end
-  resources :experiments, except: [:destroy]
+  resources :experiments, except: [:destroy] do
+    resources :runner_results, only: [:show]
+  end
   resources :prompts
   resources :workbench, only: [:index, :new, :create, :edit, :update] do
     collection do

diff --git a/db/migrate/20240821191713_add_actual_result_to_leva_dataset_records.rb b/db/migrate/20240821191713_add_actual_result_to_leva_dataset_records.rb
@@ -0,0 +1,5 @@
+class AddActualResultToLevaDatasetRecords < ActiveRecord::Migration[7.2]
+  def change
+    add_column :leva_runner_results, :actual_result, :text
+  end
+end
diff --git a/lib/leva.rb b/lib/leva.rb
@@ -56,7 +56,8 @@ def execute_and_store(experiment, dataset_record, prompt)
         experiment: experiment,
         dataset_record: dataset_record,
         prompt: prompt,
-        prediction: result
+        prediction: result,
+        actual_result: dataset_record.actual_result
       )
     end
   end

diff --git a/test/dummy/db/schema.rb b/test/dummy/db/schema.rb
@@ -10,13 +10,14 @@
 #
 # It's strongly recommended that you check this file into your version control system.
 
-ActiveRecord::Schema[7.2].define(version: 2024_08_21_183153) do
+ActiveRecord::Schema[7.2].define(version: 2024_08_21_191713) do
   create_table "leva_dataset_records", force: :cascade do |t|
     t.integer "dataset_id", null: false
     t.string "recordable_type", null: false
     t.integer "recordable_id", null: false
     t.datetime "created_at", null: false
     t.datetime "updated_at", null: false
+    t.text "actual_result"
     t.index ["dataset_id"], name: "index_leva_dataset_records_on_dataset_id"
     t.index ["recordable_type", "recordable_id"], name: "index_leva_dataset_records_on_recordable"
   end
@@ -74,6 +75,7 @@
     t.datetime "updated_at", null: false
     t.integer "prompt_version"
     t.integer "prompt_id", null: false
+    t.text "actual_result"
     t.index ["dataset_record_id"], name: "index_leva_runner_results_on_dataset_record_id"
     t.index ["experiment_id"], name: "index_leva_runner_results_on_experiment_id"
     t.index ["prompt_id"], name: "index_leva_runner_results_on_prompt_id"

diff --git a/test/models/leva/dataset_record_test.rb b/test/models/leva/dataset_record_test.rb
@@ -3,6 +3,7 @@
 # Table name: leva_dataset_records
 #
 #  id              :integer          not null, primary key
+#  actual_result   :text
 #  recordable_type :string           not null
 #  created_at      :datetime         not null
 #  updated_at      :datetime         not null

diff --git a/test/models/leva/runner_result_test.rb b/test/models/leva/runner_result_test.rb
@@ -3,6 +3,7 @@
 # Table name: leva_runner_results
 #
 #  id                :integer          not null, primary key
+#  actual_result     :text
 #  prediction        :text
 #  prompt_version    :integer
 #  created_at        :datetime         not null