{ "cells": [ { "cell_type": "markdown", "id": "005b80b1-9dbc-47cd-bde3-2c0da024b73a", "metadata": {}, "source": [ "Inspect" ] }, { "cell_type": "code", "execution_count": 15, "id": "39e4c357-02b1-42dd-8575-9d968ba25683", "metadata": {}, "outputs": [], "source": [ "csv_path = \"prediction_results.csv\" " ] }, { "cell_type": "code", "execution_count": 7, "id": "ce23b503-3adf-47de-b637-99c74c4a6633", "metadata": {}, "outputs": [ { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
QuestionPredicted AnswerCorrect Answer
2We must assume that he is a Communist, because...AD
14A water bed sells for $330 cash or $40 down an...JD
16The team that wins in a tug-of-war is the team...HA
32In a circuit of two lamps in parallel, if the ...HB
37Consider an arbitrage-free securities market m...BA
............
1127The annular space between two concentricalumin...BD
1129A 0.1 mm thick neutral density filter gives 70...IA
1130Air (100°F, 1atm) is flowing at a velocity of ...AI
1132This question refers to the following informat...AC
1134Order the following (risk, return) pairs from ...EG
\n", "

300 rows × 3 columns

\n", "
" ], "text/plain": [ " Question Predicted Answer \\\n", "2 We must assume that he is a Communist, because... A \n", "14 A water bed sells for $330 cash or $40 down an... J \n", "16 The team that wins in a tug-of-war is the team... H \n", "32 In a circuit of two lamps in parallel, if the ... H \n", "37 Consider an arbitrage-free securities market m... B \n", "... ... ... \n", "1127 The annular space between two concentricalumin... B \n", "1129 A 0.1 mm thick neutral density filter gives 70... I \n", "1130 Air (100°F, 1atm) is flowing at a velocity of ... A \n", "1132 This question refers to the following informat... A \n", "1134 Order the following (risk, return) pairs from ... E \n", "\n", " Correct Answer \n", "2 D \n", "14 D \n", "16 A \n", "32 B \n", "37 A \n", "... ... \n", "1127 D \n", "1129 A \n", "1130 I \n", "1132 C \n", "1134 G \n", "\n", "[300 rows x 3 columns]" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "\n", "Showing first 300 rows out of 3271 incorrect answers.\n", "\n", "Accuracy: 72.69%\n" ] } ], "source": [ "import pandas as pd\n", "from tabulate import tabulate\n", "from IPython.display import display\n", "\n", "def display_incorrect_answers(csv_path):\n", " df = pd.read_csv(csv_path)\n", " \n", " incorrect_answers = df[df['Predicted Answer'] != df['Correct Answer']]\n", " \n", " max_rows = 300 # Adjust as needed\n", " incorrect_answers_limited = incorrect_answers[['Question', 'Predicted Answer', 'Correct Answer']].head(max_rows)\n", " \n", " display(pd.DataFrame(incorrect_answers_limited))\n", " \n", " if len(incorrect_answers) > max_rows:\n", " print(f\"\\nShowing first {max_rows} rows out of {len(incorrect_answers)} incorrect answers.\")\n", " \n", " total_questions = len(df)\n", " incorrect_count = len(incorrect_answers)\n", " accuracy = ((total_questions - incorrect_count) / total_questions) * 100\n", " print(f\"\\nAccuracy: {accuracy:.2f}%\")\n", "\n", "display_incorrect_answers(csv_path)\n" ] }, { "cell_type": "code", "execution_count": 13, "id": "6ad0e4b2-3845-4081-a686-9d2eac98fc25", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "All 'Is Correct' rows have matching answers.\n" ] } ], "source": [ "import pandas as pd\n", "from IPython.display import display\n", "\n", "def check_correct_answers(csv_path):\n", " df = pd.read_csv(csv_path)\n", " \n", " correct_mismatches = df[(df['Is Correct'] == True) & (df['Predicted Answer'] != df['Correct Answer'])]\n", " \n", " if not correct_mismatches.empty:\n", " print(\"\\nWarning: Some 'Is Correct' rows have mismatched answers!\")\n", " display(correct_mismatches[['Question', 'Predicted Answer', 'Correct Answer', 'Is Correct']])\n", " else:\n", " print(\"\\nAll 'Is Correct' rows have matching answers.\")\n", "\n", " \n", "check_correct_answers(csv_path)\n" ] }, { "cell_type": "code", "execution_count": 19, "id": "b2a9ed17-8e39-4f81-8a27-358e948967de", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Incorrect Rows\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
QuestionPredicted AnswerCorrect AnswerIs Correct
2We must assume that he is a Communist, because...AD❌
14A water bed sells for $330 cash or $40 down an...JD❌
16The team that wins in a tug-of-war is the team...HA❌
32In a circuit of two lamps in parallel, if the ...HB❌
37Consider an arbitrage-free securities market m...BA❌
...............
11968Four years ago the owner of a shopping center ...IE❌
11972An electrical current flows along a flat plate...FNaN❌
11976A recent law school graduate took and passed t...BA❌
11977LC oscillators are used for produce a waveform...DE❌
11978A 50-hp, 500-volt shunt motor draws a line cur...DH❌
\n", "

3271 rows × 4 columns

\n", "
" ], "text/plain": [ " Question Predicted Answer \\\n", "2 We must assume that he is a Communist, because... A \n", "14 A water bed sells for $330 cash or $40 down an... J \n", "16 The team that wins in a tug-of-war is the team... H \n", "32 In a circuit of two lamps in parallel, if the ... H \n", "37 Consider an arbitrage-free securities market m... B \n", "... ... ... \n", "11968 Four years ago the owner of a shopping center ... I \n", "11972 An electrical current flows along a flat plate... F \n", "11976 A recent law school graduate took and passed t... B \n", "11977 LC oscillators are used for produce a waveform... D \n", "11978 A 50-hp, 500-volt shunt motor draws a line cur... D \n", "\n", " Correct Answer Is Correct \n", "2 D ❌ \n", "14 D ❌ \n", "16 A ❌ \n", "32 B ❌ \n", "37 A ❌ \n", "... ... ... \n", "11968 E ❌ \n", "11972 NaN ❌ \n", "11976 A ❌ \n", "11977 E ❌ \n", "11978 H ❌ \n", "\n", "[3271 rows x 4 columns]" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "def check_incorrect_answers(csv_path):\n", " df = pd.read_csv(csv_path)\n", " print(\"Incorrect Rows\")\n", " correct_mismatches = df[(df['Predicted Answer'] != df['Correct Answer'])]\n", " display(correct_mismatches[['Question', 'Predicted Answer', 'Correct Answer', 'Is Correct']])\n", "\n", "check_incorrect_answers(csv_path)\n" ] }, { "cell_type": "code", "execution_count": null, "id": "501ed411-7145-4e8e-ae07-439fea4b38f7", "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "Python 3 (ipykernel)", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.10.16" } }, "nbformat": 4, "nbformat_minor": 5 }