标签：will code your Project 000 marks 495 data Final
Final Project
MGTF 495
Kaggle Deadline :11 June 2019 11:59 PM PST
Report Due Date: 11 June 2019 11:59 PM PST
1. Instructions
The report and the code for the final should be submitted on Gradescope. To secure full marks both the
report and the code should be in sync and logically correct. Please only submit relevant and legible code.
You will need to include your kaggle id, and score on the public leader board in the report. We will not be
able to grade you otherwise. Please complete the final in your groups of 5. You are also required to
mention the kaggle ids and names of your teammates in the report.
2. Overview
For the take-home final of MGTF 495, we combine the concepts we have learned so far and apply them
to another housing price prediction problem. The problem is open-ended and you can use any method /
library you like. You will find the data and the data description in the kaggle link provided below. Solutions
will be graded on Kaggle. Please follow the link - https://www.kaggle.com/c/mgtf-sp19/ to view the
webpage, you may signup using your UCSD email id (@ucsd.edu and not @eng.ucsd.edu). Note that the
times reported on Kaggle are in UTC and not PST. Your grades will be determined by your performance
on the regression task as well as your written report listing the approaches you took.
3. Files
train.csv - contains 2,051 house listings.
test.csv - contains 879 house listings. You will need to predict Sale Prices on this data.
sample submission.csv - Your solution file needs to be of this format to be acceptable.
You can download the files on the Data tab in kaggle. The description of the columns can also be found in
the same tab. Since we are asking for your code, we will check for the originality and legitimacy of the
code. You will need to cite any code or snippets referenced for this project in your report. Any unfair
practices could end up earning you a zero on your final project.
4. Task
Sale Price Prediction - Students will need to predict the Sale Price on the test.csv. Recall from your HW
that you can train models on a train dataset to predict values on the test set. The accuracy of your
submission will be measured in terms of the root mean squared error.
The public leaderboard will show your results on half of your submitted test data, but a majority of your
score will depend on the performance on the private leaderboard which will be visible to you only after the
kaggle competition ends. (We advice you to not tune your models to overfit for the public leaderboard.)
5. Grading and Evaluation
You will be graded based on your ability to obtain a solution which outperforms the benchmarks on the
test data (the unseen portion). You will be entitled to bonus marks if you perform substantially better than
benchmarks. Following are your total marks for beating a benchmark.
Baseline: 80,000
Benchmark 1: ≤ 50,000 25 marks
Benchmark 2: ≤ 40,000 30 marks
Benchmark 3: ≤ 38,000 35 marks
Benchmark 4: ≤ 35,000 40 marks
Benchmark 5: ≤ 33,000 45 marks
Benchmark 6: ≤ 30,000 50 marks
Benchmark 7: ≤ 25,000 55 marks
Bonus Benchmark 8: ≤ 22,000 60 marks Bonus
Obtain a solution which outperforms the baseline on the seen portion of the test data (i.e., the public
leaderboard) to obtain 20 marks. This is a consolation prize in case you overfit to the leaderboard.
The report accounts for 30 marks. It should describe the approaches you took to perform the task. Make
sure that the methods you describe in the report include all the aspects of your final model
代做MGTF 495作业、Python程序设计作业代写、代做Python实验作业、代写data留学生作业
including pre-processing, feature engineering etc. (if any). The aim is to enable anyone with your
report to be able to recreate your results. Even if your model doesn’t perform well, you can obtain
marks in this section for the comprehensiveness of your analysis. You can obtain a maximum of
100 marks + 10 bonus marks in this project, which will be scaled down to 40% of the total course
assessment. To obtain good performance, you need not invent new approaches (though you are more
than welcome to!).
6 Kaggle
We have set up a Kaggle page to help you evaluate your solution.
https://www.kaggle.com/c/mgtf-sp19/
You can submit only 20 submissions per day to Kaggle. This is to ensure that you don’t learn from the test
data. Please form a validation set for measuring the performance of your model. Before the competition
ends, you need to select two top submissions on which you want us to evaluate you at the end of the
competition. If you do not select, your two best submission (based on public leaderboard) will be chosen
for you.
7. Baselines
A simple baseline solution has been provided for the task. This is included in ‘Baseline.ipynb’ among the
file available on tritoned. This contains the code for reading the train data, predicting on the test data and
also for generating a submission file. The jupyter notebook shows a simple prediction. It always predicts
the average of the SalePrice from the train.csv. Reach out to either of your TAs for any clarifications.
因为专业，所以值得信赖。如有需要，请加QQ：99515681 或邮箱：99515681@qq.com
微信：codinghelp
标签：will,code,your,Project,000,marks,495,data,Final 来源： https://www.cnblogs.com/sabd/p/10986524.html
本站声明： 1. iCode9 技术分享网（下文简称本站）提供的所有内容，仅供技术学习、探讨和分享； 2. 关于本站的所有留言、评论、转载及引用，纯属内容发起人的个人观点，与本站观点和立场无关； 3. 关于本站的所有言论和文字，纯属内容发起人的个人观点，与本站观点和立场无关； 4. 本站文章均是网友提供，不完全保证技术分享内容的完整性、准确性、时效性、风险性和版权归属；如您发现该文章侵犯了您的权益，可联系我们第一时间进行删除； 5. 本站为非盈利性的个人网站，所有内容不会用来进行牟利，也不会利用任何形式的广告来间接获益，纯粹是为了广大技术爱好者提供技术内容和技术思想的分享性交流网站。