Assignment 8

[donbowen]

The link to start is here.

Yes: v_SalePrice is supposed to be missing in the holdout data! It's like you're creating a model and then testing it in the real world, where you don't know the outcome. (I'll tell you how you would have done on Wednesday.)

So, check out the README, which has the five short steps for the assignment, and 2 tips to help you along.

1. You're creating and training a model just like last assignment. In fact, basically follow the steps from Part 2 to create and hypertune a model. Part 3 of the last assignment was giving you a chance to move beyond my simple example pipelines and try out additional ML tricks.
2. After you've tuned the parameters and settled on your model, it's time to try to use your model for real. So .fit(X_train,y_train) it on all the training data so it has the best chance to succeed. Then load the holdout data, and .predict(holdout). This will give you an array of values - these are your predictions for how much those holdout sales should be for! (in logged terms)
3. I'll compare your predictions to the actual sales.

---

Please let me know ASAP (not on the due date) if you are struggling to figure out how to assemble the submission file.

But first, please try these:

1. Fit the model from the solutions file from last assignment on all the data
2. Load and .predict using the holdout data and store those predictions as a variable.
3. Your answer file for assignment 8 should have 989 rows.
4. You can use pandas functions you know and love to output the parcel ID variable and your predictions.

If you get stuck, in your discussion board post below/email let me know which of these steps is the problem.

@LeDataSciFi/classmates-2024

---

Some tips/recap:

Assignment 8 is basically 3 parts:

1. Build a model you like/love: Basically, do a high powered version of assignment 7.
2. Do a check to make sure it's ok: Optional but suggested! The ML_template file suggests that you try your favorite on the "holdout" sample. See the section called "Final evaluation on the holdout sample", which says to:
   1. Fit your favorite model on all of X_train and y_train.
   2. Get predicted values from X_test
   3. Score them against the real values in y_test.
   4. When you do that, the point is that we are just checking to make sure that your model does about as well in the "test" data as it did in the CV folds. If this is the case, continue. (If not, your model has more variance than you expected!)
3. Prepare your final predictions to submit. Assuming you name the favoritve pipeline obj theA1pipeline, the code below is basically what you need.

# train your fav model on ALL data you have (you can now combine test and train)
theA1pipeline.fit(housing,y)   

# create predictions 
holdout = ... # load the new holdout data
holdout_X_vals = ... # remove the parcel variable
y_pred = theA1pipeline.predict(holdout_X_vals) # make predictions!

# save for output: parcel number + y_pred to csv 
df_out = pd.DataFrame({'parcel':holdout['parcel'],
                       'prediction':y_pred})

df_out.to_csv('submission/MY_PREDICTIONS.csv',index=False)

# open it... does it look like the sample version? 

[adrianmross]

2 short Q's

1. I am assuming the answer is no. Are the predictions in sample_submission.csv the actual sales data?
2. My answer file includes 990 rows. Are we supposed to exclude the header row?

-- Thanks in advance

[donbowen]

1. Haha, no. They are random.
2. Yes, it should include the header row. I have 990 rows including the header.

[donbowen]

I checked for issues with assignment 8 submissions - please check if you're listed below!

If you are NOT listed below, it means I was able to compute a score for your submission.

Remember: The deadline is tomorrow at 5pm.

Valid scores (congrats) for:

• @JinyiXu
• @adrianmross
• @dannyhubert9
• @masonschick
• @Saintwy6
• @acg425
• @alz425
• @cillianfisher
• @hanadinh
• @patwilliams04
• @reghanhesser

Submitted, but unable to compute a score:

• @elvinlee12 your submission file includes a TON of rows. I think you used your model on the training data, not the holdout.

No submission:

• Everyone else!!!

@LeDataSciFi/classmates-2024

[donbowen]

I updated the lists above this comment - now 11 of youse has submitted with valid scores.

[dannyhubert9]

Hi, just confirming, does my name being here on the list mean that you were able to see my result for ASGN 8?Please confirm when you get the chance. Danny

[donbowen]

@dannyhubert9 👍

[donbowen]

It's due soon!!! Check below for your names

I see the file, but there is a scoring error:

• @elvinlee12 - you submitted like 2000 rows of data. There should be 990 (header + 989 rows of data)
• @evandeblase - your output file needs parcel numbers. Please update using the code above and resubmit.

No file seen:

• @DannyPress
• @Keith-Cheung
• @MariaMaragkelli
• @RyanChang123
• @acw224-15
• @chrisztoh
• @leosc326
• @rz780
• @slaski07