|
|
| (не показаны 3 промежуточные версии этого же участника) |
| Строка 1: |
Строка 1: |
| − | == Main Information ==
| + | Если вы наткнулись на эту страницу и можете ее удалить - сделайте это пожалуйста |
| − | | + | |
| − | '''Deadline:''' December 11th, 23:59 <br>
| + | |
| − | <br>
| + | |
| − | '''Submission formats:'''<br>
| + | |
| − | Jupyter Notebook, submit a link to the notebook file and the dataset file via this Google form: TODO <br>
| + | |
| − | Online page (details in the pilot section). Submit a link to your project via the same Google form: TODO <br>
| + | |
| − | <br>
| + | |
| − | '''Dataset selection:''' <br>
| + | |
| − | After you choose a dataset for the project, submit it to this form: TODO <br>
| + | |
| − | When selecting a dataset, check this table with the form responses to make sure nobody else has picked it: TODO <br>
| + | |
| − | | + | |
| − | == Project Structure ==
| + | |
| − | In the project you should get a dataset and perform some visualization and analysis on the data. It should be structured as a report, showing the steps you took and the information you got about the data.<br><br>
| + | |
| − | | + | |
| − | Your project should contain the following:<br>
| + | |
| − | # '''Descriptive statistics''' for at least '''3''' numerical fields. These statistics should include at least mean, median and standard deviation of the fields.<br>
| + | |
| − | # '''Data cleanup'''. Remove rows with NaN, check that columns have the correct data type. Other steps might be needed depending on data. If your data is already clean, show that it’s clean.<br>
| + | |
| − | # '''Plots''' for at least '''3''' numerical fields. Simple line plots, scatter plots, histograms, to give an idea of what the data looks like. Choose three different types of plots.<br>
| + | |
| − | # At least '''2''' plots or outputs for '''detailed overview'''. These should be presented in the form of comparisons – plotting several lines with different conditions on the same figure, printing statistics for subsets of data, plotting two graphs next to each other, etc.<br>
| + | |
| − | # A '''hypothesis check'''. Come up with a hypothesis about your data, then plot the relevant figure and/or print the relevant statistics. Check if your hypothesis was correct. The hypothesis should be more complex than comparing two subsets of data based on a single column. More details below.<br>
| + | |
| − | # '''Data transformation'''. Modify data from other columns to make new columns. If you haven't had to do data cleanup, add at least two new columns. Otherwise, you may add just one.<br>
| + | |
| − | # '''Discussion'''. During each step make a small write-up explaining what you do and why. You don’t need to include speculation on underlying causes of your results.<br>
| + | |
| − | # ('''''for pilot students''''') Provide a '''web interface''' to your project. Ideally, you should make your project accessible over the Internet. The whole project (plots, explanations, text results) should be available in the web interface. More details below.
| + | |
| − | | + | |
| − | == Grading Criteria ==
| + | |
| − | (still to be updated)
| + | |
| − | * Statistics - up to ?? points
| + | |
| − | * Cleanup and transformation - up to ?? points
| + | |
| − | * Simple plots - up to ?? points
| + | |
| − | * Detailed plots (comparison) - up to ?? points
| + | |
| − | * Hypothesis check - up to ?? points (including explanation and formulation)
| + | |
| − | * General write-up - up to ?? points
| + | |
| − | | + | |
| − | == Details ==
| + | |
| − | | + | |
| − | ==== Dataset ====
| + | |
| − | The easiest place to look for datasets is probably kaggle.com. Select a dataset with at least 3 numeric fields. Preferably, this should not include categorical fields, however this is not a strict requirement. For example “month”, “degree of education”, “rank in top 100” can be represented as ordered numbers, but these are categorical fields. “Age”, “cost”, “total distance in km”, “number of orders” are numeric fields.
| + | |
| − | <br><br>
| + | |
| − | | + | |
| − | ==== Data cleanup ====
| + | |
| − | Check some basic statistics of your data. It should tell you whether you need to do data cleanup. For example, if you have NaNs in a few rows, you should probably remove those rows. Sometimes data is given in an inconsistent way, like numbers written as “1M+”, “10k+”, “1k+”. Convert such cases to numbers you can work with: “1000000”, “10000”, “1000”. <br>
| + | |
| − | If you end up doing data cleanup, you will have to do less data transformation.
| + | |
| − | <br><br>
| + | |
| − | | + | |
| − | ==== Overview ====
| + | |
| − | Your notebook should follow a process of exploring the dataset. At first you should show some general ideas about what kind of data you’re dealing with. Outputting some statistics and making simple plots of all data are usually good ways of doing this.
| + | |
| − | <br><br>
| + | |
| − | | + | |
| − | ==== More detailed overview ====
| + | |
| − | Then you should provide insights into potentially interesting relations in your data. For example, if a plot shows some data columns are correlated, you can compute the correlation. If you have categorical data, you can try looking at different categories separately. For example, “what is the median salary for workers with different levels of education”. If several columns seem related to each other, this is also where you can add a third level axis as color and try to show the relationship of three variables at once.
| + | |
| − | <br><br>
| + | |
| − | | + | |
| − | ==== Hypothesis checking ====
| + | |
| − | At some points the insights you get become complicated. For example “What is the correlation between salary and number of sales for employees of different education levels? Do people at all levels of education get fairly compensated for increasing the number of sales?” Make a hypothesis in this format and test it by drawing figures or computing statistics. You don’t have to use actual statistical terms and calculations. If you get the result “there is a change, but it’s quite small, so it’s not clear if it’s significant or not'', it’s totally fine. <br><br>
| + | |
| − | Another example: “petrol cars with automatic transmission sold by a dealer have a higher price than the ones sold by individuals”. Here “petrol” and “automatic transmission” are conditions on top of “dealer/individual”. Then you can plot histograms of prices and print their mean and median to see if your hypothesis was correct.
| + | |
| − | <br><br>
| + | |
| − | | + | |
| − | ==== Hypothesis checking ====
| + | |
| − | At some points the insights you get become complicated. For example “What is the correlation between salary and number of sales for employees of different education levels? Do people at all levels of education get fairly compensated for increasing the number of sales?” Make a hypothesis in this format and test it by drawing figures or computing statistics. You don’t have to use actual statistical terms and calculations. If you get the result “there is a change, but it’s quite small, so it’s not clear if it’s significant or not'', it’s totally fine. <br><br>
| + | |
| − | Another example: “petrol cars with automatic transmission sold by a dealer have a higher price than the ones sold by individuals”. Here “petrol” and “automatic transmission” are conditions on top of “dealer/individual”. Then you can plot histograms of prices and print their mean and median to see if your hypothesis was correct.
| + | |
| − | <br><br>
| + | |
| − | | + | |
| − | ==== Data transformation ====
| + | |
| − | Add two new columns to the data frame and fill them with modified data from other columns. Some examples:
| + | |
| − | * “sales divided by salary” can be worker efficiency
| + | |
| − | * if you have both money and years, you can make money inflation-adjusted
| + | |
| − | * for data containing text data that can be ordered, you can convert it to numbers. For example, ratings in text form – “bad, average, good”, education level – “high school, bachelor’s degree, master’s degree, etc” can become “1, 2, 3, 4...”.
| + | |
| − | * You may also convert between units. Ounces to ml, kilometers to miles, Celsius to Fahrenheit, etc.
| + | |
| − | If you did something similar in data cleanup, add only one column here.
| + | |
| − | <br><br>
| + | |
| − | | + | |
| − | ==== Discussion ====
| + | |
| − | Include notes about your hypothesis and the information you can see in your data. Explain in words what can be seen in the images and statistics. You don’t need to speculate on the causes of the results you get, just explain the results themselves.
| + | |
| − | For example “Here is a scatter plot comparing car prices and the number of kilometers the cars have driven. There is a downward trend, meaning that as mileage increases, the price decreases.” <br><br>
| + | |
| − | Your project page is a report. The final version should contain only the relevant calculations and present them, so that it’s clear what kind of data you’re working with, what insights you’re showing, what are your hypotheses and test results.
| + | |
| − | <br><br>
| + | |
| − | | + | |
| − | ==== Web interface ====
| + | |
| − | Pilot students should set up a web interface to the project. Potential options are:
| + | |
| − | *Flask web server + HTML
| + | |
| − | *Flask web server + API (providing data by GET requests)
| + | |
| − | *Flask web server + Telegram bot
| + | |
| − | *Streamlit
| + | |
| − | If you want to use your own option, contact your instructors and make sure they approve.
| + | |
| − | The whole project should be available through the web interface. A Streamlit page should be equivalent to the Jupyter Notebook. The web pages should contain all the information you would otherwise put into a notebook. The telegram bot should contain a menu with several pages where you’re able to get any part of the project.
| + | |
| − | <br><br>
| + | |
Если вы наткнулись на эту страницу и можете ее удалить - сделайте это пожалуйста
