Final Review

[recursory]

Summary

This project investigated Amazon reviews of cameras and attempted to make a predictive model of how many stars a review gave based on the number of "helpful" votes other users gave it. The report discovered that there wasn't a very clear relationship between these two metrics. The only explanation of the next step for this project is simply to look elsewhere for a better predictor of stars.

Data Preparation

• The first two tables, Review and Product, are made by splitting the initial csv file of camera sales by column. The new table, json_reviews, is made by organizing a dataset of one million reviews.
• Apart from removing the redundancy of the camera listings, no further tidying or explanation of tidy data is conducted for the first two sheets. The new spreadsheet was tidied by removing unnecessary columns and renaming the gross column names to much more readable ones.
• The Review and Product tables are assumed to be clean, without any demonstration. The json_reviews table is cleaned thoroughly in the course of part 2.

Modeling

• The predictive model that has been built is a predictor of the star rating of a review given the number of "helpful" votes other users have given it as a dependent variable.
• The portfolio accurately describes the purpose of the models with an explanation of how it works and what it could do if trained well.
• The explanation of the summary from the model seems to be interpreted accurately in the context of the data.

Validation

• The statistical analysis of the model in project 2 does a good job of explaining how well the relationship the model explores is established.
• The model is cross-validated correctly by training it on a set of the data and testing it on the remaining portion of the data.
• The cross-validation features an explanation about how the results indicate that the model isn't very effective. This makes sense in the context of the report and is articulated well.

R Proficiency

• R code use in this product is very good.
• It's clear that the author has experience with the language or experience finding good solutions to problems that have arisen.
• In addition, each of the code blocks have commentary explaining what is happening, so an R novice could learn through context what's happening.

Communication

• Extensive use of R markdown symbols makes the project very readable.
• The explanations of the procedures are thorough and easily understandable.
• The graphs in deliverable 2 effectively communicate the content of the datasets.
• Correlation is misspelled as "corolation" in the third deliverable.

Critical Thinking

The explanation of operationalization in the third deliverable does a good job of conveying how the project could be utilized productively, but lacks an explanation of potential consequences that could occur or variables that could affect the outcome of the project and implementation.

[NathanLovell]

Data Preparation and Modeling (17% out of 20%)

I agree with Olivia that I don't go into much detail about data preparation but that is because I believe my data was already pretty tidy. Apart from renaming and splitting my tables into more manageable chunks, I didn't feel much else was needed. My modeling was correctly done, even though no useful correlations could be made.

Validation and Operationalization (18% out of 20%)

I explained what the validation was doing and how companies could use this information. Didn't give myself perfect score because I didn't get any good data. All operationalization was speculated using my failed correlations.

R Proficiency (20% out of 20%)

All of my functions worked and I implemented my knowledge from the class correctly into my portfolio.

Communication (16% out of 20%)

While I didn't always go into the most detail, I tried to provide examples of everything I did. Some technical knowledge is expected in order to understand everything that I did in my portfolio.

Critical Thinking (10% out of 20%)

While I had everything that was needed, I did not go above and beyond to think critically. I could have come up with a better second source to try and make better correlations instead of more of basically the same data.