This notebook aims to cleans and analyzes data from our first mechanical turk user test.

This notebook uses data found here as a csv. Much more is needed to be done in order to streamline the cleaning and analysis of this data, but this notebook will serve as a kicking off point to which further investigations can be pursued.

import pandas as pd import numpy as np import plotly.graph_objects as go import plotly.express as px import json from collections import Counter pd.options.plotting.backend = "plotly" from IPython.display import HTML

HTML('''<script> function code_toggle(){ document.getElementsByClassName("css-1vxy2on")[0].style.backgroundColor = 'red'; console.log("it worked") } code_toggle() </script>''')

path = 'mkturk_results_100520.csv'

# from IPython.display import HTML # HTML('''<script> # let code_show=true; # function code_toggle() { # if (code_show){ # console.log("yay") # } else { # console.log("naw") # } # code_show = !code_show # } # </script> # <form action="javascript:code_toggle()"><input type="submit" value="Toggle"></form>''')

# from IPython.display import display, HTML # js = "<script>console.log('Hello World!');</script>" # display(HTML(js))

example

button = widgets.Button(description="Toggle Code") output = widgets.Output() display(button, output) def on_button_clicked(b): with output: code_toggle() print("Button clicked.") button.on_click(on_button_clicked)

Inspecting

task_data = pd.read_csv(path) task_data.columns

task_answers = task_data['Answer.taskAnswers'] task_answers

def build_df(path): '''Takes in path to csv file of mkturk results, seperates data into a list of individual series and returns a dataframe of just the user responses''' data = pd.read_csv(path) task_answers = data['Answer.taskAnswers'].values answer_dicts = [json.loads(task_answers[i].strip("[,]")) for i in range(len(task_answers))] cols = answer_dicts[0].keys() df = pd.DataFrame(columns=cols) for index, item in enumerate(answer_dicts): df.loc[len(df)] = answer_dicts[index] return df

task_answers_df = build_df(path) task_answers_df.head()

Our dataframe columns are now in the format Step # Slide #

task_answers_df.columns

task_answers_df[['s1s1','s1s2','s1s3','s1s4']].describe()

task_answers_df.info()

We're missing a `s4s4` column, which we should have, let's see why.

[items for items in task_answers_df['s4s3']][:1]

Looks like s4s3 contains titles from s4s3 and s4s4. We can unpack that later.

Cleaning

Step 1

Co-erce values from strings to floats

task_answers_df[['s1s1','s1s2','s1s3','s1s4']].astype('float',copy=False)

Steps 2-5

Remove slide 4 titles from s4s3 results
Build new Series for slide 4 results with new lists of lists that should not be in slide 3

Building new Series for responses to `s4s3`

s4s4 = task_answers_df['s4s3'].copy()

s4s4[0]

rm_from_slide_4 = ['episodes','offspring','grace-and-frankie','onde-day-at-at-time','fuller-house']

new_top_titles_s4s3 = [] for responses in s4s4: new_list = {key:val for key,val in responses.items() if key in rm_from_slide_4} new_top_titles_s4s3.append(new_list) task_answers_df['s4s3'] = pd.Series(new_top_titles_s4s3)

Building new Series for responses to `s4s4`

new_top_titles_s4s4 = [] for items in s4s4: new_list = {key:val for key,val in responses.items() if key not in rm_from_slide_4} new_top_titles_s4s4.append(new_list) task_answers_df['s4s4'] = pd.Series(new_top_titles_s4s4)

task_answers_df[['s4s2','s4s3','s4s4']].head()

def surface_selections(df,df_column): '''Takes in a pandas series and returns a new cleaned one removing false (non selected) items''' selected_titles = [] for all_titles in df[df_column].values: selections = [title for title, val in all_titles.items() if val == True] selected_titles.append(selections) # clean_series = pd.Series(selected_titles) clean_series = pd.Series(selected_titles).apply(lambda val: np.NaN if len(val)==0 else val) return clean_series

# columns we know have the format of {title: bool} we need to clean cols_to_clean = [ 's2s1', 's2s2', 's2s3', 's2s4', 's4s1', 's4s2', 's4s3', 's4s4', 's5s1', 's5s2', 's5s3', 's5s4' ]

for col in cols_to_clean: task_answers_df[col] = surface_selections(task_answers_df,col)

def get_toggle_val(df,df_column): '''Takes in a pandas series and returns a new cleaned one leaving only boolean state''' return df[df_column].apply(lambda val: val.get('on'))

task_answers_df['s3t1'] = get_toggle_val(task_answers_df,'s3t1')

task_answers_df.head()

Analysis

Step 1

On a scale between 1-5 how well do you think the carousel title represents the content inside it?

fig = go.Figure() fig.add_trace(go.Histogram(x=task_answers_df['s1s1'],name="Carousel 1")) fig.add_trace(go.Histogram(x=task_answers_df['s1s2'],name="Carousel 2")) fig.add_trace(go.Histogram(x=task_answers_df['s1s3'],name="Carousel 3")) fig.add_trace(go.Histogram(x=task_answers_df['s1s4'],name="Carousel 4")) fig.show()

Step 2

Select which TV Shows and Movies you would like to watch.

def get_top_responses(column): return [(key,val) for key, val in task_answers_df[column].value_counts().items()]

def plot_freq(top_responses, title): keys = pd.Series([title[0] for title in top_responses])[:10] count = pd.Series([title[1] for title in top_responses])[:10] freq_count = pd.DataFrame({'Title': keys,'Count':count}) fig = px.histogram(freq_count,x='Title', y ='Count', title=title) fig.update_xaxes(type='category') fig.show()

top_responses_s2s1 = get_top_responses('s2s1') plot_freq(top_responses_s2s1, "Top responses for Step 2 Slider 1")

top_responses_s2s2 = get_top_responses('s2s2') plot_freq(top_responses_s2s2, "Top responses for Step 2 Slider 2")

top_responses_s2s3 = get_top_responses('s2s3') plot_freq(top_responses_s2s3, "Top responses for Step 2 Slider 3")

top_responses_s2s4 = get_top_responses('s2s4') plot_freq(top_responses_s2s4, "Top responses for Step 2 Slider 4")

Top titles selected for step 2 by slider

def get_title_frequency(column): top_selections = [] for selection in task_answers_df[column].values: if type(selection) == list: for title in selection: top_selections.append(title) else: top_selections.append(selection) return pd.Series(top_selections).value_counts()

print("Step 2: Slide 1\n") print(get_title_frequency('s2s1'))

print("Step 2: Slide 2\n") print(get_title_frequency('s2s2'))

print("Step 2: Slide 3\n") print(get_title_frequency('s2s3'))

print("Step 2: Slide 4\n") print(get_title_frequency('s2s4'))

Step 4

Select what TV Shows and Movies you’ve already watched

Top group selections for Steps 4 by slider

top_responses_s4s1 = get_top_responses('s4s1') plot_freq(top_responses_s4s1, "Top responses for Step 4 Slider 1")

top_responses_s4s2 = get_top_responses('s4s2') plot_freq(top_responses_s4s2, "Top responses for Step 4 Slider 2")

top_responses_s4s3 = get_top_responses('s4s3') plot_freq(top_responses_s4s3, "Top responses for Step 4 Slider 3")

top_responses_s4s4 = get_top_responses('s4s4') plot_freq(top_responses_s4s4, "Top responses for Step 4 Slider 4")

print("Step 4: Slide 1\n") print(get_title_frequency('s4s1'))

print("Step 4: Slide 2\n") print(get_title_frequency('s4s2'))

print("Step 4: Slide 3\n") print(get_title_frequency('s4s3'))

print("Step 4: Slide 4\n") print(get_title_frequency('s4s4'))

Top group selections for Steps 5 by slider

Select what TV Shows and Movies you’ve already watched.

top_responses_s5s1 = get_top_responses('s5s1') plot_freq(top_responses_s5s1, "Top responses for Step 5 Slider 1")

top_responses_s5s2 = get_top_responses('s5s2') plot_freq(top_responses_s5s2, "Top responses for Step 5 Slider 2")

top_responses_s5s3 = get_top_responses('s5s3') plot_freq(top_responses_s5s3, "Top responses for Step 5 Slider 3")

top_responses_s5s4 = get_top_responses('s5s4') plot_freq(top_responses_s5s4, "Top responses for Step 5 Slider 4")

This notebook aims to cleans and analyzes data from our first mechanical turk user test.

example

Inspecting

We're missing a `s4s4` column, which we should have, let's see why.

Cleaning

Step 1

Steps 2-5

Building new Series for responses to `s4s3`

Building new Series for responses to `s4s4`

Analysis

Step 1

On a scale between 1-5 how well do you think the carousel title represents the content inside it?

Step 2

Select which TV Shows and Movies you would like to watch.

Top titles selected for step 2 by slider

Step 4

Select what TV Shows and Movies you’ve already watched

Top group selections for Steps 4 by slider

Top group selections for Steps 5 by slider

Select what TV Shows and Movies you’ve already watched.

Top titles selected for step 5 by slider

Recommended on Deepnote

Stock Market Analysis

The 10 Best Ways to Create NumPy Arrays

Wide Residual Networks

This notebook aims to cleans and analyzes data from our first mechanical turk user test.

example

Inspecting

We're missing a s4s4 column, which we should have, let's see why.

Cleaning

Step 1

Steps 2-5

Building new Series for responses to s4s3

Building new Series for responses to s4s4

Analysis

Step 1

On a scale between 1-5 how well do you think the carousel title represents the content inside it?

Step 2

Select which TV Shows and Movies you would like to watch.

Top titles selected for step 2 by slider

Step 4

Select what TV Shows and Movies you’ve already watched

Top group selections for Steps 4 by slider

Top group selections for Steps 5 by slider

Select what TV Shows and Movies you’ve already watched.

Top titles selected for step 5 by slider

Recommended on Deepnote

Stock Market Analysis

The 10 Best Ways to Create NumPy Arrays

Wide Residual Networks

We're missing a `s4s4` column, which we should have, let's see why.

Building new Series for responses to `s4s3`

Building new Series for responses to `s4s4`