Tension Board 2 / Tension Board 1: Data Overview and Climbing Statistics

Purpose

This notebook establishes the basic statistical landscape of the dataset before we move into hold-level analysis and predictive modelling. The main goals are:

1. to understand the size and scope of the data,
2. to compare layouts, boards, and angles at a high level,
3. to identify broad trends in grade, popularity, and quality,
4. to create a clean descriptive baseline for the later modelling notebooks.

Throughout, I treat each climb-angle entry as a separate observation unless explicitly noted otherwise. That matters because some climbs appear at multiple angles, so a unique climb count and a climb-angle count are not always the same thing.

Outputs

This notebook produces summary tables and exploratory plots that motivate the later notebooks on:

• hold usage,
• hold difficulty,
• feature engineering,
• predictive modelling,
• and deep learning.

Notebook Structure

1. Setup and Imports
2. Popularity and Temporal Trends
3. Climbing Statistics
4. Prolific Statistics
5. Conclusion

Setup and Imports

"""
==================================
Setup and imports
==================================
"""
# Imports
import pandas as pd
import sqlite3
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np

import matplotlib.patches as mpatches
import sqlite3


# Set some display options
pd.set_option('display.max_columns', None)
pd.set_option('display.max_rows', 100)

# Set style
palette=['steelblue', 'coral', 'seagreen']  #(for multi-bar graphs)

# Connect to the database
DB_PATH="../data/tb2.db"
conn = sqlite3.connect(DB_PATH)

"""
==================================
### Query our data from the DB
==================================
"""

# Query climb data
climbs_query = """
SELECT
    c.uuid,
    c.name AS climb_name,
    c.setter_username,
    c.layout_id AS layout_id,
    c.description,
    c.is_nomatch,
    c.is_listed,
    l.name AS layout_name,
    p.name AS board_name,
    c.frames,
    cs.angle,
    cs.display_difficulty,
    dg.boulder_name AS boulder_grade,
    cs.ascensionist_count,
    cs.quality_average,
    cs.fa_at
    
FROM climbs c
JOIN layouts l ON c.layout_id = l.id
JOIN products p ON l.product_id = p.id
JOIN climb_stats cs ON c.uuid = cs.climb_uuid
JOIN difficulty_grades dg ON ROUND(cs.display_difficulty) = dg.difficulty
WHERE cs.display_difficulty IS NOT NULL;
"""

# Load it into a DataFrame
df = pd.read_sql_query(climbs_query, conn)

The above query will allow us to gather basically anything we need to in order to analyze climbing statistics. We leave out information about climging holds and things like this, because they will be analyzed in a different notebook. Let's see what our DataFrame looks like.

df

	uuid	climb_name	setter_username	layout_id	description	is_nomatch	is_listed	layout_name	board_name	frames	angle	display_difficulty	boulder_grade	ascensionist_count	quality_average	fa_at
0	0004edf6aeac9618d96a3b949cd9a724	Jumping Giraffe	david.p.kunz	9		0	1	Original Layout	Tension Board	p42r1p71r4p76r1p89r2p91r3p104r2p111r2p232r2	40	24.0000	7b/V8	2	3.00000	2020-03-23 23:52:37
1	00072fbd8c22711ef3532a5017c1a5c2	Albatross	free.and.independent	9	No Matching	1	1	Original Layout	Tension Board	p31r2p49r2p52r2p53r2p87r3p92r1p99r2p118r2p120r...	25	19.2500	6b+/V4	4	3.00000	2019-10-05 01:55:14
2	0008d8af4649234054bea434aaeabaab	crossup	judemandudeman	9		0	1	Original Layout	Tension Board	p22r2p58r2p76r1p83r3p166r2p228r2p280r4	45	20.0000	6c/V5	2	2.00000	2018-01-30 03:18:13
3	000eb831d3a1e92ea8fdec2518fd77d3	For the love of Tension	willrossiter	9	No matching	1	1	Original Layout	Tension Board	p86r3p95r1p131r2p151r2p171r1p173r2p187r2p261r4...	20	18.0000	6b/V4	1	3.00000	2019-03-15 15:46:06
4	000eb831d3a1e92ea8fdec2518fd77d3	For the love of Tension	willrossiter	9	No matching	1	1	Original Layout	Tension Board	p86r3p95r1p131r2p151r2p171r1p173r2p187r2p261r4...	40	23.0000	7a+/V7	1	3.00000	2021-06-27 22:41:10
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
147041	0B714869B80248D1B698E449EE885AB0	Loch Ness Monster	bennykuttler1285	10	No matching	1	1	Tension Board 2 Mirror	Tension Board 2	p333r6p344r6p367r8p379r8p380r5p500r8p570r6p579...	45	16.0000	6a/V3	3468	2.98991	2022-12-28 17:42:57
147042	71C4F8D564D045EFA1C9F26BB949E040	$2 lap dance 💃🕺	nelldell	10	No Match	1	1	Tension Board 2 Mirror	Tension Board 2	p311r6p320r5p366r8p372r7p445r5p462r6p468r8p570r6	40	20.0000	6c/V5	3388	2.98022	2022-12-07 00:30:56
147043	B998C0712A2240E8AC858CB72E9115D5	Aftermath	tensionclimbing	10		0	1	Tension Board 2 Mirror	Tension Board 2	p350r8p370r8p464r6p569r5p589r7p685r5p718r6p767r8	40	23.0000	7a+/V7	3	3.00000	2023-07-04 20:16:10
147044	C67E01D16E2940C2AFA76B436F9541D5	curly fries	sidpintobean	10	no matching	1	1	Tension Board 2 Mirror	Tension Board 2	p485r8p497r8p692r6p716r6p722r6p726r6p740r6p753...	40	20.1111	6c/V5	18	2.88889	2024-03-22 19:39:34
147045	397494C43D6A47A1BE9DB5FA7A9351A9	duck	trevordoesinfactclimb	10	No match	1	1	Tension Board 2 Mirror	Tension Board 2	p322r6p385r6p486r5p499r8p504r8p557r7p594r6	45	28.1000	8a/V11	10	3.00000	2025-01-16 17:54:22

147046 rows × 16 columns

---

Popularity and Temporal Trends

Popularity of Tension Board

Since we do not have access to user data, we will examine the popular of the Tension Boards by counting first ascents and unique setters by year. Often it's the case that the first ascensionist is the also the setter of the climb, but not always. None the less, we group up first ascensionists by year, with an extra tidbit about how many unique setters there were.

"""
==================================
Popular of tension board by year. 
First ascents by year + unique setters by year
==================================
"""

# Convert df['fa_at'] to datetime format. (For some reason, it does not register as such)
df['fa_at'] = pd.to_datetime(df['fa_at'])

# Add a new column for the year
df['fa_year'] = df['fa_at'].dt.year

# Make a new DataFrame with year, first_ascents, and unique_setters
df_growth = df.groupby('fa_year').agg(
    first_ascents=('uuid', 'count'),
    unique_setters=('setter_username', 'nunique')
).reset_index()

# Disregard the year 2026 since the data only goes one month in. 
df_growth = df_growth[df_growth['fa_year'] < 2026]

## Plot
# Dual index plotting

fig, ax1 = plt.subplots(figsize=(12,6))

# Bar chart for first ascents
ax1.bar(df_growth['fa_year'], df_growth['first_ascents'], label='First Ascents', color='coral')
ax1.set_xlabel('Year')
ax1.set_ylabel('First Ascents')
ax1.set_title('TB First Ascents & Unique Setters over Time')
#ax1.tick_params(axis='y')

# Line chart for unique setters (secondary axis)
ax2 = ax1.twinx()
ax2.plot(df_growth['fa_year'], df_growth['unique_setters'], color='steelblue', marker='o', label='Unique Setters')
ax2.set_ylabel('Unique Setters', color='steelblue')
ax2.tick_params(axis='y', labelcolor='steelblue')

# Other stuff
fig.legend(loc='upper left', bbox_to_anchor=(0.15,0.85))

plt.xticks()
plt.savefig('../images/01_climb_stats/first_ascents_by_year.png')
plt.show()

Seasonal analysis

Next, we examine when the Tension board is most popular. Again, we will work with what we have and use first ascent data. We will plot first ascents by month, combing all years. We exclude the year 2026 because this can skew the analysis as some of the month of January has data (and clearly, 2026 is when the TB2 is the most popular, so this can actually add quite a bit bias).

"""
==================================
Season analysis: first ascents by month
==================================
"""

# First let us add a column for the month to our data
df['fa_month'] = df['fa_at'].dt.month

# Filter to years < 2026 since the data only goes one month in
df_filter = df[df['fa_year'] < 2026]

# Make a new DataFrame with month and first ascents
df_season = df_filter.groupby('fa_month').agg(
    first_ascents=('uuid', 'count'),
).reset_index()

# We also add a column for the month name. 
month_names = ['Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 
               'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec']
df_season['fa_month_name'] = df_season['fa_month'].apply(lambda x: month_names[x-1])

# Plot the data
fig,ax = plt.subplots(figsize=(12,6))
ax.bar(df_season['fa_month_name'], df_season['first_ascents'], color='coral')
ax.set_title('First Ascents by Month (All Years Combined)')
ax.set_xlabel('Month')
ax.set_ylabel('Total First Ascents')

# Save the file
plt.savefig('../images/01_climb_stats/first_ascents_by_month.png')
plt.show()

This should be what we expect: that the winter months (Dec/Jan) see the most traffic. This is probably when the outdoor climbers are hitting the boards because they're stuck inside. The warmer months see the least number of first ascents since the strong climbers are probably outdoors.

Day of Week Analysis

We can plot the number of first ascents by day of week. Removing the 2026 data shouldn't make a difference here, so we opt to keep it.

"""
==================================
Day of Week analysis
==================================
"""

# Let us add a column in our DataFrame for the day of the week.
# Note that df.dt.day_of_week will have Monday be 0 and Sunday be 6. 

df['fa_day_of_week'] = df['fa_at'].dt.day_of_week


# Make a new DataFrame with month and first ascents
df_days = df.groupby('fa_day_of_week').agg(
    first_ascents=('uuid', 'count'),
).reset_index()

# We also add a column for the month name. 
day_names = ['Mon', 'Tues', 'Wed', 'Thurs', 'Fri', 'Sat', 'Sun']
df_days['fa_day_name'] = df_days['fa_day_of_week'].apply(lambda x: day_names[x])

# Plot the data
fig,ax = plt.subplots(figsize=(12,6))
ax.bar(df_days['fa_day_name'], df_days['first_ascents'], color='coral')
ax.set_title('First Ascents by Day of Week (All Years Combined)')
ax.set_xlabel('Day')
ax.set_ylabel('Total First Ascents')

# Save the file
plt.savefig('../images/01_climb_stats/first_ascents_by_day_of_week.png')
plt.show()

Interesting, Tuesday and Wednesday have the most traffic, while Monday is the least popular.

Time of Day Analysis

We can even do a time of day analysis. Again, we will keep the 2026 data since it shouldn't affect much. It is not entirely clear that makes sense to look at this, as we don't know if the time of first ascent is recorded in local time of the climber or local time of the server. These boards are all over the world, so this may add quite a bit of variance.

"""
==================================
Time of Day analysis
==================================
"""

# Let us add a column in our DataFrame for the day of the week.
# Note that df.dt.day_of_week will have Monday be 0 and Sunday be 6. 

df['fa_hour'] = df['fa_at'].dt.hour


# Make a new DataFrame with month and first ascents
df_hour = df.groupby('fa_hour').agg(
    first_ascents=('uuid', 'count'),
).reset_index()

# We also add a column for the month name. 
#df_time['fa_day_name'] = df_time['fa_day_of_week'].apply(lambda x: day_names[x])

# Plot the data
fig,ax = plt.subplots(figsize=(12,6))
ax.bar(df_hour['fa_hour'], df_hour['first_ascents'], color='coral')
ax.set_title('First Ascents by Hour (All Years Combined)')
ax.set_xlabel('Hour')
ax.set_ylabel('Total First Ascents')

# Save the file
plt.savefig('../images/01_climb_stats/first_ascents_by_hour.png')
plt.show()

---

Climbing Statistics: Grades, Angles, Quality, and Matching

We will visualize the climbing grade distribution. Recall that we have the following table of grades (with some other unlisted grades).

difficulty	boulder_name	route_name
10	4a/V0	5b/5.9
11	4b/V0	5c/5.10a
12	4c/V0	6a/5.10b
13	5a/V1	6a+/5.10c
14	5b/V1	6b/5.10d
15	5c/V2	6b+/5.11a
16	6a/V3	6c/5.11b
17	6a+/V3	6c+/5.11c
18	6b/V4	7a/5.11d
19	6b+/V4	7a+/5.12a
20	6c/V5	7b/5.12b
21	6c+/V5	7b+/5.12c
22	7a/V6	7c/5.12d
23	7a+/V7	7c+/5.13a
24	7b/V8	8a/5.13b
25	7b+/V8	8a+/5.13c
26	7c/V9	8b/5.13d
27	7c+/V10	8b+/5.14a
28	8a/V11	8c/5.14b
29	8a+/V12	8c+/5.14c
30	8b/V13	9a/5.14d
31	8b+/V14	9a+/5.15a
32	8c/V15	9b/5.15b
33	8c+/V16	9b+/5.15c

We will use the actual difficulty in our work, and then unpack translations into boulder_name as we see fit.

Grade distribution

"""
==================================
Difficulty distribution by layout (with total)
==================================
"""

grade_counts = df['boulder_grade'].value_counts()
grade_order = df.groupby('boulder_grade')['display_difficulty'].mean().sort_values().index.tolist()
grade_counts = grade_counts.reindex(grade_order)

# Prepare data in long format
df_long = df.groupby(['boulder_grade', 'layout_name']).size().reset_index(name='count')

# Calculate total for background
df_total = df.groupby('boulder_grade').size().reset_index(name='count')
df_total['layout_name'] = 'All Layouts'

# Reindex to correct grade order
df_long['grade_order'] = df_long['boulder_grade'].map(
    {g: i for i, g in enumerate(grade_order)}
)
df_long = df_long.sort_values('grade_order')

df_total['grade_order'] = df_total['boulder_grade'].map(
    {g: i for i, g in enumerate(grade_order)}
)
df_total = df_total.sort_values('grade_order')

# Plot
fig, ax = plt.subplots(figsize=(16, 8))

# Plot "All Layouts" behind (light gray)
sns.barplot(
    data=df_total,
    x='boulder_grade',
    y='count',
    color='lightgray',
    ax=ax,
    zorder=1,
    width=0.6,
    order=grade_order
)

# Plot individual layouts (grouped) in front
sns.barplot(
    data=df_long,
    x='boulder_grade',
    y='count',
    hue='layout_name',
    palette=['steelblue', 'coral', 'seagreen'],
    ax=ax,
    zorder=2,
    order=grade_order
)

# Create custom legend with "All Layouts" included
handles, labels = ax.get_legend_handles_labels()
all_layouts_patch = mpatches.Patch(color='lightgray', label='All Layouts')
handles.insert(0, all_layouts_patch)
ax.legend(handles=handles, title='Layout', fontsize=10)

ax.set_xlabel('Grade', fontsize=11)
ax.set_ylabel('Number of Climbs', fontsize=11)
ax.set_title('Difficulty Distribution by Board Layout', fontsize=14)
ax.tick_params(axis='x', rotation=45)
ax.grid(axis='y', alpha=0.3)

plt.tight_layout()
plt.savefig('../images/01_climb_stats/difficulty_distribution_by_layout_with_total.png', dpi=150, bbox_inches='tight')
plt.show()

As a climber in North America, I tend to just use the V-grade and not look at the French grade. So let us group the V-grades together and show the distribution like that. We'll usually just stick the boulder_grade (e.g., 5c/V2) instead of grouping the V-grades though.

"""
==================================
V-Grade distribution by layout (with total)
==================================
"""

# Let's add a v_grade column and v_grade_counts
df['v_grade'] = df['boulder_grade'].str.split('/').str[1]
v_grade_counts = df['v_grade'].value_counts()
v_grade_order = df.groupby('v_grade')['display_difficulty'].mean().sort_values().index.tolist()
v_grade_counts = grade_counts.reindex(v_grade_order)

# Prepare data in long format
df_long = df.groupby(['v_grade', 'layout_name']).size().reset_index(name='count')

# Calculate total for background
df_total = df.groupby('v_grade').size().reset_index(name='count')
df_total['layout_name'] = 'All Layouts'

# Reindex to correct grade order
df_long['v_grade_order'] = df_long['v_grade'].map(
    {g: i for i, g in enumerate(v_grade_order)}
)
df_long = df_long.sort_values('v_grade_order')

df_total['v_grade_order'] = df_total['v_grade'].map(
    {g: i for i, g in enumerate(v_grade_order)}
)
df_total = df_total.sort_values('v_grade_order')

# Plot
fig, ax = plt.subplots(figsize=(16, 8))

# Plot "All Layouts" behind (light gray)
sns.barplot(
    data=df_total,
    x='v_grade',
    y='count',
    color='lightgray',
    ax=ax,
    zorder=1,
    width=0.6,
    order=v_grade_order
)

# Plot individual layouts (grouped) in front
sns.barplot(
    data=df_long,
    x='v_grade',
    y='count',
    hue='layout_name',
    palette=palette,
    ax=ax,
    zorder=2,
    order=v_grade_order
)

# Create legend with "All Layouts" included
handles, labels = ax.get_legend_handles_labels()
all_layouts_patch = mpatches.Patch(color='lightgray', label='All Layouts')
handles.insert(0, all_layouts_patch)
ax.legend(handles=handles, title='Layout', fontsize=10)

ax.set_xlabel('Grade', fontsize=11)
ax.set_ylabel('Number of Climbs', fontsize=11)
ax.set_title('Difficulty Distribution by Board Layout', fontsize=14)
ax.tick_params(axis='x', rotation=45)
ax.grid(axis='y', alpha=0.3)

plt.tight_layout()
plt.savefig('../images/01_climb_stats/v_grade_distribution_by_layout_with_total.png', dpi=150, bbox_inches='tight')
plt.show()

So the grade distribution actually varies quite a bit from board to board. Some key differences in grades are the angle at which the climb is. Note that climbs can be done at different angles.

Angle Distribution

What about the angle distribution? Since the TB1 goes from 0 to 50 and the TB2 goes from 0 to 65 (although my local board only goes to 60?), let's do an analysis on each.

"""
==================================
Angle distribution
==================================
"""

# TB1 goes up to 50 degrees, TB2 up to 65. (Although my local TB2 only goes up to 60 -- brutal climbing)

# Prepare data in long format
df_long = df.groupby(['angle', 'layout_name']).size().reset_index(name='count')

# Calculate total for background
df_total = df.groupby('angle').size().reset_index(name='count')
df_total['layout_name'] = 'All Layouts'

# Reindex to correct order
angle_order = sorted(df['angle'].unique())

# Plot
fix, ax = plt.subplots(figsize=(16,8))

# Plot All Layouts
sns.barplot(
    data=df_total,
    x='angle',
    y='count',
    color='lightgray',
    ax=ax,
    zorder=1,
    width=0.6,
    order=angle_order
)

# Plt indivudual layouts
sns.barplot(
    data=df_long,
    x='angle',
    y='count',
    hue='layout_name',
    palette=palette,
    order=angle_order,
    ax=ax,
    zorder=2
)

handles,labels = ax.get_legend_handles_labels()
all_layouts_patch = mpatches.Patch(color='lightgray', label='All Layouts')
handles.insert(0, all_layouts_patch)
ax.legend(handles=handles, title='Layout')

ax.set_xlabel('Angle')
ax.set_ylabel('Number of Climbs')
ax.set_title('Angle Distribution by Board Layout')
ax.grid(axis='y', alpha=0.3)


plt.suptitle('Angle Distribution by Board Layout')
plt.savefig('../images/01_climb_stats/angle_distribution_by_layout.png')
plt.show()

We see that for all the boards, 40 degrees is the most common angle.

Angle vs grade

How is the distribution between angles and grades? Let's do this with a heatmap.

"""
==================================
Angle vs grade
==================================
"""

fig, ax = plt.subplots(figsize=(16, 8))

# Create mapping from difficulty to V-grade
grade_mapping = df.groupby('display_difficulty')['boulder_grade'].first().to_dict()

# Plot "All Layouts" as faint background boxes
sns.boxplot(
    data=df,
    x='angle',
    y='display_difficulty',
    color='lightgray',
    order=angle_order,
    showfliers=False,
    width=0.6,
    ax=ax,
    zorder=1
)

# Plot individual layouts in front
sns.boxplot(
    data=df,
    x='angle',
    y='display_difficulty',
    hue='layout_name',
    hue_order=['Original Layout', 'Tension Board 2 Mirror', 'Tension Board 2 Spray'], # For some reason this plot goes TB2 Spray / TB1 Orig / TB2 Mirror. Simple fix.
    palette=['steelblue', 'coral', 'seagreen'],
    order=angle_order,
    showfliers=False,
    ax=ax,
    width=0.5,
    zorder=2
)

# Relabel y-axis with boulder_grades
yticks_rounded = sorted(set(int(round(t)) for t in df['display_difficulty'].unique() if not pd.isna(t)))
ylabels = [grade_mapping.get(t, '') for t in yticks_rounded]
ax.set_yticks(yticks_rounded)
ax.set_yticklabels(ylabels)

# Custom legend with "All Layouts"
handles, labels = ax.get_legend_handles_labels()
all_patch = mpatches.Patch(color='lightgray', label='All Layouts')
handles.insert(0, all_patch)
ax.legend(handles=handles, title='Layout', fontsize=10)

ax.set_xlabel('Angle (degrees)', fontsize=11)
ax.set_ylabel('V-Grade', fontsize=11)
ax.set_title('Difficulty Distribution by Angle and Layout', fontsize=14)
ax.grid(axis='y', alpha=0.3)

plt.tight_layout()
plt.savefig('../images/01_climb_stats/difficulty_by_angle_boxplot_by_layout.png', dpi=150, bbox_inches='tight')
plt.show()

The Quality of a climb

Next we examine the quality of a climb. First we look at how quality relates to the number of ascents.

"""
==================================
Climb quality vs popularity
==================================
"""

# Filter to climbs with quality ratings
df_quality = df[(df['quality_average'].notna()) & (df['quality_average'] > 0)]

# Sample for performance
df_sample = df_quality.sample(min(2000, len(df_quality)))

g = sns.jointplot(
    data=df_sample,
    x='quality_average',
    y='ascensionist_count',
    kind='scatter',
    color='teal',
    height=5
)

g.ax_joint.set_xlabel('Quality Rating')
g.ax_joint.set_ylabel('Ascensionist Count')
g.fig.suptitle('Quality vs Popularity')

plt.savefig('../images/01_climb_stats/quality_popularity.png', dpi=150, bbox_inches='tight')
plt.show()

Next we visualize the average quality vs the angle and grade, by means of a heatmap. Keep in mind that the harder the climb and steeper the angle, the less people will be doing it. So harder climbs are skewed towards people who can actually do it. The point is that, on boards, the climb quality isn't always the best metric. As such, we won't spend too much time on the quality and will only do a heatmap which takes into account all layouts.

### Average quality by angle and grade

# Filter to climbs with quality ratings
df_quality = df[(df['quality_average'].notna()) & (df['quality_average'] > 0)]


# Create pivot table
quality_pivot = df_quality.pivot_table(
    index='boulder_grade',
    columns='angle',
    values='quality_average',
    aggfunc='mean'
)
quality_pivot = quality_pivot.reindex(grade_order)
quality_pivot = quality_pivot.reindex(columns=[a for a in angle_order if a in quality_pivot.columns])

# Plot
fig, ax = plt.subplots(figsize=(16, 8))

sns.heatmap(
    quality_pivot,
    cmap='RdYlGn',
    cbar_kws={'label': 'Avg Quality Rating'},
    ax=ax
)

ax.set_xlabel('Angle (°)')
ax.set_ylabel('Grade')
ax.invert_yaxis()
ax.set_title('Average Quality Rating by Grade and Angle (All Layouts)')

plt.tight_layout()
plt.savefig('../images/01_climb_stats/quality_heatmap_all_layouts.png', dpi=150, bbox_inches='tight')
plt.show()

"Match" vs. "No Match"

Some setters opt to put the "no match" tag onto their climbs. This means that the climber is not allowed to match their hands on any hold. Let's do an analysis of the differences with regular climbs.

"""
==================================
Match vs No Match analysis
==================================
"""

# Create status column (Match vs No Match)
df['status'] = df.apply(
    lambda x: 'No Match' if (
        pd.notna(x['description']) and 'No matching' in str(x['description'])
    ) or x.get('is_nomatch', 0) == 1 else 'Matched',
    axis=1
)

# Aggregate by layout and status
df_agg = df.groupby(['layout_name', 'status']).agg(
    count=('uuid', 'count'),
    avg_ascensionists=('ascensionist_count', 'mean'),
    avg_difficulty=('display_difficulty', 'mean')
).reset_index()

# Calculate All Layouts totals
df_all = df.groupby('status').agg(
    count=('uuid', 'count'),
    avg_ascensionists=('ascensionist_count', 'mean'),
    avg_difficulty=('display_difficulty', 'mean')
).reset_index()
df_all['layout_name'] = 'All Layouts'

# Combine
df_combined = pd.concat([df_agg, df_all], ignore_index=True)

# Order
status_order = ['Matched', 'No Match']


# Plot
fig, axes = plt.subplots(1, 3, figsize=(18, 6))

for ax, metric, title in zip(axes, ['count', 'avg_difficulty', 'avg_ascensionists'], 
                               ['Total Climbs', 'Average Difficulty', 'Avg Ascensionists']):
    
    # All Layouts as background (separate for each status)
    for i, status in enumerate(status_order):
        df_bg = df_combined[(df_combined['layout_name'] == 'All Layouts') & 
                            (df_combined['status'] == status)]
        
        if not df_bg.empty:
            # Position for status
            x_pos = i
            ax.bar(
                x_pos,
                df_bg[metric].values[0],
                width=0.7,
                color='lightgray',
                zorder=1
            )
    
    # Individual layouts in front
    df_plot = df_combined[df_combined['layout_name'] != 'All Layouts']
    
    sns.barplot(
        data=df_plot,
        x='status',
        y=metric,
        hue='layout_name',
        palette=palette,
        order=status_order,
        ax=ax,
        zorder=2
    )
    
    ax.set_title(title, fontsize=12)
    ax.set_xlabel('')
    ax.set_ylabel(title if metric == 'count' else ('Grade' if metric == 'avg_difficulty' else 'Count'), fontsize=11)
    ax.legend_.remove()
    ax.grid(axis='y', alpha=0.3)

# Y-axis labels for difficulty plot
yticks = [11, 13, 15, 17, 19, 21, 23]
ylabels = [grade_mapping.get(t, f"V{t-10}") for t in yticks]
axes[1].set_yticks(yticks)
axes[1].set_yticklabels(ylabels)
axes[1].set_ylim(bottom=10)

# Add V-grade annotations on difficulty bars
for i, status in enumerate(status_order):
    for j, layout in enumerate(df['layout_name'].unique()):
        row = df_combined[(df_combined['layout_name'] == layout) & (df_combined['status'] == status)]
        if not row.empty:
            diff = row['avg_difficulty'].values[0]
            v_grade = grade_mapping.get(round(diff), '')
            # Position: x = status index + offset for layout
            x_pos = i + (j - 1) * 0.27
            axes[1].text(x_pos, diff + 0.3, v_grade, ha='center', fontsize=8, fontweight='bold')


# Custom Legend
handles,labels = ax.get_legend_handles_labels()
all_layouts_patch = mpatches.Patch(color='lightgray', label='All Layouts')
handles.insert(0, all_layouts_patch)
fig.legend(handles=handles, title='Layout', bbox_to_anchor=(1.08, 0.9))

plt.suptitle('Match vs No Match Climbs by Layout', fontsize=14, y=1.02)
plt.tight_layout()
plt.savefig('../images/01_climb_stats/match_vs_nomatch_by_layout.png', dpi=150, bbox_inches='tight')
plt.show()

So we gather the following about "no match" climbs:

• they are far fewer than "match" climbs,
• they are on average harder than "match" climbs,
• and that they tend to have a bit more ascensionists on the TB2, and less on the TB1 (although, much more overall).

"""
==================================
Match vs No Match Summary
==================================
"""

summary = df_combined.pivot_table(
    index='layout_name',
    columns='status',
    values=['count', 'avg_difficulty', 'avg_ascensionists']
).round(2)

summary

	avg_ascensionists		avg_difficulty		count
status	Matched	No Match	Matched	No Match	Matched	No Match
layout_name
All Layouts	22.11	31.22	18.18	20.16	85202.0	61844.0
Original Layout	22.27	21.46	17.36	19.62	49093.0	27163.0
Tension Board 2 Mirror	19.97	44.92	19.21	20.52	21657.0	23329.0
Tension Board 2 Spray	24.81	26.42	19.44	20.72	14452.0	11352.0

---

Prolific statistics

Here we will take note of some prolific statistics: what are the most popular climbs and who are the most popular setters?

Most popular climbs

"""
==================================
Most popular climbs
==================================
"""

# The ascensionist_count column will allow us to easily deduce the top 15 climbs. 

# Create a DataFrame with the top 15 climbs
df_popular_climbs = df.sort_values(by='ascensionist_count', ascending=False).head(15)[::-1]

# We want the y-axis to say "Climb @ angle". Let's just create a new column for this. 
df_popular_climbs['y_label'] = df_popular_climbs.apply(
    lambda row: f"{row['climb_name']} @ {row['angle']}°", axis=1
)

# Plot it
fig, ax = plt.subplots(figsize=(16,8))
bars = ax.barh(df_popular_climbs['y_label'], df_popular_climbs['ascensionist_count'], color='teal')
ax.set_xlabel('Ascensionist Count')
ax.set_title('Top 15 Most Popular Climbs (at a specific angle)')

# Add grade and angle labels
for bar, (_, row) in zip(bars, df_popular_climbs.iterrows()):
    label = f"{row['boulder_grade']} on {row['layout_name']}"
    ax.text(100,
            bar.get_y() + bar.get_height()/2,
            label,
            va='center',
            ha='left',
            color='white')


plt.savefig('../images/01_climb_stats/top_15_climbs.png', bbox_inches='tight')
plt.show()

It's unsuprising that every one of these climbs is at 40° given that 40° is the most popular angle, by a long shot.

What about an angle-agnostic analysis? What are the top climbs amonst all angles?

"""
==================================
Top 15 most popular climbs (angle agnostic)
==================================
"""

# Aggregate by climb_name (sum counts across all angles)
df_agg = df.groupby(['climb_name', 'layout_name']).agg(
    total_ascensionists=('ascensionist_count', 'sum'),
    avg_difficulty=('display_difficulty', 'mean')
).reset_index()


# Sort and select top 15
df_popular_climbs_aa = df_agg.sort_values(by='total_ascensionists', ascending=False).head(15)

df_popular_climbs_aa

# Plot
fig, ax = plt.subplots(figsize=(16, 8))
bars = ax.barh(df_popular_climbs_aa['climb_name'], df_popular_climbs_aa['total_ascensionists'], color='teal')
ax.set_xlabel('Total Ascensionist Count (All Angles)')
ax.set_title('Top 15 Most Popular Climbs (Angle Agnostic)')
ax.invert_yaxis()

# Add grade labels inside bars
for bar, (_, row) in zip(bars, df_popular_climbs.iterrows()):
    # Create a label like "(Tension Board 2 Mirror)"
    label = f"{row['layout_name']}" 
    
    ax.text(
        100,  # Position inside bar
        bar.get_y() + bar.get_height() / 2,
        label,
        va='center',
        ha='left',
        color='white'
    )

plt.tight_layout()
plt.savefig('../images/01_climb_stats/top_15_climbs_angle_agnostic.png', bbox_inches='tight')
plt.show()

What about the top climbs (with/without angles) for each of the board layouts? Let's do with angles first.

"""
==================================
Top 15 most popular climbs by layout
==================================
"""

layouts = df['layout_name'].unique()

for layout in layouts:
    # Filter data for this layout
    df_layout = df[df['layout_name'] == layout]
    
    # Sort by popularity and take top 15
    df_top = df_layout.sort_values(by='ascensionist_count', ascending=False).head(15).reset_index(drop=True)
    
    # Select desired columns
    df_display = df_top[['climb_name', 'angle', 'boulder_grade', 'ascensionist_count']].copy()
    
    # Rename columns for display
    df_display.columns = ['Name', 'Angle', 'Grade', 'Ascensionists']
    
    # Format angle as string with degree symbol
    df_display['Angle'] = df_display['Angle'].astype(int).astype(str) + '°'
    
    # Reset index to show rank 1-15
    df_display.index = df_display.index + 1
    
    print(f"\n{layout}\n")
    display(df_display)

Original Layout

	Name	Angle	Grade	Ascensionists
1	Pre-game	40°	4a/V0	10378
2	Laser Camera	40°	6a/V3	8457
3	lefty loosy	40°	5a/V1	7248
4	Riopio	40°	4a/V0	6961
5	Sea Freight	40°	5c/V2	6953
6	Pre-game	30°	4a/V0	6863
7	Bubbles	40°	6b+/V4	6653
8	Laser Camera	30°	5a/V1	6409
9	Holler	40°	6a/V3	6288
10	Jiggles	40°	6b/V4	5568
11	Shimmy	40°	6c/V5	5489
12	Bubbles	30°	6a/V3	5473
13	Gunter	40°	4a/V0	5461
14	It's Alive!	30°	5c/V2	5357
15	It's Alive!	40°	5b/V1	5301

Tension Board 2 Mirror

	Name	Angle	Grade	Ascensionists
1	Masquerade	40°	6a/V3	12123
2	Drain from the Brain	40°	6c/V5	10242
3	Compliments To The Climber	40°	6b/V4	10204
4	Putty	40°	5c/V2	10162
5	Propagation	40°	6b/V4	9640
6	Thunderstruck	40°	6b/V4	9352
7	All Plastic	40°	5a/V1	9052
8	You Look Great Today	40°	6a/V3	8881
9	love is an open door.	40°	6c/V5	8693
10	Sherman	40°	6a/V3	8212
11	hop scotch	40°	6b/V4	7936
12	Dear John	40°	5c/V2	7821
13	Dancing in the Moonlight	40°	6b/V4	7109
14	Doomscroll	40°	4c/V0	7107
15	Poseidon	40°	6c/V5	7078

Tension Board 2 Spray

	Name	Angle	Grade	Ascensionists
1	Pizza Box	40°	6a/V3	6649
2	Aw, shoot	40°	6b/V4	5453
3	Write This Way	40°	6a/V3	5145
4	Shoulder Rust	40°	5a/V1	5018
5	Nacho Mango	40°	5c/V2	4841
6	Frictionless	40°	6c/V5	4536
7	Put It Up	40°	6b/V4	4499
8	Pour Cece	40°	6a/V3	4225
9	Authorized	40°	6c/V5	4207
10	It’s A Mid-West Alcohol	40°	6b/V4	4081
11	Center Left	40°	6c/V5	4008
12	Lost Glove	40°	6b/V4	3563
13	Perfect Pizza Party	40°	6a/V3	3420
14	Lightyears	40°	5c/V2	3379
15	The Where	40°	6b/V4	3310

On the TB2, it looks like 40 degrees constantly wins out. It is cool to see that, on the TB1, at least one climb made it to the top 15 twice, at two different angles! Congrats "It's Alive."

Now let us do a per-board angle agnostic analysis.

"""
==================================
Top 15 climbs by layout (angle agnostic)
==================================
"""

layouts = df['layout_name'].unique()

# Aggregate counts and collect angles
df_agg = df.groupby(['climb_name', 'layout_name']).agg(
    total_ascensionists=('ascensionist_count', 'sum')
).reset_index()

for layout in layouts:
    df_layout = df_agg[df_agg['layout_name'] == layout]
    df_top = df_layout.sort_values(by='total_ascensionists', ascending=False).head(15).reset_index(drop=True)
    
    df_display = df_top[['climb_name', 'total_ascensionists']].copy()
    df_display.columns = ['Name', 'Total Ascensionists']
    
    # Appropriate index
    df_display.index = df_display.index + 1
    
    print(f"\n### {layout}\n")
    display(df_display)

### Original Layout

	Name	Total Ascensionists
1	Pre-game	27468
2	Laser Camera	20968
3	Bubbles	16161
4	Sea Freight	15264
5	Holler	14275
6	It's Alive!	13545
7	Dull Scissors	11488
8	lefty loosy	11275
9	Getting By	11116
10	Go Figure	10483
11	Intro To Power	10377
12	Switching Places	9748
13	Shimmy	9427
14	Foles Gold	9339
15	Big Pinch Pinchin'	8823

### Tension Board 2 Mirror

	Name	Total Ascensionists
1	Masquerade	19655
2	Putty	19445
3	Compliments To The Climber	16844
4	Thunderstruck	16121
5	Further	15981
6	Propagation	15606
7	Poseidon	15549
8	All Plastic	15135
9	You Look Great Today	14607
10	Doomscroll	14150
11	Drain from the Brain	12910
12	Prime	11384
13	Endearing	10946
14	Dancing in the Moonlight	10855
15	Guided By Angels	10757

### Tension Board 2 Spray

	Name	Total Ascensionists
1	Pizza Box	11884
2	Nacho Mango	10325
3	Write This Way	9299
4	Shoulder Rust	8749
5	Marshmallow Dragon	8001
6	Authorized	7387
7	The Where	7325
8	Aw, shoot	7006
9	Perfect Pizza Party	6575
10	Frictionless	6536
11	New Gold	6529
12	Pour Cece	6022
13	Put It Up	6012
14	It’s A Mid-West Alcohol	5863
15	Bring The Ruckus	5689

Prolific setters

Next, we will make a simple table of the most prolific setters by board.

"""
==================================
Top 10 setters by layout
==================================
"""

# Make a DataFrame for the setters
df_setters = df.groupby(['setter_username', 'layout_name']).agg(
    climb_count=('uuid', 'nunique')
).reset_index()

layouts = df['layout_name'].unique()

for layout in layouts:
    df_layout = df_setters[df_setters['layout_name'] == layout]
    df_top = df_layout.sort_values(by='climb_count', ascending=False).head(10).reset_index(drop=True)

    df_display = df_top[['setter_username', 'climb_count']].copy()
    df_display.columns = ['Username', 'Climbs']

    # Reset index to show rank
    df_display.index = df_display.index + 1

    # Display
    print(f"\n{layout}\n")
    display(df_display)
    

Original Layout

	Username	Climbs
1	adamf1234	466
2	jonlackman	383
3	willanglin	332
4	kylejosephharding	328
5	str8_crimpin	247
6	tmon	246
7	gibbs	240
8	topoutclimbinggym	229
9	senderone	227
10	memphisben	198

Tension Board 2 Mirror

	Username	Climbs
1	tensionclimbing	513
2	SocksonBlocs	372
3	limberlimb	353
4	jaketiger111	320
5	mo3_3az	307
6	lijahl	306
7	iansutherland	218
8	theruz	203
9	AlexK	203
10	nicholson.brendan	174

Tension Board 2 Spray

	Username	Climbs
1	MaxClark	246
2	ianmek	217
3	tensionclimbing	175
4	Jeremy_Fullerton	171
5	danielwoodseyetattoo	148
6	Fatbeninco	136
7	rprops	130
8	SPIGGOTT	125
9	milo_forbes	118
10	cotton125	116

---

Conclusion

At this point we have a board-level and climb-level picture of the dataset. In particular, we now know:

• how large the dataset is,
• how the grade and angle distributions vary across layouts,
• which climbs and setters appear most often,
• and where simple descriptive trends begin to show up.

That gives us enough context to move from global statistics to hold-level structure. The next notebook focuses on hold usage patterns and board heatmaps, where we stop asking only how many climbs there are and start asking which physical parts of the board are driving those climbs.