def picks_waitlist_user_heatmap(users: pd.DataFrame):
init_plt()
heatmap = users.groupby([UserColumns.LIFETIME_ORDERS_BUCKET, UserColumns.LIFETIME_AOV_BUCKET]).sum()[
UserColumns.IS_PICKS_WAITLIST
].unstack(
).fillna(
0
).astype(
int
)
sns.heatmap(heatmap, cmap=sns.light_palette(Colors.PINK_DARK), annot=True, fmt=",")
plt.gcf().axes[0].invert_yaxis()
plt.gcf().suptitle("Primary Picks Waitlist Users")
plt.xlabel("User AOV")
plt.ylabel("User Lifetime Orders")
plt.show()
heatmap = users.groupby([UserColumns.ORDERS_PER_QUARTER_BUCKET, UserColumns.LIFETIME_AOV_BUCKET]).sum()[
UserColumns.IS_PICKS_WAITLIST
].unstack(
).fillna(
0
).astype(
int
)
sns.heatmap(heatmap, cmap=sns.light_palette(Colors.PINK_DARK), annot=True, fmt=",")
plt.gcf().axes[0].invert_yaxis()
plt.gcf().suptitle("Primary Picks Waitlist Users")
plt.xlabel("AOV")
plt.ylabel("Orders per Quarter")
plt.show()
def quality_onetimers(users: pd.DataFrame):
init_plt()
candidates = users[
(users[UserColumns.LIFETIME_ORDERS] < 2) &
(users[UserColumns.LIFETIME_GPR] > 90.0)
].copy(
)
candidates["month"] = candidates[UserColumns.FIRST_ORDER_MONTH].apply(
lambda d: d.month
)
sns.lineplot(
x=UserColumns.FIRST_ORDER_MONTH,
y=UserColumns.FIRST_ORDER_MONTH,
estimator= lambda x: len(x),
data=users,
)
plt.gcf().suptitle("Quality Onetimers by Date")
plt.xlabel("First Order Date")
plt.ylabel("Users")
plt.show()
grouped = candidates.groupby("month").agg({"email": "count"}).rename(
columns={"email": "users"}
)
grouped["percentage"] = grouped["users"] / len(candidates) * 100.0
grouped.plot(y="percentage", kind="bar", color=Colors.GREEN_LIGHT)
plt.gcf().suptitle("Quality Onetimers by Month")
plt.xlabel("First Order Month")
plt.ylabel("% of Users")
plt.show()
def order_frequency_over_time(users: pd.DataFrame):
init_plt()
candidates = users[
(users[UserColumns.LIFETIME_ORDERS] > 1) &
(~ users[UserColumns.AVG_DAYS_BETWEEN_ORDERS].isna()) &
(~ users[UserColumns.AVG_DAYS_BETWEEN_VISITS].isna())
]
def marketing_opt_out(users: pd.DataFrame):
init_plt()
heatmap = users.groupby([UserColumns.LIFETIME_ORDERS_BUCKET, UserColumns.LIFETIME_AOV_BUCKET]).sum()[
UserColumns.IS_MARKETING_OPT_OUT
].unstack(
).fillna(
0
).astype(
int
)
sns.heatmap(heatmap, cmap=sns.light_palette(Colors.PINK_DARK), annot=True, fmt=",")
plt.gcf().axes[0].invert_yaxis()
plt.gcf().suptitle("Marketing Opt-Out Users")
plt.xlabel("AOV")
plt.ylabel("Lifetime Orders")
plt.show()
total_users_heatmap = users.groupby(
[UserColumns.LIFETIME_ORDERS_BUCKET, UserColumns.LIFETIME_AOV_BUCKET]
).count(
)[UserColumns.EMAIL].unstack(
).fillna(
0
).astype(
int
)
percentage_map = heatmap / total_users_heatmap
sns.heatmap(
percentage_map,
cmap=sns.light_palette(Colors.PINK_DARK),
annot=True,
fmt=".0%"
)
plt.gcf().axes[0].invert_yaxis()
plt.gcf().suptitle("Marketing Opt-Out Users")
plt.xlabel("AOV")
plt.ylabel("Lifetime Orders")
plt.show()
heatmap = users.groupby(
[UserColumns.ORDERS_PER_QUARTER_BUCKET, UserColumns.LIFETIME_AOV_BUCKET]
).sum()[
UserColumns.IS_MARKETING_OPT_OUT
].unstack(
).fillna(
0
).astype(
int
)
sns.heatmap(heatmap, cmap=sns.light_palette(Colors.PINK_DARK), annot=True, fmt=",")
plt.gcf().axes[0].invert_yaxis()
plt.gcf().suptitle("Marketing Opt-Out Users")
plt.xlabel("AOV")
plt.ylabel("Orders per Quarter")
plt.show()
def products_per_bag(start: date = date(2018, 12, 1), end: date = date(2019, 2, 25)):
init_plt()
user_date_stats = sql_to_df('views_per_bag.sql', start=start, end=end)
user_date_stats["bag_indicator"] = user_date_stats["bag_indicator"].astype(float)
sns.lineplot(x="product", y="bag_indicator", data=user_date_stats)
plt.gcf().suptitle("Product Views vs Bags")
plt.xlabel("Product Views")
plt.ylabel("Likelihood of Bagging")
plt.xlim(0, 25)
plt.show()
def orders_per_year(users: pd.DataFrame):
init_plt()
sns.distplot(users[UserColumns.ORDERS_PER_YEAR], kde=False)
plt.gcf().suptitle("Orders per Year")
plt.xlim(0, 10)
plt.ylabel("Users")
plt.xlabel("Orders per Year")
plt.show()
display(users[UserColumns.ORDERS_PER_YEAR].describe(
percentiles=[.1, .2, .3, .4, .5, .6, .7, .8, .9, .95]))
def user_paths(
users: pd.DataFrame,
start: date = date(2018, 1, 1),
end: date = date(2019, 1, 1),
event_names: List[str] = [Events.PDP, Events.BAG, Events.CHECKOUT],
palette: Dict[str, str] = {
Events.PDP: Colors.PINK_MEDIUM,
Events.BAG: Colors.YELLOW_LIGHT,
Events.CHECKOUT: Colors.GREEN_MEDIUM,
},
x_jitter: float = None,
y_jitter: float = None,
title: str = "User Journeys",
):
init_plt()
events = get_events(list(users[EventColumns.EMAIL].values),
start=start,
end=end)
users["user_index"] = range(1, len(users) + 1)
events = events.join(
users.set_index(UserColumns.EMAIL)[["user_index"]],
on=UserColumns.EMAIL,
how="inner",
)
events = events[events[EventColumns.NAME].isin(event_names)]
events["epoch"] = events[EventColumns.TIME].apply(lambda d: d.timestamp())
if x_jitter:
events = _jitter(events, "epoch", x_jitter)
if y_jitter:
events = _jitter(events, "user_index", y_jitter)
fig, ax = plt.subplots(figsize=(15, 10))
sns.scatterplot(
y="user_index",
x="epoch",
hue=EventColumns.NAME,
palette=palette,
hue_order=event_names,
style=EventColumns.PLATFORM,
ax=ax,
data=events,
s=75,
)
plt.gcf().suptitle(title)
plt.ylabel("Individual User")
plt.xlabel("Date")
steps = pd.date_range(start, end, freq='MS').to_pydatetime()
plt.xticks(
[step.timestamp() for step in steps],
[step.date().isoformat() for step in steps],
rotation="vertical",
)
plt.legend(loc='center right', bbox_to_anchor=(1.25, 0.5), ncol=1)
plt.yticks(list(range(1, len(users) + 1)))
plt.show()
def common_products(transactions: pd.DataFrame):
init_plt()
candidates = transactions[transactions[TransactionColumns.ORDER_INDEX] > 1.0][
[
TransactionColumns.ORDER_ID,
TransactionColumns.ORDER_INDEX,
TransactionColumns.COMMON_PRODUCTS_RATIO,
]
].copy(
)
candidates["order_number"] = pd.cut(
candidates[TransactionColumns.ORDER_INDEX],
bins=[2, 3, 4, 5, 100],
labels=["2", "3", "4", "5+"],
right=False,
include_lowest=True,
)
candidates["product_ratio_bucket"] = pd.cut(
candidates[TransactionColumns.COMMON_PRODUCTS_RATIO],
bins=list(range(0, 91, 10)) + [101],
labels=[
"0-9",
"10-19",
"20-29",
"30-39",
"40-49",
"50-59",
"60-69",
"70-79",
"80-89",
"90-100",
],
right=False,
include_lowest=True,
)
order_buckets = candidates.groupby("order_number").size().rename(
"order_number_total"
)
grouped = candidates.groupby(["order_number", "product_ratio_bucket"]).size(
).rename(
"transactions"
)
grouped = grouped.to_frame().join(order_buckets, on="order_number").reset_index()
grouped["percentage"] = grouped["transactions"] / grouped[
"order_number_total"
] * 100.0
sns.barplot(
x="product_ratio_bucket", y="percentage", hue="order_number", data=grouped
)
plt.ylabel("% of Orders")
plt.xlabel("% of Previously Purchased Styles per Order")
plt.xticks(rotation="vertical")
plt.gcf().suptitle("Previously Purchased Styles in Repeat Orders")
plt.show()
def sales(transactions: pd.DataFrame):
init_plt()
sale_transactions = transactions[transactions[TransactionColumns.HAS_SALE_ITEM] > 0]
sale_transactions.groupby(TransactionColumns.ORDER_MONTH).size().rename(
"orders"
).to_frame(
).plot(
y="orders", kind="bar"
)
plt.gcf().suptitle("Sale Orders by Date")
plt.ylabel("Sale Orders")
plt.xlabel("Date")
plt.show()
users = transactions.groupby(TransactionColumns.EMAIL).agg(
{
TransactionColumns.HAS_SALE_ITEM: "sum",
TransactionColumns.ORDER_ID: "count",
TransactionColumns.ITEM_TOTAL: "mean",
}
).rename(
columns={
TransactionColumns.HAS_SALE_ITEM: "sale_orders",
TransactionColumns.ORDER_ID: "orders",
TransactionColumns.ITEM_TOTAL: "aov",
}
)
users["order_count_bucket"] = pd.cut(
users["orders"],
bins=[1, 2, 3, 4, 5, 6, 11, users["orders"].max() + 1],
labels=["1", "2", "3", "4", "5", "6-10", "11+"],
right=False,
)
users["aov_bucket"] = pd.cut(
users["aov"],
bins=[0, 50, 101, users["aov"].max() + 1],
labels=["<50", "50-100", "100+"],
right=False,
)
heatmap = users.groupby(["order_count_bucket", "aov_bucket"]).sum()[
"sale_orders"
].unstack(
).fillna(
0
).astype(
int
)
sns.heatmap(heatmap, cmap=sns.light_palette(Colors.PINK_DARK), annot=True, fmt=",")
plt.gcf().axes[0].invert_yaxis()
plt.gcf().suptitle("Orders w/ a Sale Item")
plt.xlabel("User AOV")
plt.ylabel("User Lifetime Orders")
plt.show()
def kids_rate_heatmap(users: pd.DataFrame):
init_plt()
heatmap = users.groupby([UserColumns.LIFETIME_ORDERS_BUCKET, UserColumns.LIFETIME_AOV_BUCKET]).mean()[
UserColumns.FIRST_ORDER_HAS_KIDS
].unstack(
).fillna(
0
)
sns.heatmap(heatmap, cmap=sns.light_palette(Colors.PINK_DARK), annot=True, fmt=".0%")
plt.gcf().axes[0].invert_yaxis()
plt.gcf().suptitle("Users with Kids in First Order")
plt.xlabel("AOV")
plt.ylabel("Lifetime Orders")
plt.show()
return heatmap
def subscription_op_size(users: pd.DataFrame, margin=0.65, shipping=6.0):
init_plt()
sorted = users.sort_values(UserColumns.ORDERS_PER_YEAR, ascending=False)[[
UserColumns.ORDERS_PER_YEAR, UserColumns.LIFETIME_AOV
]].copy()
sorted["one"] = 1.0
def new_subscription_orders(orders_per_year: int):
return max(0, 4.0 - math.floor(orders_per_year))
sorted["new_orders"] = sorted["orders_per_year"].apply(
new_subscription_orders)
sorted["new_revenue"] = sorted["new_orders"] * 68
sorted["new_profit"] = sorted["new_orders"] * 23.10
sorted["cumulative_new_revenue"] = sorted["new_revenue"].cumsum()
sorted["cumulative_new_profit"] = sorted["new_profit"].cumsum()
sorted["cumulative_users"] = sorted["one"].cumsum()
sns.lineplot(x="cumulative_users", y="cumulative_new_revenue", data=sorted)
plt.xlim(0, 50000)
plt.ylim(0, 2000000)
plt.gcf().suptitle("Additional Yearly Revenue based on Subscription Users")
plt.xlabel("Subscription Users")
plt.ylabel("Additional Yearly Revenue")
plt.show()
sns.lineplot(x="cumulative_users", y="cumulative_new_profit", data=sorted)
plt.xlim(0, 50000)
plt.ylim(0, 600000)
plt.gcf().suptitle("Additional Yearly Profit based on Subscription Users")
plt.xlabel("Subscription Users")
plt.ylabel("Additional Yearly Profit")
plt.show()
percents = list(range(1, 51))
opp_size = pd.DataFrame({
"coverage_percentage":
percents,
"cumulative_new_revenue":
np.percentile(sorted["cumulative_new_revenue"], percents),
"cumulative_new_profit":
np.percentile(sorted["cumulative_new_profit"], percents),
"cumulative_users":
np.percentile(sorted["cumulative_users"], percents)
})
display(opp_size)
def units_per_order(transactions: pd.DataFrame):
init_plt()
cols = TransactionColumns
units = cols.UNITS
repeats = transactions[transactions[cols.IS_REPEAT] > 0.0]
sns.distplot(transactions[units], kde=False)
plt.gcf().suptitle("Units per Order")
plt.xlabel("Units per Order")
plt.ylabel("Orders")
plt.show()
sns.distplot(repeats[units], kde=False)
plt.gcf().suptitle("Units per Repeat Order")
plt.xlabel("Units per Order")
plt.ylabel("Repeat Orders")
plt.show()
sns.lineplot(x=cols.ORDER_DATE, y=cols.UNITS, hue=cols.IS_REPEAT, data=transactions)
plt.ylim(0, 10)
plt.legend(plt.legend().get_patches(), ["first_order", "repeat_order"])
plt.gcf().suptitle("Units per Order")
plt.xlabel("Date")
plt.ylabel("Average Units Per Order")
plt.show()
def guest_users(transactions: pd.DataFrame):
init_plt()
is_guest = TransactionColumns.IS_GUEST
users = transactions.groupby(TransactionColumns.EMAIL).agg(
{
is_guest: "max",
TransactionColumns.ORDER_ID: "count",
TransactionColumns.ITEM_TOTAL: "mean",
}
).rename(
columns={
TransactionColumns.ORDER_ID: "orders", TransactionColumns.ITEM_TOTAL: "aov"
}
)
users["order_count_bucket"] = pd.cut(
users["orders"],
bins=[1, 2, 3, 4, 5, 6, 11, users["orders"].max() + 1],
labels=["1", "2", "3", "4", "5", "6-10", "11+"],
right=False,
)
users["aov_bucket"] = pd.cut(
users["aov"],
bins=[0, 50, 101, users["aov"].max() + 1],
labels=["<50", "50-100", "100+"],
right=False,
)
guest_group = users.groupby(TransactionColumns.IS_GUEST).size().rename(
"is_guest_transactions"
)
grouped = users.groupby(["order_count_bucket", is_guest]).size().rename(
"transactions"
).to_frame(
).reset_index(
)
grouped = grouped.join(guest_group, on=is_guest)
grouped["percentage"] = grouped["transactions"] / grouped[
"is_guest_transactions"
] * 100.0
sns.barplot(
x="order_count_bucket", y="percentage", hue=is_guest, data=grouped, ci=False
)
plt.gcf().suptitle("Lifetime Orders by Signed-In vs Guest")
plt.xlabel("Lifetime Orders")
plt.ylabel("% of Users")
plt.legend(plt.legend().get_patches(), ["signed_in", "guest"])
plt.show()
guest_counts = users[users[TransactionColumns.IS_GUEST] > 0.0].groupby(
["aov_bucket", "order_count_bucket"]
).size(
).rename(
"guest_users"
).to_frame(
)
matrix = users.groupby(["aov_bucket", "order_count_bucket"]).size().rename(
"users"
).to_frame(
)
matrix = matrix.join(guest_counts, how="left").fillna(0.0)
matrix["guest_percentage"] = matrix["guest_users"] / matrix["users"]
graph = matrix["guest_percentage"].unstack().T
display(graph)
sns.heatmap(graph, cmap=sns.light_palette(Colors.PINK_DARK), annot=True)
plt.gcf().axes[0].invert_yaxis()
plt.gcf().suptitle("Guest User Penetration by AOV and Lifetime Orders")
plt.xlabel("AOV")
plt.ylabel("Lifetime Orders")
plt.show()
def order_frequency(users: pd.DataFrame):
init_plt()
sns.lineplot(
x=UserColumns.FIRST_ORDER_MONTH,
y=UserColumns.FIRST_ORDER_MONTH,
estimator= lambda x: len(x),
data=users,
)
plt.gcf().suptitle("Users Acquired by Date")
plt.xlabel("First Order Date")
plt.ylabel("Users")
plt.show()
sns.distplot(users[UserColumns.LIFETIME_ORDERS], kde=False)
plt.gcf().suptitle("Lifetime Orders")
plt.xlabel("Lifetime Orders")
plt.ylabel("Users")
plt.show()
candidates = users[
(users[UserColumns.LIFETIME_ORDERS] > 1) &
(~ users[UserColumns.AVG_DAYS_BETWEEN_ORDERS].isna()) &
(~ users[UserColumns.AVG_DAYS_BETWEEN_VISITS].isna())
].copy(
)
sns.distplot(candidates[UserColumns.LIFETIME_ORDERS], kde=False)
plt.gcf().suptitle("Lifetime Orders > 1")
plt.xlabel("Lifetime Orders")
plt.ylabel("Users")
plt.show()
sns.distplot(
users[UserColumns.LIFETIME_ORDERS],
bins=[0, 1, 2, 3, 5, 10, 15, 20, 30, 100],
hist_kws=dict(cumulative=True),
)
plt.gcf().suptitle("Lifetime Orders CDF < 20")
plt.xlabel("Lifetime Orders")
plt.ylabel("CDF")
plt.xticks([1, 2, 3, 5, 10, 15, 20, 30, 100])
plt.xlim(0, 20)
plt.show()
sns.distplot(candidates[UserColumns.AVG_DAYS_BETWEEN_ORDERS], kde=False)
plt.gcf().suptitle("Average Days Between Orders")
plt.xlabel("Average Days Between Ordres")
plt.ylabel("Users")
plt.show()
sns.distplot(
candidates[UserColumns.AVG_DAYS_BETWEEN_ORDERS], hist_kws=dict(cumulative=True)
)
plt.gcf().suptitle("Average Days Between Orders < 300 CDF")
plt.xlabel("Average Days Between Orders")
plt.ylabel("CDF")
plt.xlim(0, 300)
plt.show()
sns.countplot(candidates[UserColumns.FIRST_ORDER_DIVISION])
plt.gcf().suptitle("Users by First Order Division")
plt.xlabel("Division")
plt.ylabel("Users")
plt.show()
sns.distplot(candidates[UserColumns.AVG_DAYS_BETWEEN_VISITS], kde=False)
plt.gcf().suptitle("Average Days Between Visits")
plt.xlabel("Average Days Between Visits")
plt.ylabel("Users")
plt.show()
sns.distplot(
candidates[UserColumns.AVG_DAYS_BETWEEN_VISITS], hist_kws=dict(cumulative=True)
)
plt.gcf().suptitle("Average Days Between Visits CDF")
plt.xlabel("Average Days Between Visits")
plt.ylabel("CDF")
plt.xlim(0, 300)
plt.show()
melted = pd.melt(
candidates,
id_vars=[
UserColumns.EMAIL,
UserColumns.LIFETIME_ORDERS,
UserColumns.FIRST_ORDER_DIVISION,
],
value_vars=[
UserColumns.AVG_DAYS_BETWEEN_ORDERS, UserColumns.AVG_DAYS_BETWEEN_VISITS
],
)
melted["variable"] = melted["variable"].apply(
lambda v: "orders" if v == UserColumns.AVG_DAYS_BETWEEN_ORDERS else "visits"
)
sns.lineplot(x=UserColumns.LIFETIME_ORDERS, y="value", hue="variable", data=melted)
plt.gcf().suptitle("Days between Events by Lifetime Orders")
plt.ylabel("Days between Events")
plt.xlabel("Lifetime Orders")
plt.xlim(2, 10)
plt.show()
sns.lineplot(
x=UserColumns.LIFETIME_ORDERS,
y="value",
hue="variable",
style=UserColumns.FIRST_ORDER_DIVISION,
data=melted[melted[UserColumns.FIRST_ORDER_DIVISION] != "unknown"],
)
plt.gcf().suptitle("Days between Events by Lifetime Orders")
plt.ylabel("Days between Events")
plt.xlabel("Lifetime Orders")
plt.xlim(2, 10)
plt.show()