Track Your Performance

This tutorial shows how to retrieve and analyze your submission history programmaticaly, track scores over time, and identify your best-performing strategies.

In [ ]:

%pip install crowdcent-challenge altair polars
import crowdcent_challenge as cc
import polars as pl
import altair as alt

%pip install crowdcent-challenge altair polars import crowdcent_challenge as cc import polars as pl import altair as alt

For this tutorial, you will need:

1. CrowdCent account: register for free
2. CrowdCent API Key: generate an API key from your user profile
3. Some scored submissions: Submit predictions and wait for scoring

Load API key¶

In [2]:

CROWDCENT_API_KEY = "API_KEY_HERE"

CROWDCENT_API_KEY = "API_KEY_HERE"

Initialize the client¶

In [ ]:

client = cc.ChallengeClient(
    challenge_slug="hyperliquid-ranking",
    api_key=CROWDCENT_API_KEY
)

client = cc.ChallengeClient( challenge_slug="hyperliquid-ranking", api_key=CROWDCENT_API_KEY )

2025-12-30 15:08:08,401 - INFO - ChallengeClient initialized for 'hyperliquid-ranking' at URL: https://crowdcent.com/api

Get Performance History¶

Use get_performance() to retrieve all your scored submissions as a list of dicts:

In [4]:

history = client.get_performance()
history_df = pl.DataFrame(history).with_columns(
    pl.col("release_date").str.to_date(),
    pl.col("submitted_at").str.to_datetime(),
)
history_df

history = client.get_performance() history_df = pl.DataFrame(history).with_columns( pl.col("release_date").str.to_date(), pl.col("submitted_at").str.to_datetime(), ) history_df

2025-12-30 15:08:08,479 - INFO - Fetching submission history for 'hyperliquid-ranking'...
2025-12-30 15:08:10,410 - INFO - Loaded 949 scored submissions.

Out[4]:

shape: (949, 14)

id	slot	release_date	submitted_at	status	score_ndcg@40_10d	score_ndcg@40_30d	score_spearman_10d	score_spearman_30d	percentile_ndcg@40_10d	percentile_ndcg@40_30d	percentile_spearman_10d	percentile_spearman_30d	composite_percentile
i64	i64	date	datetime[μs, UTC]	str	f64	f64	f64	f64	f64	f64	f64	f64	f64
5880	5	2025-12-28	2025-12-28 14:44:37.055198 UTC	"pending"	0.337716	0.340841	-0.533123	-0.523227	8.108108	8.108108	10.810811	10.810811	9.459459
5879	4	2025-12-28	2025-12-28 14:44:36.327841 UTC	"pending"	0.421611	0.419226	-0.417002	-0.384453	62.162162	48.648649	70.27027	62.162162	60.810811
5878	3	2025-12-28	2025-12-28 14:44:35.920831 UTC	"pending"	0.385517	0.388523	-0.499784	-0.464013	16.216216	18.918919	27.027027	35.135135	24.324324
5877	2	2025-12-28	2025-12-28 14:44:34.445854 UTC	"pending"	0.452958	0.423994	-0.47497	-0.469166	81.081081	54.054054	59.459459	29.72973	56.081081
5876	1	2025-12-28	2025-12-28 14:44:34.074837 UTC	"pending"	0.414794	0.492791	-0.415315	-0.242053	56.756757	89.189189	72.972973	83.783784	75.675676
…	…	…	…	…	…	…	…	…	…	…	…	…	…
14	2	2025-06-06	2025-06-06 16:04:20.309350 UTC	"evaluated"	0.647897	0.63497	0.145175	0.133588	66.666667	66.666667	66.666667	66.666667	null
13	1	2025-06-06	2025-06-06 16:04:19.820281 UTC	"evaluated"	0.647897	0.63497	0.145175	0.133588	66.666667	66.666667	66.666667	66.666667	null
12	2	2025-06-05	2025-06-05 15:06:38.631117 UTC	"evaluated"	0.540538	0.651426	0.004378	0.119459	66.666667	66.666667	66.666667	66.666667	null
11	3	2025-06-05	2025-06-05 15:04:02.653942 UTC	"evaluated"	0.55855	0.690496	0.051245	0.227518	100.0	100.0	100.0	100.0	null
10	1	2025-06-05	2025-06-05 14:37:09.450635 UTC	"evaluated"	0.540538	0.651426	0.004378	0.119459	66.666667	66.666667	66.666667	66.666667	null

Each dict includes:

• id, slot, release_date, submitted_at, status — submission metadata
• score_* keys — raw metric scores (e.g., score_spearman_10d)
• percentile_* keys — your percentile rank vs other participants
• composite_percentile — overall ranking (if available)

Analyze Performance by Slot¶

If you're using multiple slots for different strategies, compare them:

In [5]:

# Average percentile by slot
slot_performance = (
    history_df.group_by("slot")
    .agg(
        pl.col("composite_percentile").mean().alias("avg_percentile"),
        pl.col("composite_percentile").std().alias("std_percentile"),
        pl.col("composite_percentile").count().alias("submissions"),
    )
    .sort("avg_percentile", descending=True)
)
slot_performance

# Average percentile by slot slot_performance = ( history_df.group_by("slot") .agg( pl.col("composite_percentile").mean().alias("avg_percentile"), pl.col("composite_percentile").std().alias("std_percentile"), pl.col("composite_percentile").count().alias("submissions"), ) .sort("avg_percentile", descending=True) ) slot_performance

Out[5]:

shape: (5, 4)

slot	avg_percentile	std_percentile	submissions
i64	f64	f64	u32
3	68.601376	20.018743	181
4	67.555864	22.640585	181
5	62.624793	26.325876	168
2	53.486718	21.407269	194
1	42.384833	21.720835	194

Visualize Performance Over Time¶

Plot your percentile trajectory with a moving average:

In [6]:

window_size = 5

# Add rolling averages for percentile metrics
history_df = history_df.sort("release_date").with_columns(
    [
        pl.col("composite_percentile")
        .rolling_mean(window_size=window_size)
        .over("slot")
        .alias(f"composite_percentile_ma{window_size}"),
    ]
)

window_size = 5 # Add rolling averages for percentile metrics history_df = history_df.sort("release_date").with_columns( [ pl.col("composite_percentile") .rolling_mean(window_size=window_size) .over("slot") .alias(f"composite_percentile_ma{window_size}"), ] )

In [7]:

chart = (
    alt.Chart(history_df)
    .mark_line(point=True)
    .encode(
        x=alt.X("release_date:T", title="Inference Period"),
        y=alt.Y("composite_percentile_ma5:Q", title="Percentile"),
        color=alt.Color("slot:N", title="Slot"),
        tooltip=["release_date", "slot", "composite_percentile"],
    )
    .properties(title="Performance Over Time", width=600, height=300)
)
chart

chart = ( alt.Chart(history_df) .mark_line(point=True) .encode( x=alt.X("release_date:T", title="Inference Period"), y=alt.Y("composite_percentile_ma5:Q", title="Percentile"), color=alt.Color("slot:N", title="Slot"), tooltip=["release_date", "slot", "composite_percentile"], ) .properties(title="Performance Over Time", width=600, height=300) ) chart

Out[7]: