cudf-analytics

Name: cudf-analytics
Author: langchain-ai

192installs3.7kforks

How to install cudf-analytics?

langchain-ai/deepagents/cudf-analytics

$npx -y skills add langchain-ai/deepagents --skill cudf-analytics

Installs into the current project.

Summary

Use for GPU-accelerated data analysis on datasets, CSVs, or tabular data using NVIDIA cuDF. Triggers when tasks involve groupby aggregations, statistical summaries, anomaly detection, or large-scale data profiling.

Files · 1

View on GitHub

SKILL.md

1	# cuDF Analytics Skill
2
3	GPU-accelerated data analysis using NVIDIA RAPIDS cuDF. cuDF provides a pandas-like API that runs on NVIDIA GPUs, enabling massive speedups on large datasets.
4
5	## When to Use This Skill
6
7	Use this skill when:
8	- Analyzing CSV files, datasets, or tabular data
9	- Computing statistical summaries (mean, median, std, quartiles)
10	- Performing groupby aggregations
11	- Detecting anomalies or outliers in data
12	- Profiling datasets with millions of rows
13	- Computing correlation matrices
14
15	## Initialization (REQUIRED)
16
17	Always start every script with this boilerplate. It tests actual GPU operations, not just import.
18
19	```python
20	import pandas as pd
21
22	try:
23	import cudf
24	# Smoke-test: verify GPU compute AND host transfer both work
25	_test = cudf.Series([1, 2, 3])
26	assert _test.sum() == 6
27	assert _test.to_pandas().tolist() == [1, 2, 3]
28	GPU = True
29	except Exception as e:
30	print(f"[GPU] cudf unavailable, falling back to pandas: {e}")
31	GPU = False
32
33	def read_csv(path):
34	return cudf.read_csv(path) if GPU else pd.read_csv(path)
35
36	def to_pd(df):
37	"""Convert cuDF DataFrame/Series to pandas. Use this instead of .to_pandas() directly."""
38	if not GPU:
39	return df
40	try:
41	return df.to_pandas()
42	except Exception as e:
43	print(f"[GPU] .to_pandas() failed, using Arrow fallback: {e}")
44	return df.to_arrow().to_pandas()
45	```
46
47	## Quick Reference
48
49	cuDF mirrors the pandas API. Common operations:
50
51	### Read Data
52	```python
53	df = read_csv("data.csv")
54	```
55
56	### Statistical Summary
57	```python
58	# Use to_pd() when you need pandas output
59	summary = to_pd(df[["value", "score"]].describe())
60
61	# Scalar values work directly with float()
62	mean_val = float(df["value"].mean())
63	q1 = float(df["value"].quantile(0.25))
64
65	# Correlation
66	corr = float(df["value"].corr(df["score"]))
67	```
68
69	### Groupby Aggregation
70	```python
71	result = df.groupby("category").agg({
72	"revenue": ["sum", "mean", "count"],
73	"quantity": ["sum", "mean"],
74	})
75	result_pd = to_pd(result)
76	```
77
78	### Anomaly Detection (IQR Method)
79	```python
80	col = "value"
81	Q1 = float(df[col].quantile(0.25))
82	Q3 = float(df[col].quantile(0.75))
83	IQR = Q3 - Q1
84	lower = Q1 - 1.5 * IQR
85	upper = Q3 + 1.5 * IQR
86	outliers = to_pd(df[(df[col] < lower) \| (df[col] > upper)])
87	```
88
89	### Anomaly Detection (Z-Score Method)
90	```python
91	mean = float(df[col].mean())
92	std = float(df[col].std())
93	df["z_score"] = (df[col] - mean) / std
94	anomalies = to_pd(df[df["z_score"].abs() > 3])
95	```
96
97	### Filtering and Selection
98	```python
99	# Filter rows
100	filtered = df[df["status"] == "active"]
101
102	# Select columns
103	subset = df[["name", "revenue", "date"]]
104
105	# Sort
106	sorted_df = df.sort_values("revenue", ascending=False)
107
108	# Convert to pandas for final output / iteration
109	result_pd = to_pd(sorted_df)
110	```
111
112	## Data Type Requirements
113
114	cuDF requires explicit type specification for optimal performance:
115	- Use `float32` or `float64` for numeric data
116	- Use `int32` or `int64` for integer data
117	- String columns use cuDF's string dtype automatically
118
119	## Output Guidelines
120
121	When reporting analysis results:
122	- Include dataset dimensions (rows x columns)
123	- Show key statistics in formatted tables
124	- Highlight notable patterns, trends, or anomalies
125	- Provide both summary statistics and specific examples
126	- Note any data quality issues (missing values, outliers)

People who liked this also viewed

6 picks

Type

cudf-analytics

192installs3.7kforks

View on GitHub

How to install cudf-analytics?

langchain-ai/deepagents/cudf-analytics

$npx -y skills add langchain-ai/deepagents --skill cudf-analytics

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# cuDF Analytics Skill
2
3	GPU-accelerated data analysis using NVIDIA RAPIDS cuDF. cuDF provides a pandas-like API that runs on NVIDIA GPUs, enabling massive speedups on large datasets.
4
5	## When to Use This Skill
6
7	Use this skill when:
8	- Analyzing CSV files, datasets, or tabular data
9	- Computing statistical summaries (mean, median, std, quartiles)
10	- Performing groupby aggregations
11	- Detecting anomalies or outliers in data
12	- Profiling datasets with millions of rows
13	- Computing correlation matrices
14
15	## Initialization (REQUIRED)
16
17	Always start every script with this boilerplate. It tests actual GPU operations, not just import.
18
19	```python
20	import pandas as pd
21
22	try:
23	import cudf
24	# Smoke-test: verify GPU compute AND host transfer both work
25	_test = cudf.Series([1, 2, 3])
26	assert _test.sum() == 6
27	assert _test.to_pandas().tolist() == [1, 2, 3]
28	GPU = True
29	except Exception as e:
30	print(f"[GPU] cudf unavailable, falling back to pandas: {e}")
31	GPU = False
32
33	def read_csv(path):
34	return cudf.read_csv(path) if GPU else pd.read_csv(path)
35
36	def to_pd(df):
37	"""Convert cuDF DataFrame/Series to pandas. Use this instead of .to_pandas() directly."""
38	if not GPU:
39	return df
40	try:
41	return df.to_pandas()
42	except Exception as e:
43	print(f"[GPU] .to_pandas() failed, using Arrow fallback: {e}")
44	return df.to_arrow().to_pandas()
45	```
46
47	## Quick Reference
48
49	cuDF mirrors the pandas API. Common operations:
50
51	### Read Data
52	```python
53	df = read_csv("data.csv")
54	```
55
56	### Statistical Summary
57	```python
58	# Use to_pd() when you need pandas output
59	summary = to_pd(df[["value", "score"]].describe())
60
61	# Scalar values work directly with float()
62	mean_val = float(df["value"].mean())
63	q1 = float(df["value"].quantile(0.25))
64
65	# Correlation
66	corr = float(df["value"].corr(df["score"]))
67	```
68
69	### Groupby Aggregation
70	```python
71	result = df.groupby("category").agg({
72	"revenue": ["sum", "mean", "count"],
73	"quantity": ["sum", "mean"],
74	})
75	result_pd = to_pd(result)
76	```
77
78	### Anomaly Detection (IQR Method)
79	```python
80	col = "value"
81	Q1 = float(df[col].quantile(0.25))
82	Q3 = float(df[col].quantile(0.75))
83	IQR = Q3 - Q1
84	lower = Q1 - 1.5 * IQR
85	upper = Q3 + 1.5 * IQR
86	outliers = to_pd(df[(df[col] < lower) \| (df[col] > upper)])
87	```
88
89	### Anomaly Detection (Z-Score Method)
90	```python
91	mean = float(df[col].mean())
92	std = float(df[col].std())
93	df["z_score"] = (df[col] - mean) / std
94	anomalies = to_pd(df[df["z_score"].abs() > 3])
95	```
96
97	### Filtering and Selection
98	```python
99	# Filter rows
100	filtered = df[df["status"] == "active"]
101
102	# Select columns
103	subset = df[["name", "revenue", "date"]]
104
105	# Sort
106	sorted_df = df.sort_values("revenue", ascending=False)
107
108	# Convert to pandas for final output / iteration
109	result_pd = to_pd(sorted_df)
110	```
111
112	## Data Type Requirements
113
114	cuDF requires explicit type specification for optimal performance:
115	- Use `float32` or `float64` for numeric data
116	- Use `int32` or `int64` for integer data
117	- String columns use cuDF's string dtype automatically
118
119	## Output Guidelines
120
121	When reporting analysis results:
122	- Include dataset dimensions (rows x columns)
123	- Show key statistics in formatted tables
124	- Highlight notable patterns, trends, or anomalies
125	- Provide both summary statistics and specific examples
126	- Note any data quality issues (missing values, outliers)

People who liked this also viewed

6 picks

Type