microclaw Data Analysis
Compute, don't eyeball. Load the data, check it, then answer the actual question with numbers.
Check tooling first: python3 -c "import pandas" — if missing, fall back to the csv/statistics stdlib (see the csv-tools skill).
Explore first (shape, types, missing)
python3 - <<'PY'
import pandas as pd
df = pd.read_csv("data.csv") # or read_excel / read_json
print(df.shape)
print(df.dtypes)
print(df.isna().sum()) # missing values per column
print(df.describe(include="all")) # summary stats
PY
Aggregate, group, correlate
python3 - <<'PY'
import pandas as pd
df = pd.read_csv("data.csv")
print(df.groupby("category")["amount"].agg(["count","mean","sum"]))
print(df.select_dtypes("number").corr()) # correlation matrix
PY
Plot (save, then send as attachment)
python3 - <<'PY'
import pandas as pd, matplotlib
matplotlib.use("Agg") # headless
import matplotlib.pyplot as plt
df = pd.read_csv("data.csv")
df.groupby("month")["sales"].sum().plot(kind="bar")
plt.tight_layout(); plt.savefig("tmp/chart.png", dpi=120)
print("wrote tmp/chart.png")
PY
Workflow
- Restate the question and what would answer it.
- Load + sanity-check (shape, types, missing, obvious outliers).
- Compute the specific metrics; don't dump everything.
- Report the findings ("北区销量是南区的 2.3 倍，主要集中在 Q4"), with the numbers behind them.
- Note caveats: small samples, missing data, correlation ≠ causation.
Guidance
- Save charts under
tmp/and send viasend_messagewithattachment_path. - Don't over-plot; one clear chart that answers the question beats five decorative ones.
- State assumptions (how you handled missing/dupes) so results are reproducible.
- If pandas isn't installed and the task is light, use
csv-tools; if heavy, say what's needed.