Crafte Sessions analysis 2026 09 29 2026 10 29
sessions_analysis-2026-09-29-2026-10-29.csv · 2,246 rows · 15 columns · definition v3 saved 29 Oct 2026 14:08
This feature set was built from column definition v1; the definition is now v3. Rebuild to pick up the changes.
Feature set v3 built 29 Oct 2026 12:50
Snapshots
2,246
End of each active day
Entities
2,014
2,246 rows used
Inputs used
21 of 49
after encoding
PCA
Off
14 components would explain 90%
Scaling
Standard
What to do
From this build's numbers: 21 inputs now, 21 if you take every suggestion. Buttons only tick boxes on the Inputs tab and options on Settings & rebuild; nothing changes until you rebuild.- Fine as is Leave PCA off
It takes 14 components to hold 90% of what your 21 inputs say, so the inputs mostly carry different information. PCA would save little and the segments would be harder to explain.
- Fine as is Scaling: keep Standard
For segments every input should get an equal say. Standard gives all 21 inputs the same weight. Min-max would squash inputs with a long tail, so they'd barely count.
Compare the scalings
Scaling Inputs with a real say Who dominates Standard 21 of 21 all equal Robust 20.8 of 21 fairly even Min-max 8.1 of 21 mostly device=mobile and landingPage=/ (39%) None 2.7 of 21 mostly minutes and tenure_days (68%)
Correlated pairs are flagged from |r| ≥ 0.9, a common rule of thumb; nothing is removed automatically. The Correlations tab shows the pairs behind each suggestion.
Strongly correlated inputs
|r| ≥ 0.9No pairs above the threshold.
How the inputs relate
The 20 inputs most tied up with others, grouped so related ones sit together. Red move together, blue move opposite.Nothing to remove: no pair reaches 0.9. The strongest is
minutes and pages (+0.70).- Nothing to remove: no pair reaches 0.9. The strongest is
minutesandpages(+0.70). - Each cell is the correlation between the input on its row and the one in its column, across all snapshots, from −1 to +1. Only the lower half is shown: the upper half would repeat it. Red: they rise together. Blue: one rises as the other falls. Near 0: unrelated.
- Grey diagonal: an input against itself. Hatched: two values of the same column (device=mobile and device=tablet); only one can be 1, so their link is built in and means nothing.
- The decision: pairs at 0.9 or above say the same thing twice; keep one. Below that, keep both: related inputs still carry their own information.
minutesandpages(+0.70): they tend to rise together. Related but not duplicates; in a model they share their effect, so each can look weaker on its own.rowsandpages(+0.60): they tend to rise together. Related but not duplicates; in a model they share their effect, so each can look weaker on its own.tenure_daysandpages(+0.52): they tend to rise together. Related but not duplicates; in a model they share their effect, so each can look weaker on its own.newVisitorandsource=Facebook(+0.41): they tend to rise together. A moderate link worth knowing, not a problem.- 84% of the pairs shown are near 0 (under 0.1), so most inputs carry separate information, which suits segmentation and models.
| actions[add_checkout] | actions[add_recommended] | actions[co_cardpay] | actions[add_search] | rows | tenure_days | minutes | pages | newVisitor | source=Facebook | source=Direct | landingPage=/product/75-christmas-cake-boxed | landingPage=/product/79-chocolate-fruit-cake-500g-boxed | landingPage=/ | landingPage=other | actions[sub_start] | actions[co_fail] | device=desktop | device=mobile | device=tablet | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
actions[add_checkout] | ||||||||||||||||||||
actions[add_recommended] | ||||||||||||||||||||
actions[co_cardpay] | ||||||||||||||||||||
actions[add_search] | ||||||||||||||||||||
rows | -0.0 | -0.0 | 0.1 | -0.0 | ||||||||||||||||
tenure_days | -0.0 | -0.0 | 0.1 | -0.0 | ||||||||||||||||
minutes | 0.0 | 0.0 | 0.2 | 0.0 | 0.4 | 0.3 | ||||||||||||||
pages | 0.1 | 0.1 | 0.3 | 0.1 | 0.6 | 0.5 | 0.7 | |||||||||||||
newVisitor | 0.0 | 0.0 | 0.0 | 0.0 | -0.3 | -0.2 | -0.2 | -0.3 | ||||||||||||
source=Facebook | 0.0 | -0.0 | 0.0 | 0.0 | -0.1 | -0.1 | -0.2 | -0.2 | 0.4 | |||||||||||
source=Direct | -0.0 | 0.0 | -0.0 | -0.0 | 0.0 | -0.0 | 0.1 | 0.1 | -0.2 | |||||||||||
landingPage=/product/75-christmas-cake-boxed | -0.0 | -0.0 | 0.1 | -0.0 | 0.1 | 0.1 | 0.1 | 0.1 | 0.0 | 0.0 | -0.0 | |||||||||
landingPage=/product/79-chocolate-fruit-cake-500g-boxed | -0.0 | -0.0 | 0.0 | -0.0 | -0.0 | -0.0 | -0.0 | 0.1 | 0.0 | 0.0 | -0.0 | |||||||||
landingPage=/ | 0.0 | 0.0 | -0.1 | 0.0 | -0.1 | -0.1 | -0.1 | -0.2 | 0.0 | 0.2 | -0.2 | |||||||||
landingPage=other | -0.0 | -0.0 | 0.0 | -0.0 | 0.1 | 0.1 | 0.1 | 0.1 | -0.1 | -0.2 | 0.2 | |||||||||
actions[sub_start] | -0.0 | -0.0 | 0.0 | 0.1 | 0.0 | 0.0 | -0.0 | -0.0 | -0.0 | 0.0 | -0.0 | |||||||||
actions[co_fail] | -0.0 | -0.0 | -0.0 | 0.0 | 0.0 | 0.0 | -0.0 | -0.0 | -0.0 | 0.0 | -0.0 | |||||||||
device=desktop | -0.0 | -0.0 | 0.0 | -0.0 | 0.1 | 0.0 | -0.0 | 0.0 | -0.0 | -0.1 | 0.1 | -0.0 | -0.0 | -0.0 | 0.0 | -0.0 | -0.0 | |||
device=mobile | 0.0 | 0.0 | -0.0 | 0.0 | -0.0 | -0.0 | 0.1 | 0.0 | 0.0 | 0.1 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | |||
device=tablet | -0.0 | -0.0 | -0.0 | -0.0 | -0.0 | 0.0 | -0.1 | -0.0 | -0.0 | 0.0 | -0.0 | 0.0 | 0.0 | 0.0 | -0.0 | 0.1 | 0.1 |
Outlined cells are at or above the 0.9 threshold. Hatched cells are two values of the same column (only one can be 1, so they always pull apart) and are left out of the ranking. "+N identical" means other inputs match that one exactly and are folded into its row.
Variation explained
14 of 21 components are needed for 90%: the inputs carry mostly separate information, so PCA wouldn't shrink them much. Keep it off.
- 14 of 21 components are needed for 90%: the inputs carry mostly separate information, so PCA wouldn't shrink them much. Keep it off.
- PCA (principal component analysis) rewrites the inputs as new combined ones, called components, ordered by how much of the variation between snapshots each captures.
- Bars: the share each component captures on its own. The first here takes 15%.
- Line: the running total. The dashed lines mark 90% and how many components it takes to get there.
- The decision: if a few components reach 90% (under about 40% of the inputs), the inputs largely repeat each other and PCA simplifies them. If it takes most of them, there's little to compress: keep PCA off, because the inputs themselves are easier to explain.
- A steep drop after the first few bars (an 'elbow') also says those first components hold most of the structure.
Each componentRunning total
12 components explain 80% of the variation, 14 explain 90%; 16 explain 95%. PCA is off: the segments use the inputs directly, and this chart is only for comparison. Fitted on 2,246 snapshots.
What PCA would combine
PCA is off, so none of this is used: it shows how inputs would group if you turned it on.Component 1 is mostly
pages (+0.47), rows (+0.43), tenure_days (+0.40). Component 2 is mostly source=Direct (+0.51), source=Facebook (-0.50).- Component 1 is mostly
pages(+0.47),rows(+0.43),tenure_days(+0.40). Component 2 is mostlysource=Direct(+0.51),source=Facebook(-0.50). - Each column is a component and each row an input. A cell (the loading) is how much that input contributes to the component: big red or blue means it matters, near 0 means it doesn't.
- Same colour down a column: those inputs rise together in that component. Opposite colours: the component contrasts them (high in one, low in the other).
- To name a component, read its two or three biggest cells: e.g. 'busy session' if pages and minutes are both strongly red.
- If PCA is on, segments are built from these components, so a segment described as 'high on component 1' means high on whatever component 1 is made of.
| PC1 | PC2 | PC3 | PC4 | PC5 | PC6 | PC7 | PC8 | PC9 | PC10 | |
|---|---|---|---|---|---|---|---|---|---|---|
landingPage=/product/474-port-and-prune-pudding-800g-log-in-cloth | +0.02 | +0.01 | -0.06 | -0.25 | -0.17 | -0.16 | -0.36 | -0.07 | +0.73 | -0.04 |
actions[add_recommended] | +0.03 | -0.01 | -0.03 | +0.06 | +0.11 | +0.13 | +0.36 | -0.36 | +0.27 | -0.68 |
device=mobile | -0.02 | -0.24 | -0.67 | +0.12 | -0.05 | +0.03 | -0.05 | +0.02 | -0.01 | +0.01 |
device=desktop | +0.05 | +0.21 | +0.39 | -0.04 | -0.01 | -0.64 | +0.25 | -0.06 | -0.01 | -0.02 |
actions[add_checkout] | +0.01 | -0.03 | -0.02 | +0.03 | +0.07 | +0.03 | +0.33 | -0.29 | +0.10 | +0.64 |
landingPage=/ | -0.20 | -0.22 | +0.20 | +0.60 | +0.09 | -0.03 | -0.01 | +0.04 | -0.01 | +0.00 |
actions[add_search] | +0.04 | -0.06 | -0.02 | -0.04 | +0.59 | -0.14 | -0.12 | +0.34 | +0.17 | +0.04 |
device=tablet | -0.02 | +0.10 | +0.49 | -0.12 | +0.07 | +0.57 | -0.17 | +0.03 | +0.02 | +0.00 |
actions[co_cardpay] | +0.14 | -0.12 | -0.00 | -0.18 | +0.57 | -0.15 | -0.12 | +0.06 | -0.01 | -0.01 |
source=Direct | +0.23 | +0.51 | -0.15 | +0.20 | +0.06 | +0.03 | -0.04 | +0.00 | +0.00 | -0.03 |
landingPage=/product/79-chocolate-fruit-cake-500g-boxed | +0.04 | +0.03 | -0.08 | -0.44 | -0.08 | +0.20 | +0.51 | +0.47 | -0.05 | -0.05 |
source=Facebook | -0.28 | -0.50 | +0.12 | -0.24 | -0.03 | -0.04 | +0.05 | -0.05 | -0.01 | +0.01 |
pages | +0.47 | -0.18 | +0.01 | +0.02 | +0.14 | +0.05 | +0.15 | -0.03 | +0.05 | +0.00 |
landingPage=/product/75-christmas-cake-boxed | +0.08 | -0.05 | +0.01 | -0.24 | +0.15 | +0.07 | -0.30 | -0.45 | -0.45 | -0.12 |
rows | +0.43 | -0.27 | +0.14 | +0.02 | -0.27 | -0.06 | -0.08 | +0.06 | -0.01 | -0.00 |
tenure_days | +0.40 | -0.28 | +0.14 | +0.00 | -0.29 | -0.06 | -0.10 | +0.05 | -0.00 | +0.00 |
minutes | +0.38 | -0.15 | -0.04 | +0.09 | +0.20 | +0.08 | +0.14 | -0.10 | -0.02 | +0.01 |
actions[sub_start] | +0.02 | -0.01 | +0.06 | +0.00 | +0.11 | +0.23 | +0.18 | -0.32 | +0.34 | +0.28 |
landingPage=other | +0.16 | +0.27 | -0.15 | -0.28 | -0.07 | -0.07 | -0.11 | -0.20 | -0.12 | +0.15 |
newVisitor | -0.26 | -0.15 | -0.01 | -0.26 | +0.07 | -0.15 | +0.07 | -0.21 | -0.04 | -0.03 |
actions[co_fail] | -0.01 | +0.01 | +0.06 | -0.01 | +0.03 | +0.19 | -0.21 | +0.12 | +0.13 | +0.07 |
Inputs
Tick "Leave out", then rebuild21 of 49 inputs are used.
- 21 of 49 inputs are used.
- Each row is one input the segments and models see. A category becomes one input per value (one-hot); a list becomes one count per item.
- Mean / Std dev / Range: after the transform (log, etc.), before scaling. Zeros: the share of snapshots at 0; very high means the input rarely says anything.
- Most correlated with: its closest other input; at 0.9 or above, one of the two can go.
- Outcome link: its strongest correlation with any outcome. Very high (0.8+) usually means it's part of the outcome (leakage): leave it out.
- Leave out: tick, then Rebuild feature set. Click an input's name for its distribution.
| Input | From column | Encoding | Mean | Std dev | Range | Zeros | Filled | Most correlated with | Outcome link | Leave out |
|---|---|---|---|---|---|---|---|---|---|---|
| (activity) | number · log | 0.748 | 0.18 | 0.693 – 2.3 | 0.0% | +0.60 pages | +0.20 | |||
| left out | (activity) | number · log | 0.748 | 0.18 | 0.693 – 2.3 | 0.0% | ||||
| (activity) | number · log | 0.166 | 0.552 | 0 – 3.04 | 89.7% | +0.52 pages | +0.19 | |||
| minutes | number · log | 0.526 | 1.37 | 0 – 7.42 | 81.1% | +0.70 pages | +0.44 | |||
| pages | number · log | 0.939 | 0.496 | 0.693 – 4.62 | 0.0% | +0.70 minutes | +0.62 | |||
| source | one-hot | 0.958 | 0.2 | 0 – 1 | 4.2% | +0.41 newVisitor | -0.24 | |||
| source | one-hot | 0.0298 | 0.17 | 0 – 1 | 97.0% | -0.23 newVisitor | +0.23 | |||
| left out | source | one-hot | 0.00801 | 0.0892 | 0 – 1 | 99.2% | ||||
| left out | source | one-hot | 0.00223 | 0.0471 | 0 – 1 | 99.8% | ||||
| left out | source | one-hot | 0.00089 | 0.0298 | 0 – 1 | 99.9% | ||||
| left out | source | one-hot | 0.000445 | 0.0211 | 0 – 1 | 100.0% | ||||
| left out | source | one-hot | 0.000445 | 0.0211 | 0 – 1 | 100.0% | ||||
| landingPage | one-hot | 0.825 | 0.38 | 0 – 1 | 17.5% | -0.22 source=Direct | -0.12 | |||
| landingPage | one-hot | 0.0392 | 0.194 | 0 – 1 | 96.1% | +0.06 pages | -0.03 | |||
| landingPage | one-hot | 0.0142 | 0.119 | 0 – 1 | 98.6% | -0.04 minutes | -0.02 | |||
| landingPage | one-hot | 0.0111 | 0.105 | 0 – 1 | 98.9% | +0.12 actions[co_cardpay] | +0.07 | |||
| left out | landingPage | one-hot | 0.0098 | 0.0985 | 0 – 1 | 99.0% | ||||
| left out | landingPage | one-hot | 0.0089 | 0.0939 | 0 – 1 | 99.1% | ||||
| left out | landingPage | one-hot | 0.00623 | 0.0787 | 0 – 1 | 99.4% | ||||
| left out | landingPage | one-hot | 0.00534 | 0.0729 | 0 – 1 | 99.5% | ||||
| left out | landingPage | one-hot | 0.0049 | 0.0698 | 0 – 1 | 99.5% | ||||
| left out | landingPage | one-hot | 0.00445 | 0.0666 | 0 – 1 | 99.6% | ||||
| left out | landingPage | one-hot | 0.00401 | 0.0632 | 0 – 1 | 99.6% | ||||
| left out | landingPage | one-hot | 0.00401 | 0.0632 | 0 – 1 | 99.6% | ||||
| left out | landingPage | one-hot | 0.00356 | 0.0596 | 0 – 1 | 99.6% | ||||
| left out | landingPage | one-hot | 0.00312 | 0.0557 | 0 – 1 | 99.7% | ||||
| left out | landingPage | one-hot | 0.00312 | 0.0557 | 0 – 1 | 99.7% | ||||
| landingPage | one-hot (other) | 0.0525 | 0.223 | 0 – 1 | 94.7% | -0.21 source=Facebook | +0.05 | |||
| device | one-hot | 0.75 | 0.433 | 0 – 1 | 25.0% | +0.06 source=Facebook | +0.05 | |||
| device | one-hot | 0.132 | 0.338 | 0 – 1 | 86.8% | +0.05 actions[co_fail] | -0.04 | |||
| device | one-hot | 0.118 | 0.323 | 0 – 1 | 88.2% | -0.11 source=Facebook | +0.03 | |||
| newVisitor | number | 0.993 | 0.0841 | 0 – 1 | 0.7% | +0.41 source=Facebook | -0.14 | |||
| left out | actions | number · log | 0.052 | 0.188 | 0 – 1.39 | 92.7% | ||||
| left out | actions | number · log | 0.0267 | 0.134 | 0 – 1.1 | 96.2% | ||||
| left out | actions | number · log | 0.0258 | 0.132 | 0 – 1.1 | 96.3% | ||||
| left out | actions | number · log | 0.0249 | 0.13 | 0 – 1.1 | 96.4% | ||||
| left out | actions | number · log | 0.0249 | 0.13 | 0 – 1.1 | 96.4% | ||||
| left out | actions | number · log | 0.0228 | 0.126 | 0 – 1.1 | 96.8% | ||||
| left out | actions | number · log | 0.0187 | 0.113 | 0 – 1.1 | 97.3% | ||||
| left out | actions | number · log | 0.0136 | 0.0961 | 0 – 0.693 | 98.0% | ||||
| left out | actions | number · log | 0.0136 | 0.0961 | 0 – 0.693 | 98.0% | ||||
| left out | actions | number · log | 0.0116 | 0.09 | 0 – 1.1 | 98.4% | ||||
| left out | actions | number · log | 0.00772 | 0.0727 | 0 – 0.693 | 98.9% | ||||
| actions | number · log | 0.00679 | 0.0683 | 0 – 0.693 | 99.0% | +0.26 pages | +0.54 | |||
| actions | number · log | 0.00247 | 0.0413 | 0 – 0.693 | 99.6% | +0.10 pages | +0.19 | |||
| actions | number · log | 0.000617 | 0.0207 | 0 – 0.693 | 99.9% | +0.06 pages | +0.09 | |||
| actions | number · log | 0.000309 | 0.0146 | 0 – 0.693 | 100.0% | +0.10 pages | +0.11 | |||
| actions | number · log | 0.000309 | 0.0146 | 0 – 0.693 | 100.0% | +0.05 device=tablet | +0.07 | |||
| actions | number · log | 0.000309 | 0.0146 | 0 – 0.693 | 100.0% | +0.07 pages | +0.07 |
Feature set versions
| Version | Built | Definition | Snapshots | Inputs | PCA | Scaling |
|---|---|---|---|---|---|---|
| v8 | 30 Oct 2026 12:27 | v3 | End of each active day | 18 | off | Standard |
| v7 | 29 Oct 2026 17:16 | v3 | End of each active day | 18 | off | Standard |
| v6 | 29 Oct 2026 14:08 | v3 | End of each active day | 18 | off | Standard |
| v5 | 29 Oct 2026 14:01 | v2 | End of each active day | 18 | off | Standard |
| v4 | 29 Oct 2026 13:09 | v2 | End of each active day | 21 | off | Standard |
| v3 | 29 Oct 2026 12:50 | v1 | End of each active day | 21 | off | Standard |
| v2 | 29 Oct 2026 12:28 | v1 | End of each active day | 49 | off | Standard |
| v1 | 29 Oct 2026 12:19 | v1 | End of each active day | 49 | off | Standard |