Outils

Plotly express : import plotly.express as px

Données

Evolutions

Graphes possibles :

line (ligne)
histogram (barre)

Multiline

2 axes :

fig = px.line(df, 
    x='date', 
    y=['sum', 'median'], 
    markers=True # Affichage des points
 ) 

3 axes :

fig = px.line(df, 
    x='date', 
    y='annonces', 
    color='region',
 ) 

Manuel :

fig = go.Figure()

for i, column_value in enumerate(df[column_line].unique()):
    df_subset = df[df[column_line] == column_value]
    fig.add_trace(
        go.Scatter(
            x=df_subset[column_x], 
            y=df_subset[column_y], 
            mode='lines', 
            name=str(column_value),
            fillcolor=colors[i]
        )
    )
    
for i, row in df.groupby(column_line).first().reset_index().iterrows():
    fig.add_annotation(
        x=row[column_x], 
        y=row[column_y], 
        text=row[column_line],
        showarrow=False, 
        yshift=10,
        xshift=10,
        font=dict(family="Arial", size=14, weight='bold', color=colors[i])
    )

Corrélations

Graphes possibles :

scatter_matrix (matrice de corrélation)
imshow (heatmap)

N-N

fig = px.scatter_matrix(df,
    dimensions=columns,
    color='year',
)

fig.update_traces(diagonal_visible=False)

fig = px.imshow(
    df_graph.loc[:, columns].corr(),
    zmin=-1,
    zmax=1,
    text_auto=".2f",
    color_continuous_scale="rdbu", # -1 : Rouge, 0 : Blanc, 1 : Bleu
)

X-Y

df_corrs = pd.DataFrame({
    "Feature": columns_x,
})

for target in columns_y:
    df_corrs[target] = df[[target, *columns_x]].corr().loc[target, columns_x].values

fig = px.imshow(
    df_corrs.set_index("Feature").T,
    text_auto="0.2f",
    color_continuous_scale="rdbu",
    zmin=-1,
    zmax=1,
)

Outliers

fig = px.box(
    df, 
    x=x, 
    y=y,
)

Affichage

Lignes

Couleurs :

fig = px.line(df,
    color_discrete_sequence=color,
)

Annoter les lignes :

for i, serie_name in enumerate([trace.name for trace in fig.data]):
    subset = df.loc[df[color].astype(str) == serie_name].reset_index()
    fig.add_annotation(
        x=subset[x][0],  
        y=subset[y][0],  
        text=serie_name,
        showarrow=False,
        yshift=0,
        xshift=-20,
        textangle=0,
        font=dict(
            family="Arial", 
            size=14, 
            color=colors[i]
        )
    )

Modification des tracès :

fig.for_each_trace(
    lambda t: t.update(
        name = names[t.name],
        legendgroup = names[t.name],
        hovertemplate = t.hovertemplate.replace(t.name, names[t.name])
    )
)

Axes

Nom des axes :

fig = px.line(df,
    labels={ # Renomage des axes
        'date_day': 'Date', 
        'annonces': "Nombre d'annonces", 
        'region': 'Région'
    }, 
 ) 

Orientation de la légende sur X : fig.update_xaxes(tickangle=45)

Layout

fig.update_layout(
    title="Nombre d'annonces par jour",
    height=600,
    width=1000,
    showlegend=True,
)

Export

Image : fig.write_image(f"../graphs/{filename}.jpg", width=1920, height=1080, scale=2)
Html : fig.write_html(f"../graphs/{filename}.html")

Notes

Exemple de schéma de couleurs de plotly express : px.colors.qualitative.Dark24

colors_32 = [
    "#CC00FF", # Violet clair
    "#9933FF", # Violet foncé
    "#6633FF", # Violet
    "#0066FF", # Bleu foncé
    "#3399FF", # Bleu moyen
    "#00FFFF", # Cyan
    "#33CCFF", # Bleu clair
    "#33CCCC", # Turquoise
    "#33CC99", # Vert bleuâtre
    "#33CC33", # Vert émeraude
    "#66CC00", # Vert foncé
    "#99CC00", # Vert moyen
    "#FFCC00", # Orange clair
    "#FF9933", # Orange
    "#FF6347", # Orange vif
    "#FF0000", # Rouge vif
    "#CC00FF", # Violet clair
    "#9933FF", # Violet foncé
    "#6633FF", # Violet
    "#0066FF", # Bleu foncé
    "#3399FF", # Bleu moyen
    "#00FFFF", # Cyan
    "#33CCFF", # Bleu clair
    "#33CCCC", # Turquoise
    "#33CC99", # Vert bleuâtre
    "#33CC33", # Vert émeraude
    "#66CC00", # Vert foncé
    "#99CC00", # Vert moyen
    "#FFCC00", # Orange clair
    "#FF9933", # Orange
    "#FF6347", # Orange vif
    "#FF0000"  # Rouge vif
]

Machine Learning - Graphiques