syntaxmatrix 2.5.6__py3-none-any.whl → 2.6.2__py3-none-any.whl

syntaxmatrix/preface.py

@@ -26,6 +26,7 @@ __all__ = [
     "SB_boxplot",
     "SB_scatterplot",
     "SB_heatmap",
+    "viz_stacked_bar",
 
     # core helpers (all underscore-prefixed functions)
     "_SMX_caption_from_ctx",
@@ -68,12 +69,12 @@ except Exception:
     class _Dummy:
         def __getattr__(self, name):
             def _f(*a, **k):
-                from syntaxmatrix.display import show
+                from syntaxmatrix.display_html import show
                 show('⚠ seaborn not available; plot skipped.')
             return _f
     sns = _Dummy()
 
-from syntaxmatrix.display import show as _SMX_base_show
+from syntaxmatrix.display_html import show as _SMX_base_show
 
 
 boxplot = barplot = histplot = distplot = lineplot = countplot = heatmap = pairplot = None
@@ -134,11 +135,32 @@ def _SMX_axes_have_titles(fig=None):
 def _SMX_export_png():
     import io, base64
     fig = plt.gcf()
+
+    # If the figure has no real data, skip exporting to avoid blank images.
+    try:
+        axes = fig.get_axes()
+        has_data = any(
+            getattr(ax, "has_data", lambda: False)()
+            for ax in axes
+        )
+    except Exception:
+        has_data = True  # fail open: better a plot than nothing if check breaks
+
+    if not has_data:
+        try:
+            from syntaxmatrix.display_html import show as _show
+            _show("⚠ Plot skipped: figure has no data to export.")
+        except Exception:
+            pass
+        plt.close(fig)
+        return
+
     try:
         if not _SMX_axes_have_titles(fig):
             fig.suptitle(_SMX_caption_from_ctx(), fontsize=10)
     except Exception:
         pass
+
     buf = io.BytesIO()
     plt.savefig(buf, format='png', bbox_inches='tight')
     buf.seek(0)
@@ -150,7 +172,7 @@ def _SMX_export_png():
             "style='max-width:100%;height:auto;border:1px solid #ccc;border-radius:4px;'/>"
         )
     )
-    plt.close()
+    plt.close(fig)
 
 
 def _pick_df():
@@ -228,7 +250,7 @@ def _safe_plot(func, *args, **kwargs):
             ax = plt.gca()
         try:
             if hasattr(ax, 'has_data') and not ax.has_data():
-                from syntaxmatrix.display import show as _show
+                from syntaxmatrix.display_html import show as _show
                 _show('⚠ Empty plot: no data drawn.')
         except Exception:
             pass
@@ -238,7 +260,7 @@ def _safe_plot(func, *args, **kwargs):
             pass
         return ax
     except Exception as e:
-        from syntaxmatrix.display import show as _show
+        from syntaxmatrix.display_html import show as _show
         _show(f'⚠ Plot skipped: {type(e).__name__}: {e}')
         return None
 
@@ -398,7 +420,7 @@ def SB_heatmap(*a, **k):
         except Exception:
             data = None
     if data is None:
-        from syntaxmatrix.display import show as _show
+        from syntaxmatrix.display_html import show as _show
         _show('⚠ Heatmap skipped: no data.')
         return None
     if not _missing and hasattr(sns, 'heatmap'):
@@ -442,6 +464,85 @@ def SB_heatmap(*a, **k):
     return _safe_plot(lambda **kw: _mat_heat())
 
 
+def viz_stacked_bar(df=None, x=None, hue=None, normalise=True, top_k=8):
+    """
+    Stacked (optionally percentage-stacked) bar chart for two categorical columns.
+
+    - df: optional dataframe. If None, falls back to the active `df` via _pick_df().
+    - x: base categorical axis (e.g. 'state').
+    - hue: second categorical (e.g. 'body').
+    - normalise: if True, show percentages by x; else raw counts.
+    """
+    from syntaxmatrix.display_html import show as _show
+
+    d = df if df is not None else _pick_df()
+    if d is None:
+        _show("⚠ Stacked bar skipped: no dataframe.")
+        return None
+
+    # Choose categorical candidates with reasonable cardinality
+    cat_cols = [
+        c for c in d.columns
+        if (d[c].dtype == "object" or str(d[c].dtype).startswith("category"))
+        and d[c].nunique(dropna=True) > 1
+        and d[c].nunique(dropna=True) <= 30
+    ]
+
+    if x is None or x not in d.columns:
+        x = cat_cols[0] if cat_cols else None
+    if hue is None or hue not in d.columns:
+        remaining = [c for c in cat_cols if c != x]
+        hue = remaining[0] if remaining else None
+
+    if x is None or hue is None:
+        _show("⚠ Stacked bar skipped: need two categorical columns.")
+        return None
+
+    work = d[[x, hue]].dropna()
+    if work.empty:
+        _show("⚠ Stacked bar skipped: no data.")
+        return None
+
+    def _draw():
+        _work = work.copy()
+
+        # Compress minor hue categories into "Other" for readability
+        keep_h = _work[hue].astype(str).value_counts().index[:top_k]
+        _work[hue] = _work[hue].astype(str).where(
+            _work[hue].astype(str).isin(keep_h),
+            other="Other",
+        )
+
+        tab = pd.crosstab(_work[x].astype(str), _work[hue].astype(str))
+
+        try:
+            _show(tab)
+        except Exception:
+            pass
+
+        plot_tab = tab.copy()
+        ylabel = "Count"
+        if normalise:
+            plot_tab = plot_tab.div(plot_tab.sum(axis=1), axis=0) * 100
+            ylabel = "Percentage"
+
+        ax = plot_tab.plot(kind="bar", stacked=True, figsize=(8, 4))
+        title = f"{hue} composition by {x}"
+        if normalise:
+            title += " (%)"
+
+        if not (ax.get_title() or "").strip():
+            ax.set_title(title)
+        ax.set_xlabel(str(x))
+        ax.set_ylabel(ylabel)
+        plt.xticks(rotation=45, ha="right")
+        plt.tight_layout()
+        return ax
+
+    # NOTE: _safe_plot handles empty plots and layout, but does NOT export PNGs.
+    return _safe_plot(lambda **kw: _draw())
+
+
 def _safe_concat(objs, **kwargs):
     import pandas as _pd
     if objs is None:
@@ -453,23 +554,24 @@ def _safe_concat(objs, **kwargs):
     except Exception as e:
         smx_show(f'⚠ concat skipped: {e}')
         return _pd.DataFrame()
-
+    
 
 def _SMX_OHE(**k):
     # normalise arg name across sklearn versions
-    if 'sparse' in k and 'sparse_output' not in k:
-        k['sparse_output'] = k.pop('sparse')
-    k.setdefault('handle_unknown', 'ignore')
-    k.setdefault('sparse_output', False)
+    if "sparse" in k and "sparse_output" not in k:
+        k["sparse_output"] = k.pop("sparse")
+    k.setdefault("handle_unknown", "ignore")
+    k.setdefault("sparse_output", False)
     try:
-        sig = inspect.signature(OneHotEncoder)
-        if 'sparse_output' not in sig.parameters and 'sparse_output' in k:
-            k['sparse'] = k.pop('sparse_output')
-    except Exception:
-        if 'sparse_output' in k:
-            k['sparse'] = k.pop('sparse_output')
-    return OneHotEncoder(**k)
-
+        if "sparse_output" not in inspect.signature(OneHotEncoder).parameters:
+            if "sparse_output" in k:
+                k["sparse"] = k.pop("sparse_output")
+        return OneHotEncoder(**k)
+    except TypeError:
+        if "sparse_output" in k:
+            k["sparse"] = k.pop("sparse_output")
+        return OneHotEncoder(**k)
+        
 
 def _SMX_mm(a, b):
     try:
@@ -489,13 +591,32 @@ def _SMX_mm(a, b):
 
 
 def _SMX_call(fn, *a, **k):
+    """Safe metric invocation that can handle older sklearn signatures.
+
+    - If the metric accepts the provided keywords, it just runs.
+    - If we hit "unexpected keyword argument 'squared'", we drop that kw
+      and retry. When the caller asked for squared=False with
+      mean_squared_error, we emulate RMSE by taking the square root of
+      the returned MSE.
+    """
+    squared_flag = k.get("squared", None)
     try:
         return fn(*a, **k)
     except TypeError as e:
         msg = str(e)
         if "unexpected keyword argument 'squared'" in msg:
-            k.pop('squared', None)
-            return fn(*a, **k)
+            # remove unsupported kw and retry
+            k.pop("squared", None)
+            result = fn(*a, **k)
+            # emulate squared=False for old sklearn.mean_squared_error
+            try:
+                if squared_flag is False and getattr(fn, "__name__", "") == "mean_squared_error":
+                    import numpy as _np
+                    return float(_np.asarray(result, dtype=float) ** 0.5)
+            except Exception:
+                # if anything goes wrong, just fall back to the raw result
+                pass
+            return result
         raise
 
 

syntaxmatrix/profiles.py

@@ -2,7 +2,7 @@
 from openai import OpenAI
 from google import genai
 import anthropic
-
+from datetime import date
 from syntaxmatrix.llm_store import list_profiles, load_profile
 
 # Preload once at import-time
@@ -31,6 +31,7 @@ def get_profiles():
 def get_client(profile):
     
     provider = profile["provider"].lower()
+    model = profile["model"]
     api_key = profile["api_key"]
 
     #1 - Google - gemini series
@@ -41,25 +42,42 @@ def get_client(profile):
     if provider == "openai":    
         return OpenAI(api_key=api_key)
     
-    #3 - xAI - grok series
+    #3 - Anthropic claude series
+    if provider == "anthropic": 
+        return anthropic.Anthropic(api_key=api_key)
+    
+    #4 - xAI - grok series
     if provider == "xai":   
         return OpenAI(api_key=api_key, base_url="https://api.x.ai/v1")
     
-    #4 - DeepSeek chat model
-    if provider == "deepseek":  
-        return OpenAI(api_key=api_key, base_url="https://api.deepseek.com")
+    #5 - DeepSeek chat model
+    if provider == "deepseek":
+        url_special = "https://api.deepseek.com/v3.2_speciale_expires_on_20251215"
+        url_default = "https://api.deepseek.com"
+        expiry_date = date(2025, 12, 15)
+        today = date.today()
+        if today < expiry_date:
+            base_url = url_special
+        else:
+            base_url = url_default
+        return OpenAI(api_key=api_key, base_url=base_url)
     
-    #5 - Moonshot chat model
+    #6 - Moonshot chat model
     if provider == "moonshot":  #5
         return OpenAI(api_key=api_key, base_url="https://api.moonshot.ai/v1")
     
-    #6 - Alibaba qwen series
+    #7 - Alibaba qwen series
     if provider == "alibaba":   #6
         return OpenAI(api_key=api_key, base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",)
     
-    #7 - Anthropic claude series
-    if provider == "anthropic": #7
-        return anthropic.Anthropic(api_key=api_key)
+    #8 - ZAI GLM series
+    if provider == "zai":
+        url = ""
+        if profile.get('name') == "coding":
+            url = "https://api.z.ai/api/coding/paas/v4"
+        else: url = "https://api.z.ai/api/paas/v4"
+        return OpenAI(api_key=api_key, base_url=url)
+    
     
 def drop_cached_profile_by_name(profile_name: str) -> bool:
     """