Merge pull request #1 from app2scale/testing

agent/backend/data.py

@@ -6,12 +6,15 @@ import numpy as np
 class ExplorationDataset(Dataset):
     def __init__(self, df: pd.DataFrame, 
                 input_cols,
-                output_cols):
+                output_cols, transform_dict = None):
         super().__init__()
         self.df = df
         self.input_cols = input_cols
         self.output_cols = output_cols
-        self.transform_dict = self.transform_fit(df)
+        if transform_dict is None:
+            self.transform_dict = self.transform_fit(df)
+        else:
+            self.transform_dict = transform_dict
         self.input_transformed = torch.tensor(self.input_transform(df[input_cols]).values).to(torch.float)
         self.output_transformed = torch.tensor(self.output_transform(df[output_cols]).values).to(torch.float)
                 

agent/backend/utils.py

@@ -3,22 +3,51 @@ from tqdm import tqdm
 from torch.utils.data import Dataset, DataLoader
 from functools import partial
 import pandas as pd
+from typing import Dict, Union, List
+import itertools
+import solara
 
 from .data import ExplorationDataset
 from .models import Perceptron
 from .loss import loss_mape
 
-
-
-def train(df, model_name, input_cols, output_cols, trn_ratio, 
+def predict_dict(model, ds, inputs: Dict[str, Union[List[int], List[float]]]):
+    combinations = itertools.product(*inputs.values())
+    input_cols = list(inputs.keys())
+
+    input_for_df = {col: [] for col in input_cols}
+    for combination in combinations:
+        for input_col, value in zip(input_cols, combination):
+            input_for_df[input_col].append(value)
+
+    input_df = pd.DataFrame(input_for_df)
+    input_df_transformed = ds.input_transform(input_df)
+    #print(input_df)
+    #print(input_df_transformed)
+    ds_new = ExplorationDataset(input_df, input_cols=ds.input_cols, 
+                output_cols=[], transform_dict=ds.transform_dict)
+    dl_new = DataLoader(ds_new, batch_size=len(input_df), shuffle=False)
+    for x, y in dl_new:
+        output = model.forward(x)
+    output_for_df = {col_name: output[:,col_index].detach().numpy() for col_index, col_name in enumerate(ds.output_cols)}
+    output_df = pd.DataFrame(output_for_df)
+    output_df_transformed = ds.output_inv_transform(output_df)
+    #print(output_df)
+    #print(output_df_transformed)
+    return input_df, output_df_transformed
+
+
+
+def train(ds: ExplorationDataset, model_name, trn_ratio, 
           batch_size_trn, batch_size_val, optimizer_name, learning_rate,
           max_epoch, loss_name, seed):
     torch.manual_seed(seed)
+    input_cols, output_cols = ds.input_cols, ds.output_cols
+    df = ds.df
     if model_name == "Perceptron":
         model = Perceptron(in_features=len(input_cols), out_features=len(output_cols))
     if loss_name == "mape":
         loss_fn = loss_mape
-    ds = ExplorationDataset(df, input_cols=input_cols, output_cols=output_cols)
 
     trn_size = int(len(ds)*trn_ratio)
     val_size = len(ds) - trn_size
@@ -103,6 +132,7 @@ def predict_dataloader(model, dataloader):
             predictions = torch.cat([predictions, y_pred], dim=0)
             targets = torch.cat([targets, y], dim=0)
         return predictions, targets
+        
 
 def evaluate(model, dataloader, loss_fn):
     with torch.no_grad():
@@ -140,3 +170,47 @@ def update_policy(model, rewards, log_probabilities, gamma, learning_rate, optim
     # policy_gradient.backward()
     policy_gradient.backward(retain_graph=True)
     optimizer.step()
+
+
+@solara.component
+def Plot1D(x: List, y: List, title='title',xlabel='xlabel', ylabel='ylabel', force_render=0):
+    options = {
+        'title': {
+            'text': title,
+            'left': 'center'},
+        'tooltip': {
+            'trigger': 'axis',
+            'axisPointer': {
+                'type': 'cross'
+            }
+        },
+        'xAxis': {
+            'axisTick': {
+                'alignWithLabel': True
+            },
+            'data': x,
+            'name': xlabel,
+            'nameLocation': 'middle',
+            'nameTextStyle': {'verticalAlign': 'top','padding': [10, 0, 0, 0]}
+        },
+        'yAxis': [
+            {
+                'type': 'value',
+                'name': ylabel,
+                'position': 'left',
+                'alignTicks': True,
+                'axisLine': {
+                    'show': True,
+                    'lineStyle': {'color': 'green'}}
+            },
+        ],
+        'series': [
+            {
+            'name': ylabel,
+            'data': y,
+            'type': 'line',
+            'yAxisIndex': 1
+            },
+        ],
+    }
+    solara.FigureEcharts(option=options)

agent/dashboard/__init__.py

@@ -1,6 +1,6 @@
 import solara
 
-route_order = ["/","data","training","testing"]
+route_order = ["/","data","training","testing","inference"]
 
 @solara.component
 def Page():

agent/dashboard/testing.py

@@ -19,14 +19,18 @@ def force_render():
 
 @solara.component
 def ScatterPlot(predictions, render_count):
-    for col in predictions.keys():
-        with solara.Row():
-            solara_px.scatter(
-                predictions[col]['training'],
-                x = 'prediction',
-                y = 'target',
-                title=f'{col}'
-            )
+    with solara.Card(title='Visual checks of the model performance',
+                        subtitle='''Scatter plots of predicted and target outputs. More the dots
+                        populate around the main diagonal (y=x), better the model performance.'''):
+        for col in predictions.keys():
+                with solara.Row():
+                    for dataset in predictions[col].keys():
+                        solara_px.scatter(
+                            predictions[col][dataset],
+                            x = 'prediction',
+                            y = 'target',
+                            title=f'{col} - {dataset}'
+                    )
             
 
 @solara.component

agent/dashboard/training.py

@@ -3,6 +3,7 @@ import pandas as pd
 from typing import Optional, cast
 import solara.express as solara_px
 from .data import state
+from ..backend.data import ExplorationDataset
 from ..backend.utils import train
 from ..backend.loss import loss_mape
 
@@ -21,6 +22,7 @@ local_state = solara.reactive(
         'loss_plot_data': solara.reactive({'epoch': [], 'trn_loss': [], 'val_loss': []}),
         'render_count': solara.reactive(0),
         'model': solara.reactive(None),
+        'ds': solara.reactive(None),
         'seed': solara.reactive(42),
     }
     )
@@ -85,7 +87,9 @@ def ExecutePanel(df):
         epoch_list = []
         trn_loss_list = []
         val_loss_list = []
-        for epoch, trn_loss, val_loss, model in train(dff, "Perceptron", input_cols, output_cols, trn_ratio,
+        ds = ExplorationDataset(dff, input_cols, output_cols)
+        local_state.value['ds'].set(ds)
+        for epoch, trn_loss, val_loss, model in train(ds, "Perceptron", trn_ratio,
               batch_size_trn, batch_size_val, optimizer_name, learning_rate,
               max_epoch, loss_name, seed):
             epoch_list.append(epoch)