updated the evalueate.py

[redwankarimsony]

1. Previously three functions were used and in all of the model was loaded separately and the data loader was loaded again. Now the model is loaded only once and prediction is generated only once.
2. Accuracy and MSE is newly created.
3. The accuracy and classification report is now being saved in classification_report inside the directory.

[redwankarimsony]

ran the whole code. works fine. here are a few observations:

• the structure looks top-notch. onek guchano. dekhe mone hocche kono pro level er kaj run kortesi.
• (will be done at the end) readme.md needs to be more noob friendly.
• (will be done at the end) remove unnecessary files/folders (checkpoints, etc), or put them in gitignore
• (let me first create the best model) provide a way to use the best model (?) train.py execution er shomoy
• (need them in case something doesn't go right) er shomoy besh onekgula message ashe, like "Successfully opened dynamic library libcudart.so"... these look like error messages. egula suppress kora jay ki?
• (done) model related graphs pdf hishebe store kora better. resizable text thake. and font size arektu boro kora possible @Tasnim IUT CSE'15 can help in this regard. evaluate.py execution er shomoy
• (taken care of) console model summary 2 bar dekhaise, eita may be train e dekhale better hoy. and sathe save o kora jay..
• (still could't figure it out) confusion matrix er top text kete jay..
• (done) console e je f1 score etc dekhaise, eita save kore rakha valo.
• (eita korsilam check korar jonno, will remove at the end) console e shobgula misclassified samples er name print korar dorkar nai. shudhu count print korai enough i guess.